mirror/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-01-18 13:02:28 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions
mariadb/sql/sql_table.cc

7059 lines
215 KiB
C++
Raw Normal View History

WL#1687 - Optimize UNION DISTINCT ... UNION ALL (again). HEAP: Copies the key count to a backup variable and sets the key count to zero. That way, no HEAP function will ever try to touch any index. Re-enabling is done by copying back the backup variable. To avoid memory leak at table close, disable deletes all index trees. Re-enabling must be done with empty indexes and data anyway. Otherwise, the indexes would need to be repaired, wich HEAP is not capable of. MyISAM: Only the key_map is cleared and set. Re-enabling must be done with empty indexes and data. Otherwise, repair needs to be done which will enable all keys persistently. The former implementation disabled only non-unique keys and maked this persistent. The new implementation additionally can disable all keys, but only without making this persistent. Re-enabling all keys can be done without repair, if data file and indexes are empty.
2004-05-06 15:53:01 +02:00
/* Copyright (C) 2000-2004 MySQL AB
Import changeset
2000-07-31 21:29:14 +02:00
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
Many files: Changed header to GPL version 2 only
2006-12-23 20:17:15 +01:00
the Free Software Foundation; version 2 of the License.
Import changeset
2000-07-31 21:29:14 +02:00
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */
/* drop and alter of tables */
#include "mysql_priv.h"
Increased max possible max_allowed_packet to 1G Small optimization to not do external locking of temporary MyISAM tables.
2002-09-19 16:49:41 +02:00
#include <hash.h>
Import changeset
2000-07-31 21:29:14 +02:00
#include <myisam.h>
Fixed problem when comparing a key for a multi-byte-character set. (bug 152) Use 0x.... as strings if 'new' mode. (bug 152) Don't report -max on windows when InnoDB is enabled. (bug 332) Reset current_linfo; This could cause a hang when doing PURGE LOGS. Fix for row numbers in EXPLAIN (bug 322) Fix that USE_FRM works for all table types (bug 97)
2003-04-27 21:12:08 +02:00
#include <my_dir.h>
Fix for bugs #5892/6182/8751/8758/10994 (based on Antony's patch) "Triggers have the wrong namespace" "Triggers: duplicate names allowed" "Triggers: CREATE TRIGGER does not accept fully qualified names" "SHOW TRIGGERS"
2005-07-19 18:06:49 +02:00
#include "sp_head.h"
#include "sql_trigger.h"
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
#include "sql_show.h"
Import changeset
2000-07-31 21:29:14 +02:00
#ifdef __WIN__
#include <io.h>
#endif
renamedb.test, renamedb.result: new file Many files: WL#757 RENAME DATABASE
2006-02-13 08:49:28 +01:00
int creating_table= 0; // How many mysql_create_table are running
after merge fixes
2004-09-07 16:56:31 +02:00
const char *primary_key_name="PRIMARY";
Import changeset
2000-07-31 21:29:14 +02:00
static bool check_if_keyname_exists(const char *name,KEY *start, KEY *end);
static char *make_unique_key_name(const char *field_name,KEY *start,KEY *end);
static int copy_data_between_tables(TABLE *from,TABLE *to,
Merge mysql.com:/home/dlenev/mysql-5.0-bg18437-3 into mysql.com:/home/dlenev/mysql-5.1-bg18437
2006-07-02 00:12:53 +02:00
List<create_field> &create, bool ignore,
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
uint order_num, ORDER *order,
Merge example.com:/work/bug24395-v2/my50 into example.com:/work/bug24395-v2/my51
2006-12-01 11:37:33 +01:00
ha_rows *copied,ha_rows *deleted,
enum enum_enable_or_disable keys_onoff);
Fixes while reviewing new code Added option --count to mysqlshow (to show number of rows) Fixed possible core dump in information schema
2005-05-06 10:39:30 +02:00
static bool prepare_blob_field(THD *thd, create_field *sql_field);
Bug#6877 MySQL should give an error if the requested table type is not available Implement new SQL mode - NO_ENGINE_SUBSTITUTION
2005-06-17 23:14:44 +02:00
static bool check_engine(THD *thd, const char *table_name,
Bug #17210 Create temp table call to ha_ndbcluster::create_handler_files caused core - do not try to create ndb temp tables even if default storage is ndb
2006-02-17 17:12:35 +01:00
HA_CREATE_INFO *create_info);
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
static int mysql_prepare_table(THD *thd, HA_CREATE_INFO *create_info,
List<create_field> *fields,
List<Key> *keys, bool tmp_table,
uint *db_options,
handler *file, KEY **key_info_buffer,
uint *key_count, int select_field_count);
Import changeset
2000-07-31 21:29:14 +02:00
Making old tables seen with "#mysql50#" prefix, which makes it possible to run RENAME TABLE on old tables when upgrading from 5.0. TODO: A stored procedure to rename all tables and databases with old name format into new format, it will simplify upgrade. sql_table.cc: Making old tables seen with "#mysql50#" prefix. Adding warning into .err log when an old name is found. sql_show.cc: Skip non-directories before filename_to_tablename call, to avoid unnecessary warnings. strfunc.cc: Adding "error" argument to strconvert() mysql_priv.h: Adding "error" agrument to strconvert()
2005-12-31 09:34:39 +01:00
#define MYSQL50_TABLE_NAME_PREFIX "#mysql50#"
#define MYSQL50_TABLE_NAME_PREFIX_LENGTH 9
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
/*
Translate a file name to a table name (WL #1324).
SYNOPSIS
filename_to_tablename()
from The file name in my_charset_filename.
to OUT The table name in system_charset_info.
to_length The size of the table name buffer.
RETURN
Table name length.
*/
Making old tables seen with "#mysql50#" prefix, which makes it possible to run RENAME TABLE on old tables when upgrading from 5.0. TODO: A stored procedure to rename all tables and databases with old name format into new format, it will simplify upgrade. sql_table.cc: Making old tables seen with "#mysql50#" prefix. Adding warning into .err log when an old name is found. sql_show.cc: Skip non-directories before filename_to_tablename call, to avoid unnecessary warnings. strfunc.cc: Adding "error" argument to strconvert() mysql_priv.h: Adding "error" agrument to strconvert()
2005-12-31 09:34:39 +01:00
uint filename_to_tablename(const char *from, char *to, uint to_length)
{
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
uint errors;
uint res;
DBUG_ENTER("filename_to_tablename");
DBUG_PRINT("enter", ("from '%s'", from));
if (!memcmp(from, tmp_file_prefix, tmp_file_prefix_length))
Making old tables seen with "#mysql50#" prefix, which makes it possible to run RENAME TABLE on old tables when upgrading from 5.0. TODO: A stored procedure to rename all tables and databases with old name format into new format, it will simplify upgrade. sql_table.cc: Making old tables seen with "#mysql50#" prefix. Adding warning into .err log when an old name is found. sql_show.cc: Skip non-directories before filename_to_tablename call, to avoid unnecessary warnings. strfunc.cc: Adding "error" argument to strconvert() mysql_priv.h: Adding "error" agrument to strconvert()
2005-12-31 09:34:39 +01:00
{
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
/* Temporary table name. */
res= (strnmov(to, from, to_length) - to);
}
else
{
res= strconvert(&my_charset_filename, from,
system_charset_info, to, to_length, &errors);
if (errors) // Old 5.0 name
{
res= (strxnmov(to, to_length, MYSQL50_TABLE_NAME_PREFIX, from, NullS) -
to);
sql_print_error("Invalid (old?) table or database name '%s'", from);
/*
TODO: add a stored procedure for fix table and database names,
and mention its name in error log.
*/
}
Making old tables seen with "#mysql50#" prefix, which makes it possible to run RENAME TABLE on old tables when upgrading from 5.0. TODO: A stored procedure to rename all tables and databases with old name format into new format, it will simplify upgrade. sql_table.cc: Making old tables seen with "#mysql50#" prefix. Adding warning into .err log when an old name is found. sql_show.cc: Skip non-directories before filename_to_tablename call, to avoid unnecessary warnings. strfunc.cc: Adding "error" argument to strconvert() mysql_priv.h: Adding "error" agrument to strconvert()
2005-12-31 09:34:39 +01:00
}
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
DBUG_PRINT("exit", ("to '%s'", to));
DBUG_RETURN(res);
Making old tables seen with "#mysql50#" prefix, which makes it possible to run RENAME TABLE on old tables when upgrading from 5.0. TODO: A stored procedure to rename all tables and databases with old name format into new format, it will simplify upgrade. sql_table.cc: Making old tables seen with "#mysql50#" prefix. Adding warning into .err log when an old name is found. sql_show.cc: Skip non-directories before filename_to_tablename call, to avoid unnecessary warnings. strfunc.cc: Adding "error" argument to strconvert() mysql_priv.h: Adding "error" agrument to strconvert()
2005-12-31 09:34:39 +01:00
}
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
/*
Translate a table name to a file name (WL #1324).
SYNOPSIS
tablename_to_filename()
from The table name in system_charset_info.
to OUT The file name in my_charset_filename.
to_length The size of the file name buffer.
RETURN
File name length.
*/
Making old tables seen with "#mysql50#" prefix, which makes it possible to run RENAME TABLE on old tables when upgrading from 5.0. TODO: A stored procedure to rename all tables and databases with old name format into new format, it will simplify upgrade. sql_table.cc: Making old tables seen with "#mysql50#" prefix. Adding warning into .err log when an old name is found. sql_show.cc: Skip non-directories before filename_to_tablename call, to avoid unnecessary warnings. strfunc.cc: Adding "error" argument to strconvert() mysql_priv.h: Adding "error" agrument to strconvert()
2005-12-31 09:34:39 +01:00
uint tablename_to_filename(const char *from, char *to, uint to_length)
{
bug#17870 Table names conflict with Windows device names It was impossible to create some table names on Windows (e.g. LPT1, AUX, COM1, etc). Fixed to pad dangerous names with thee "at" signs (e.g. LPT1@@@, AUX@@@, COM1@@@, and so on).
2006-04-11 15:16:14 +02:00
uint errors, length;
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
DBUG_ENTER("tablename_to_filename");
DBUG_PRINT("enter", ("from '%s'", from));
sql_table.cc: Typo fix. Thanks Serg for noticing this.
2006-01-12 08:41:57 +01:00
if (from[0] == '#' && !strncmp(from, MYSQL50_TABLE_NAME_PREFIX,
MYSQL50_TABLE_NAME_PREFIX_LENGTH))
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
DBUG_RETURN((uint) (strmake(to, from+MYSQL50_TABLE_NAME_PREFIX_LENGTH,
to_length-1) -
(from + MYSQL50_TABLE_NAME_PREFIX_LENGTH)));
bug#17870 Table names conflict with Windows device names It was impossible to create some table names on Windows (e.g. LPT1, AUX, COM1, etc). Fixed to pad dangerous names with thee "at" signs (e.g. LPT1@@@, AUX@@@, COM1@@@, and so on).
2006-04-11 15:16:14 +02:00
length= strconvert(system_charset_info, from,
&my_charset_filename, to, to_length, &errors);
if (check_if_legal_tablename(to) &&
length + 4 < to_length)
{
memcpy(to + length, "@@@", 4);
length+= 3;
}
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
DBUG_PRINT("exit", ("to '%s'", to));
DBUG_RETURN(length);
Making old tables seen with "#mysql50#" prefix, which makes it possible to run RENAME TABLE on old tables when upgrading from 5.0. TODO: A stored procedure to rename all tables and databases with old name format into new format, it will simplify upgrade. sql_table.cc: Making old tables seen with "#mysql50#" prefix. Adding warning into .err log when an old name is found. sql_show.cc: Skip non-directories before filename_to_tablename call, to avoid unnecessary warnings. strfunc.cc: Adding "error" argument to strconvert() mysql_priv.h: Adding "error" agrument to strconvert()
2005-12-31 09:34:39 +01:00
}
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
/*
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
Creates path to a file: mysql_data_dir/db/table.ext
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
SYNOPSIS
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
build_table_filename()
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
buff Where to write result in my_charset_filename.
bufflen buff size
db Database name in system_charset_info.
table_name Table name in system_charset_info.
ext File extension.
flags FN_FROM_IS_TMP or FN_TO_IS_TMP or FN_IS_TMP
table_name is temporary, do not change.
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
NOTES
Uses database and table name, and extension to create
a file name in mysql_data_dir. Database and table
names are converted from system_charset_info into "fscs".
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
Unless flags indicate a temporary table name.
'db' is always converted.
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
'ext' is not converted.
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
The conversion suppression is required for ALTER TABLE. This
statement creates intermediate tables. These are regular
(non-temporary) tables with a temporary name. Their path names must
be derivable from the table name. So we cannot use
build_tmptable_filename() for them.
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
RETURN
path length
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
*/
uint build_table_filename(char *buff, size_t bufflen, const char *db,
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
const char *table_name, const char *ext, uint flags)
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
{
uint length;
char dbbuff[FN_REFLEN];
char tbbuff[FN_REFLEN];
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
DBUG_ENTER("build_table_filename");
if (flags & FN_IS_TMP) // FN_FROM_IS_TMP | FN_TO_IS_TMP
strnmov(tbbuff, table_name, sizeof(tbbuff));
else
VOID(tablename_to_filename(table_name, tbbuff, sizeof(tbbuff)));
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
VOID(tablename_to_filename(db, dbbuff, sizeof(dbbuff)));
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
length= strxnmov(buff, bufflen, mysql_data_home, "/", dbbuff,
"/", tbbuff, ext, NullS) - buff;
DBUG_PRINT("exit", ("buff: '%s'", buff));
DBUG_RETURN(length);
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
}
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
/*
Creates path to a file: mysql_tmpdir/#sql1234_12_1.ext
SYNOPSIS
build_tmptable_filename()
thd The thread handle.
buff Where to write result in my_charset_filename.
bufflen buff size
NOTES
Uses current_pid, thread_id, and tmp_table counter to create
a file name in mysql_tmpdir.
RETURN
path length
*/
Bug#15904 --tmpdir requires trailing slash '/' - Use same code for creating the temptable filename both from "mysql_create_table" and "mysql_create_like_table"
2006-01-23 13:28:42 +01:00
uint build_tmptable_filename(THD* thd, char *buff, size_t bufflen)
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
{
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
uint length;
Bug#15904 --tmpdir requires trailing slash '/' - Use same code for creating the temptable filename both from "mysql_create_table" and "mysql_create_like_table"
2006-01-23 13:28:42 +01:00
char tmp_table_name[tmp_file_prefix_length+22+22+22+3];
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
DBUG_ENTER("build_tmptable_filename");
Bug#15904 --tmpdir requires trailing slash '/' - Use same code for creating the temptable filename both from "mysql_create_table" and "mysql_create_like_table"
2006-01-23 13:28:42 +01:00
my_snprintf(tmp_table_name, sizeof(tmp_table_name),
"%s%lx_%lx_%x",
tmp_file_prefix, current_pid,
thd->thread_id, thd->tmp_table++);
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
strxnmov(buff, bufflen, mysql_tmpdir, "/", tmp_table_name, reg_ext, NullS);
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
length= unpack_filename(buff, buff);
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
DBUG_PRINT("exit", ("buff: '%s'", buff));
DBUG_RETURN(length);
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
}
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
/*
Return values for compare_tables().
If you make compare_tables() non-static, move them to a header file.
*/
#define ALTER_TABLE_DATA_CHANGED 1
#define ALTER_TABLE_INDEX_CHANGED 2
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
/*
SYNOPSIS
mysql_copy_create_list()
orig_create_list Original list of created fields
inout::new_create_list Copy of original list
RETURN VALUES
FALSE Success
TRUE Memory allocation error
DESCRIPTION
mysql_prepare_table destroys the create_list and in some cases we need
this lists for more purposes. Thus we copy it specifically for use
by mysql_prepare_table
*/
static int mysql_copy_create_list(List<create_field> *orig_create_list,
Bug #17169: Partitions: out of memory if add partition and unique. When creating a new partition, a bogus memory allocation problem was reported.
2006-02-16 17:15:36 +01:00
List<create_field> *new_create_list)
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
{
List_iterator<create_field> prep_field_it(*orig_create_list);
create_field *prep_field;
DBUG_ENTER("mysql_copy_create_list");
while ((prep_field= prep_field_it++))
{
create_field *field= new create_field(*prep_field);
if (!field || new_create_list->push_back(field))
{
mem_alloc_error(2);
DBUG_RETURN(TRUE);
}
}
DBUG_RETURN(FALSE);
}
/*
SYNOPSIS
mysql_copy_key_list()
orig_key Original list of keys
inout::new_key Copy of original list
RETURN VALUES
FALSE Success
TRUE Memory allocation error
DESCRIPTION
mysql_prepare_table destroys the key list and in some cases we need
this lists for more purposes. Thus we copy it specifically for use
by mysql_prepare_table
*/
static int mysql_copy_key_list(List<Key> *orig_key,
List<Key> *new_key)
{
List_iterator<Key> prep_key_it(*orig_key);
Key *prep_key;
Bug #17169: Partitions: out of memory if add partition and unique. When creating a new partition, a bogus memory allocation problem was reported.
2006-02-16 17:15:36 +01:00
DBUG_ENTER("mysql_copy_key_list");
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
while ((prep_key= prep_key_it++))
{
List<key_part_spec> prep_columns;
List_iterator<key_part_spec> prep_col_it(prep_key->columns);
key_part_spec *prep_col;
Key *temp_key;
while ((prep_col= prep_col_it++))
{
key_part_spec *prep_key_part;
Bug #17169: Partitions: out of memory if add partition and unique. When creating a new partition, a bogus memory allocation problem was reported.
2006-02-16 17:15:36 +01:00
if (!(prep_key_part= new key_part_spec(*prep_col)))
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
{
mem_alloc_error(sizeof(key_part_spec));
DBUG_RETURN(TRUE);
}
if (prep_columns.push_back(prep_key_part))
{
mem_alloc_error(2);
DBUG_RETURN(TRUE);
}
}
Bug #17169: Partitions: out of memory if add partition and unique. When creating a new partition, a bogus memory allocation problem was reported.
2006-02-16 17:15:36 +01:00
if (!(temp_key= new Key(prep_key->type, prep_key->name,
Cleanups after review of WL#602 Fixed warnings from test suite Some fixes in mysql-test-run script to catch more warnings
2006-05-03 18:40:52 +02:00
&prep_key->key_create_info,
Bug #17169: Partitions: out of memory if add partition and unique. When creating a new partition, a bogus memory allocation problem was reported.
2006-02-16 17:15:36 +01:00
prep_key->generated,
Added support for key_block_size for key and table level (WL#602) Added support for key_block_size to MyISAM. Simplify interface to 'new Key' to make it easier to add new key options. mysqld option --new is used to define where key options are printed. (In 5.3 we should move all key options to after key part definition to avoid problem with reserved names) Fixed some compiler warnings and a memory leak in ssl
2006-05-03 14:59:17 +02:00
prep_columns)))
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
{
mem_alloc_error(sizeof(Key));
DBUG_RETURN(TRUE);
}
if (new_key->push_back(temp_key))
{
mem_alloc_error(2);
DBUG_RETURN(TRUE);
}
}
DBUG_RETURN(FALSE);
}
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
/*
--------------------------------------------------------------------------
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
MODULE: DDL log
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
-----------------
This module is used to ensure that we can recover from crashes that occur
in the middle of a meta-data operation in MySQL. E.g. DROP TABLE t1, t2;
We need to ensure that both t1 and t2 are dropped and not only t1 and
also that each table drop is entirely done and not "half-baked".
To support this we create log entries for each meta-data statement in the
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
ddl log while we are executing. These entries are dropped when the
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
operation is completed.
At recovery those entries that were not completed will be executed.
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
There is only one ddl log in the system and it is protected by a mutex
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
and there is a global struct that contains information about its current
state.
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
History:
First version written in 2006 by Mikael Ronstrom
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
--------------------------------------------------------------------------
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
typedef struct st_global_ddl_log
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
/*
We need to adjust buffer size to be able to handle downgrades/upgrades
where IO_SIZE has changed. We'll set the buffer size such that we can
handle that the buffer size was upto 4 times bigger in the version
that wrote the DDL log.
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
char file_entry_buf[4*IO_SIZE];
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
char file_name_str[FN_REFLEN];
char *file_name;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DDL_LOG_MEMORY_ENTRY *first_free;
DDL_LOG_MEMORY_ENTRY *first_used;
uint num_entries;
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
File file_id;
uint name_len;
WL 2826: Error handling for ALTER TABLE for partitioning Step 14: First version of table log for add/Drop partition
2006-02-09 11:05:05 +01:00
uint io_size;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
bool inited;
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
bool recovery_phase;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
} GLOBAL_DDL_LOG;
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
GLOBAL_DDL_LOG global_ddl_log;
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
pthread_mutex_t LOCK_gdl;
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
#define DDL_LOG_ENTRY_TYPE_POS 0
#define DDL_LOG_ACTION_TYPE_POS 1
#define DDL_LOG_PHASE_POS 2
#define DDL_LOG_NEXT_ENTRY_POS 4
#define DDL_LOG_NAME_POS 8
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
#define DDL_LOG_NUM_ENTRY_POS 0
#define DDL_LOG_NAME_LEN_POS 4
WL 2826: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
#define DDL_LOG_IO_SIZE_POS 8
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Read one entry from ddl log file
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
read_ddl_log_file_entry()
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
entry_no Entry number to read
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
RETURN VALUES
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
TRUE Error
FALSE Success
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
static bool read_ddl_log_file_entry(uint entry_no)
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
{
bool error= FALSE;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
File file_id= global_ddl_log.file_id;
char *file_entry_buf= (char*)global_ddl_log.file_entry_buf;
uint io_size= global_ddl_log.io_size;
DBUG_ENTER("read_ddl_log_file_entry");
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
Fix windows compile problem
2006-04-22 13:29:05 +02:00
if (my_pread(file_id, (byte*)file_entry_buf, io_size, io_size * entry_no,
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
MYF(MY_WME)) != io_size)
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
error= TRUE;
DBUG_RETURN(error);
}
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Write one entry from ddl log file
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
write_ddl_log_file_entry()
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
entry_no Entry number to read
RETURN VALUES
TRUE Error
FALSE Success
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
static bool write_ddl_log_file_entry(uint entry_no)
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
{
bool error= FALSE;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
File file_id= global_ddl_log.file_id;
char *file_entry_buf= (char*)global_ddl_log.file_entry_buf;
DBUG_ENTER("write_ddl_log_file_entry");
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
Fix windows compile problem
2006-04-22 13:29:05 +02:00
if (my_pwrite(file_id, (byte*)file_entry_buf,
WL 2826: Error handling of ALTER TABLE for partitioning First step for handling errors in ALTER TABLE for partitioning
2006-02-14 14:22:21 +01:00
IO_SIZE, IO_SIZE * entry_no, MYF(MY_WME)) != IO_SIZE)
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
error= TRUE;
DBUG_RETURN(error);
}
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Write ddl log header
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
write_ddl_log_header()
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
RETURN VALUES
TRUE Error
FALSE Success
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
static bool write_ddl_log_header()
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
{
uint16 const_var;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
bool error= FALSE;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DBUG_ENTER("write_ddl_log_header");
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
int4store(&global_ddl_log.file_entry_buf[DDL_LOG_NUM_ENTRY_POS],
global_ddl_log.num_entries);
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
const_var= FN_LEN;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
int4store(&global_ddl_log.file_entry_buf[DDL_LOG_NAME_LEN_POS],
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
const_var);
WL 2826: Error handling for ALTER TABLE for partitioning Step 14: First version of table log for add/Drop partition
2006-02-09 11:05:05 +01:00
const_var= IO_SIZE;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
int4store(&global_ddl_log.file_entry_buf[DDL_LOG_IO_SIZE_POS],
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
const_var);
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (write_ddl_log_file_entry(0UL))
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
{
sql_print_error("Error writing ddl log header");
DBUG_RETURN(TRUE);
}
VOID(sync_ddl_log());
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
DBUG_RETURN(error);
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
}
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Create ddl log file name
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
create_ddl_log_file_name()
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
file_name Filename setup
RETURN VALUES
NONE
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
static inline void create_ddl_log_file_name(char *file_name)
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
strxmov(file_name, mysql_data_home, "/", "ddl_log.log", NullS);
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Read header of ddl log file
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
read_ddl_log_header()
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
RETURN VALUES
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
> 0 Last entry in ddl log
0 No entries in ddl log
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
DESCRIPTION
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
When we read the ddl log header we get information about maximum sizes
of names in the ddl log and we also get information about the number
of entries in the ddl log.
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
static uint read_ddl_log_header()
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
char *file_entry_buf= (char*)global_ddl_log.file_entry_buf;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
char file_name[FN_REFLEN];
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
uint entry_no;
WL 2826: Error handling for ALTER TABLE for partitioning Step 14: First version of table log for add/Drop partition
2006-02-09 11:05:05 +01:00
bool successful_open= FALSE;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DBUG_ENTER("read_ddl_log_header");
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
create_ddl_log_file_name(file_name);
WL 2826: Error handling of ALTER TABLE for partitioning Fixed a number of bugs
2006-04-19 22:08:59 +02:00
if ((global_ddl_log.file_id= my_open(file_name,
O_RDWR | O_BINARY, MYF(MY_WME))) >= 0)
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (read_ddl_log_file_entry(0UL))
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
/* Write message into error log */
sql_print_error("Failed to read ddl log file in recovery");
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
WL 2826: Error handling for ALTER TABLE for partitioning Step 14: First version of table log for add/Drop partition
2006-02-09 11:05:05 +01:00
else
successful_open= TRUE;
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
}
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
entry_no= uint4korr(&file_entry_buf[DDL_LOG_NUM_ENTRY_POS]);
global_ddl_log.name_len= uint4korr(&file_entry_buf[DDL_LOG_NAME_LEN_POS]);
WL 2826: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
if (successful_open)
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
global_ddl_log.io_size= uint4korr(&file_entry_buf[DDL_LOG_IO_SIZE_POS]);
WL 2826: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
DBUG_ASSERT(global_ddl_log.io_size <=
sizeof(global_ddl_log.file_entry_buf));
}
WL 2826: Error handling of ALTER TABLE for partitioning Bug fixes
2006-02-20 22:22:19 +01:00
else
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
{
entry_no= 0;
}
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
global_ddl_log.first_free= NULL;
global_ddl_log.first_used= NULL;
global_ddl_log.num_entries= 0;
VOID(pthread_mutex_init(&LOCK_gdl, MY_MUTEX_INIT_FAST));
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
DBUG_RETURN(entry_no);
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
}
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Read a ddl log entry
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
read_ddl_log_entry()
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
read_entry Number of entry to read
out:entry_info Information from entry
RETURN VALUES
TRUE Error
FALSE Success
DESCRIPTION
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Read a specified entry in the ddl log
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
bool read_ddl_log_entry(uint read_entry, DDL_LOG_ENTRY *ddl_log_entry)
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
char *file_entry_buf= (char*)&global_ddl_log.file_entry_buf;
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
uint inx;
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
uchar single_char;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DBUG_ENTER("read_ddl_log_entry");
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (read_ddl_log_file_entry(read_entry))
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
DBUG_RETURN(TRUE);
}
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
ddl_log_entry->entry_pos= read_entry;
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
single_char= file_entry_buf[DDL_LOG_ENTRY_TYPE_POS];
ddl_log_entry->entry_type= (enum ddl_log_entry_code)single_char;
single_char= file_entry_buf[DDL_LOG_ACTION_TYPE_POS];
ddl_log_entry->action_type= (enum ddl_log_action_code)single_char;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
ddl_log_entry->phase= file_entry_buf[DDL_LOG_PHASE_POS];
ddl_log_entry->next_entry= uint4korr(&file_entry_buf[DDL_LOG_NEXT_ENTRY_POS]);
ddl_log_entry->name= &file_entry_buf[DDL_LOG_NAME_POS];
inx= DDL_LOG_NAME_POS + global_ddl_log.name_len;
ddl_log_entry->from_name= &file_entry_buf[inx];
inx+= global_ddl_log.name_len;
ddl_log_entry->handler_name= &file_entry_buf[inx];
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
DBUG_RETURN(FALSE);
}
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Initialise ddl log
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
init_ddl_log()
Cleanups after review of WL#602 Fixed warnings from test suite Some fixes in mysql-test-run script to catch more warnings
2006-05-03 18:40:52 +02:00
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
DESCRIPTION
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Write the header of the ddl log file and length of names. Also set
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
number of entries to zero.
Cleanups after review of WL#602 Fixed warnings from test suite Some fixes in mysql-test-run script to catch more warnings
2006-05-03 18:40:52 +02:00
RETURN VALUES
TRUE Error
FALSE Success
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
static bool init_ddl_log()
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
{
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
char file_name[FN_REFLEN];
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DBUG_ENTER("init_ddl_log");
WL 2826: Step 12 More table log code
2006-02-07 16:26:34 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (global_ddl_log.inited)
Cleanups after review of WL#602 Fixed warnings from test suite Some fixes in mysql-test-run script to catch more warnings
2006-05-03 18:40:52 +02:00
goto end;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
global_ddl_log.io_size= IO_SIZE;
WL 2826: Error handling of ALTER TABLE for partitioning Fixed a number of bugs
2006-04-19 22:08:59 +02:00
create_ddl_log_file_name(file_name);
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if ((global_ddl_log.file_id= my_create(file_name,
CREATE_MODE,
O_RDWR | O_TRUNC | O_BINARY,
MYF(MY_WME))) < 0)
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
/* Couldn't create ddl log file, this is serious error */
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
sql_print_error("Failed to open ddl log file");
DBUG_RETURN(TRUE);
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
Fixes
2006-04-03 18:26:35 +02:00
global_ddl_log.inited= TRUE;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (write_ddl_log_header())
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
Cleanups after review of WL#602 Fixed warnings from test suite Some fixes in mysql-test-run script to catch more warnings
2006-05-03 18:40:52 +02:00
VOID(my_close(global_ddl_log.file_id, MYF(MY_WME)));
Fixes
2006-04-03 18:26:35 +02:00
global_ddl_log.inited= FALSE;
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
DBUG_RETURN(TRUE);
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
Cleanups after review of WL#602 Fixed warnings from test suite Some fixes in mysql-test-run script to catch more warnings
2006-05-03 18:40:52 +02:00
end:
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
DBUG_RETURN(FALSE);
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
}
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Execute one action in a ddl log entry
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
execute_ddl_log_action()
ddl_log_entry Information in action entry to execute
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
RETURN VALUES
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
TRUE Error
FALSE Success
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
*/
WL2826: Error handling of ALTER TABLE for partitioning Actual error reported in my_errno from my_delete
2006-04-20 02:57:42 +02:00
static int execute_ddl_log_action(THD *thd, DDL_LOG_ENTRY *ddl_log_entry)
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
bool frm_action= FALSE;
LEX_STRING handler_name;
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
handler *file= NULL;
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
MEM_ROOT mem_root;
WL2826: Error handling of ALTER TABLE for partitioning Actual error reported in my_errno from my_delete
2006-04-20 02:57:42 +02:00
int error= TRUE;
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
char to_path[FN_REFLEN];
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
char from_path[FN_REFLEN];
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
char *par_ext= (char*)".par";
handlerton *hton;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DBUG_ENTER("execute_ddl_log_action");
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (ddl_log_entry->entry_type == DDL_IGNORE_LOG_ENTRY_CODE)
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
{
DBUG_RETURN(FALSE);
}
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
handler_name.str= (char*)ddl_log_entry->handler_name;
handler_name.length= strlen(ddl_log_entry->handler_name);
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
init_sql_alloc(&mem_root, TABLE_ALLOC_BLOCK_SIZE, 0);
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
if (!strcmp(ddl_log_entry->handler_name, reg_ext))
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
frm_action= TRUE;
else
{
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
TABLE_SHARE dummy;
WL 2826: Error handling of ALTER TABLE for partitioning Fixed a number of bugs
2006-04-19 22:08:59 +02:00
hton= ha_resolve_by_name(thd, &handler_name);
if (!hton)
{
my_error(ER_ILLEGAL_HA, MYF(0), ddl_log_entry->handler_name);
goto error;
}
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
bzero(&dummy, sizeof(TABLE_SHARE));
file= get_new_handler(&dummy, &mem_root, hton);
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
if (!file)
WL 2826: Error handling of ALTER TABLE for partitioning First step for handling errors in ALTER TABLE for partitioning
2006-02-14 14:22:21 +01:00
{
mem_alloc_error(sizeof(handler));
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
goto error;
WL 2826: Error handling of ALTER TABLE for partitioning First step for handling errors in ALTER TABLE for partitioning
2006-02-14 14:22:21 +01:00
}
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
}
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
switch (ddl_log_entry->action_type)
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
case DDL_LOG_REPLACE_ACTION:
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
case DDL_LOG_DELETE_ACTION:
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
if (ddl_log_entry->phase == 0)
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
{
if (frm_action)
{
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
strxmov(to_path, ddl_log_entry->name, reg_ext, NullS);
if ((error= my_delete(to_path, MYF(MY_WME))))
{
WL2826: Error handling of ALTER TABLE for partitioning Actual error reported in my_errno from my_delete
2006-04-20 02:57:42 +02:00
if (my_errno != ENOENT)
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
break;
}
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
strxmov(to_path, ddl_log_entry->name, par_ext, NullS);
VOID(my_delete(to_path, MYF(MY_WME)));
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
#endif
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
}
else
{
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
if ((error= file->delete_table(ddl_log_entry->name)))
{
if (error != ENOENT && error != HA_ERR_NO_SUCH_TABLE)
break;
}
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
}
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
if ((deactivate_ddl_log_entry(ddl_log_entry->entry_pos)))
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
break;
VOID(sync_ddl_log());
error= FALSE;
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
if (ddl_log_entry->action_type == DDL_LOG_DELETE_ACTION)
break;
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
}
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
DBUG_ASSERT(ddl_log_entry->action_type == DDL_LOG_REPLACE_ACTION);
/*
Fall through and perform the rename action of the replace
action. We have already indicated the success of the delete
action in the log entry by stepping up the phase.
*/
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
}
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
case DDL_LOG_RENAME_ACTION:
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
error= TRUE;
if (frm_action)
{
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
strxmov(to_path, ddl_log_entry->name, reg_ext, NullS);
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
strxmov(from_path, ddl_log_entry->from_name, reg_ext, NullS);
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
if (my_rename(from_path, to_path, MYF(MY_WME)))
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
break;
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
strxmov(to_path, ddl_log_entry->name, par_ext, NullS);
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
strxmov(from_path, ddl_log_entry->from_name, par_ext, NullS);
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
VOID(my_rename(from_path, to_path, MYF(MY_WME)));
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
#endif
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
}
else
{
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
if (file->rename_table(ddl_log_entry->from_name,
ddl_log_entry->name))
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
break;
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
}
if ((deactivate_ddl_log_entry(ddl_log_entry->entry_pos)))
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
break;
VOID(sync_ddl_log());
error= FALSE;
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
break;
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
}
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
default:
DBUG_ASSERT(0);
break;
}
delete file;
error:
free_root(&mem_root, MYF(0));
DBUG_RETURN(error);
WL 2826: Step 10 New methods to read/write/initialise table log
2006-02-06 21:47:03 +01:00
}
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Get a free entry in the ddl log
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
get_free_ddl_log_entry()
out:active_entry A ddl log memory entry returned
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
RETURN VALUES
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
TRUE Error
FALSE Success
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
static bool get_free_ddl_log_entry(DDL_LOG_MEMORY_ENTRY **active_entry,
bool *write_header)
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DDL_LOG_MEMORY_ENTRY *used_entry;
DDL_LOG_MEMORY_ENTRY *first_used= global_ddl_log.first_used;
DBUG_ENTER("get_free_ddl_log_entry");
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (global_ddl_log.first_free == NULL)
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (!(used_entry= (DDL_LOG_MEMORY_ENTRY*)my_malloc(
sizeof(DDL_LOG_MEMORY_ENTRY), MYF(MY_WME))))
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
sql_print_error("Failed to allocate memory for ddl log free list");
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
DBUG_RETURN(TRUE);
}
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
global_ddl_log.num_entries++;
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
used_entry->entry_pos= global_ddl_log.num_entries;
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
*write_header= TRUE;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
else
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
used_entry= global_ddl_log.first_free;
global_ddl_log.first_free= used_entry->next_log_entry;
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
*write_header= FALSE;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
/*
Move from free list to used list
*/
used_entry->next_log_entry= first_used;
used_entry->prev_log_entry= NULL;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
global_ddl_log.first_used= used_entry;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
if (first_used)
first_used->prev_log_entry= used_entry;
*active_entry= used_entry;
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
DBUG_RETURN(FALSE);
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
}
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
External interface methods for the DDL log Module
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
---------------------------------------------------
*/
/*
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
write_ddl_log_entry()
ddl_log_entry Information about log entry
out:entry_written Entry information written into
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
RETURN VALUES
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
TRUE Error
FALSE Success
DESCRIPTION
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
A careful write of the ddl log is performed to ensure that we can
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
handle crashes occurring during CREATE and ALTER TABLE processing.
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
bool write_ddl_log_entry(DDL_LOG_ENTRY *ddl_log_entry,
DDL_LOG_MEMORY_ENTRY **active_entry)
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
{
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
bool error, write_header;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DBUG_ENTER("write_ddl_log_entry");
if (init_ddl_log())
{
DBUG_RETURN(TRUE);
}
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
global_ddl_log.file_entry_buf[DDL_LOG_ENTRY_TYPE_POS]=
(char)DDL_LOG_ENTRY_CODE;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
global_ddl_log.file_entry_buf[DDL_LOG_ACTION_TYPE_POS]=
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
(char)ddl_log_entry->action_type;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
global_ddl_log.file_entry_buf[DDL_LOG_PHASE_POS]= 0;
int4store(&global_ddl_log.file_entry_buf[DDL_LOG_NEXT_ENTRY_POS],
ddl_log_entry->next_entry);
DBUG_ASSERT(strlen(ddl_log_entry->name) < FN_LEN);
WL 2826: Error handling of ALTER TABLE for partitioning Final review fixes
2006-04-21 14:35:25 +02:00
strmake(&global_ddl_log.file_entry_buf[DDL_LOG_NAME_POS],
ddl_log_entry->name, FN_LEN - 1);
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (ddl_log_entry->action_type == DDL_LOG_RENAME_ACTION ||
ddl_log_entry->action_type == DDL_LOG_REPLACE_ACTION)
Bug fixes + debug stuff
2006-02-10 05:57:54 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DBUG_ASSERT(strlen(ddl_log_entry->from_name) < FN_LEN);
WL 2826: Error handling of ALTER TABLE for partitioning Final review fixes
2006-04-21 14:35:25 +02:00
strmake(&global_ddl_log.file_entry_buf[DDL_LOG_NAME_POS + FN_LEN],
ddl_log_entry->from_name, FN_LEN - 1);
Bug fixes + debug stuff
2006-02-10 05:57:54 +01:00
}
WL 2826: Error handling of ALTER TABLE for partitioning A number of fixes
2006-02-09 20:13:22 +01:00
else
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
global_ddl_log.file_entry_buf[DDL_LOG_NAME_POS + FN_LEN]= 0;
DBUG_ASSERT(strlen(ddl_log_entry->handler_name) < FN_LEN);
WL 2826: Error handling of ALTER TABLE for partitioning Final review fixes
2006-04-21 14:35:25 +02:00
strmake(&global_ddl_log.file_entry_buf[DDL_LOG_NAME_POS + (2*FN_LEN)],
ddl_log_entry->handler_name, FN_LEN - 1);
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (get_free_ddl_log_entry(active_entry, &write_header))
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
DBUG_RETURN(TRUE);
}
error= FALSE;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (write_ddl_log_file_entry((*active_entry)->entry_pos))
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
{
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
error= TRUE;
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
sql_print_error("Failed to write entry_no = %u",
(*active_entry)->entry_pos);
}
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
if (write_header && !error)
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
VOID(sync_ddl_log());
if (write_ddl_log_header())
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
error= TRUE;
}
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
if (error)
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
release_ddl_log_memory_entry(*active_entry);
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
DBUG_RETURN(error);
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
}
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Write final entry in the ddl log
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
write_execute_ddl_log_entry()
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
first_entry First entry in linked list of entries
to execute, if 0 = NULL it means that
the entry is removed and the entries
are put into the free list.
WL 2826: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
complete Flag indicating we are simply writing
info about that entry has been completed
in:out:active_entry Entry to execute, 0 = NULL if the entry
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
is written first time and needs to be
returned. In this case the entry written
is returned in this parameter
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
RETURN VALUES
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
TRUE Error
FALSE Success
DESCRIPTION
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
This is the last write in the ddl log. The previous log entries have
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
already been written but not yet synched to disk.
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
We write a couple of log entries that describes action to perform.
This entries are set-up in a linked list, however only when a first
execute entry is put as the first entry these will be executed.
This routine writes this first
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
*/
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
bool write_execute_ddl_log_entry(uint first_entry,
bool complete,
DDL_LOG_MEMORY_ENTRY **active_entry)
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning New function to set exec_log_entry Moving code outside lock of table log Fixing some initialisations of list reference to avoid releasing wrong parts
2006-02-10 12:30:51 +01:00
bool write_header= FALSE;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
char *file_entry_buf= (char*)global_ddl_log.file_entry_buf;
DBUG_ENTER("write_execute_ddl_log_entry");
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (init_ddl_log())
{
DBUG_RETURN(TRUE);
}
WL 2826: Error handling of ALTER TABLE for partitioning A little more complete handling
2006-02-09 20:20:21 +01:00
if (!complete)
{
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
/*
We haven't synched the log entries yet, we synch them now before
writing the execute entry. If complete is true we haven't written
any log entries before, we are only here to write the execute
entry to indicate it is done.
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
VOID(sync_ddl_log());
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
file_entry_buf[DDL_LOG_ENTRY_TYPE_POS]= (char)DDL_LOG_EXECUTE_CODE;
WL 2826: Error handling of ALTER TABLE for partitioning A little more complete handling
2006-02-09 20:20:21 +01:00
}
else
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
file_entry_buf[DDL_LOG_ENTRY_TYPE_POS]= (char)DDL_IGNORE_LOG_ENTRY_CODE;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
file_entry_buf[DDL_LOG_ACTION_TYPE_POS]= 0; /* Ignored for execute entries */
file_entry_buf[DDL_LOG_PHASE_POS]= 0;
int4store(&file_entry_buf[DDL_LOG_NEXT_ENTRY_POS], first_entry);
file_entry_buf[DDL_LOG_NAME_POS]= 0;
file_entry_buf[DDL_LOG_NAME_POS + FN_LEN]= 0;
file_entry_buf[DDL_LOG_NAME_POS + 2*FN_LEN]= 0;
WL 2826: Error handling of ALTER TABLE for partitioning A number of fixes
2006-02-09 20:13:22 +01:00
if (!(*active_entry))
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (get_free_ddl_log_entry(active_entry, &write_header))
WL 2826: Error handling of ALTER TABLE for partitioning A number of fixes
2006-02-09 20:13:22 +01:00
{
DBUG_RETURN(TRUE);
}
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (write_ddl_log_file_entry((*active_entry)->entry_pos))
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
sql_print_error("Error writing execute entry in ddl log");
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
release_ddl_log_memory_entry(*active_entry);
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
DBUG_RETURN(TRUE);
}
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
VOID(sync_ddl_log());
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
if (write_header)
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (write_ddl_log_header())
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
release_ddl_log_memory_entry(*active_entry);
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
DBUG_RETURN(TRUE);
}
}
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
DBUG_RETURN(FALSE);
}
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
For complex rename operations we need to deactivate individual entries.
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
deactivate_ddl_log_entry()
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
entry_no Entry position of record to change
RETURN VALUES
TRUE Error
FALSE Success
DESCRIPTION
During replace operations where we start with an existing table called
t1 and a replacement table called t1#temp or something else and where
we want to delete t1 and rename t1#temp to t1 this is not possible to
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
do in a safe manner unless the ddl log is informed of the phases in
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
the change.
Delete actions are 1-phase actions that can be ignored immediately after
being executed.
Rename actions from x to y is also a 1-phase action since there is no
interaction with any other handlers named x and y.
Replace action where drop y and x -> y happens needs to be a two-phase
action. Thus the first phase will drop y and the second phase will
rename x -> y.
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
bool deactivate_ddl_log_entry(uint entry_no)
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
char *file_entry_buf= (char*)global_ddl_log.file_entry_buf;
DBUG_ENTER("deactivate_ddl_log_entry");
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (!read_ddl_log_file_entry(entry_no))
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (file_entry_buf[DDL_LOG_ENTRY_TYPE_POS] == DDL_LOG_ENTRY_CODE)
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (file_entry_buf[DDL_LOG_ACTION_TYPE_POS] == DDL_LOG_DELETE_ACTION ||
file_entry_buf[DDL_LOG_ACTION_TYPE_POS] == DDL_LOG_RENAME_ACTION ||
(file_entry_buf[DDL_LOG_ACTION_TYPE_POS] == DDL_LOG_REPLACE_ACTION &&
file_entry_buf[DDL_LOG_PHASE_POS] == 1))
file_entry_buf[DDL_LOG_ENTRY_TYPE_POS]= DDL_IGNORE_LOG_ENTRY_CODE;
else if (file_entry_buf[DDL_LOG_ACTION_TYPE_POS] == DDL_LOG_REPLACE_ACTION)
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DBUG_ASSERT(file_entry_buf[DDL_LOG_PHASE_POS] == 0);
file_entry_buf[DDL_LOG_PHASE_POS]= 1;
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
}
else
{
DBUG_ASSERT(0);
}
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
if (write_ddl_log_file_entry(entry_no))
{
sql_print_error("Error in deactivating log entry. Position = %u",
entry_no);
DBUG_RETURN(TRUE);
}
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
}
}
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
else
{
sql_print_error("Failed in reading entry before deactivating it");
DBUG_RETURN(TRUE);
}
DBUG_RETURN(FALSE);
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
}
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Sync ddl log file
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
sync_ddl_log()
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
RETURN VALUES
TRUE Error
FALSE Success
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
bool sync_ddl_log()
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
{
bool error= FALSE;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DBUG_ENTER("sync_ddl_log");
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
if ((!global_ddl_log.recovery_phase) &&
init_ddl_log())
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
{
DBUG_RETURN(TRUE);
}
if (my_sync(global_ddl_log.file_id, MYF(0)))
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
{
/* Write to error log */
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
sql_print_error("Failed to sync ddl log");
WL 2826: Error handling of ALTER TABLE for partitioning Make lots of fixes to handle the most complex case of reorganising of partitions where two-phased processes are needed in some cases.
2006-02-10 22:36:01 +01:00
error= TRUE;
}
DBUG_RETURN(error);
}
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
/*
Release a log memory entry
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
release_ddl_log_memory_entry()
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
log_memory_entry Log memory entry to release
RETURN VALUES
NONE
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
void release_ddl_log_memory_entry(DDL_LOG_MEMORY_ENTRY *log_entry)
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DDL_LOG_MEMORY_ENTRY *first_free= global_ddl_log.first_free;
DDL_LOG_MEMORY_ENTRY *next_log_entry= log_entry->next_log_entry;
DDL_LOG_MEMORY_ENTRY *prev_log_entry= log_entry->prev_log_entry;
DBUG_ENTER("release_ddl_log_memory_entry");
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
global_ddl_log.first_free= log_entry;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
log_entry->next_log_entry= first_free;
if (prev_log_entry)
prev_log_entry->next_log_entry= next_log_entry;
else
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
global_ddl_log.first_used= next_log_entry;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
if (next_log_entry)
next_log_entry->prev_log_entry= prev_log_entry;
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
DBUG_VOID_RETURN;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Execute one entry in the ddl log. Executing an entry means executing
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
a linked list of actions.
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
execute_ddl_log_entry()
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
first_entry Reference to first action in entry
RETURN VALUES
TRUE Error
FALSE Success
*/
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
bool execute_ddl_log_entry(THD *thd, uint first_entry)
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DDL_LOG_ENTRY ddl_log_entry;
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
uint read_entry= first_entry;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DBUG_ENTER("execute_ddl_log_entry");
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
pthread_mutex_lock(&LOCK_gdl);
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
do
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (read_ddl_log_entry(read_entry, &ddl_log_entry))
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
/* Write to error log and continue with next log entry */
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
sql_print_error("Failed to read entry = %u from ddl log",
read_entry);
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
break;
}
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DBUG_ASSERT(ddl_log_entry.entry_type == DDL_LOG_ENTRY_CODE ||
ddl_log_entry.entry_type == DDL_IGNORE_LOG_ENTRY_CODE);
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
if (execute_ddl_log_action(thd, &ddl_log_entry))
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
{
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
/* Write to error log and continue with next log entry */
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
sql_print_error("Failed to execute action for entry = %u from ddl log",
read_entry);
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
break;
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
}
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
read_entry= ddl_log_entry.next_entry;
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
} while (read_entry);
WL 2826: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
pthread_mutex_unlock(&LOCK_gdl);
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
DBUG_RETURN(FALSE);
}
WL 2826: Error handling of ALTER TABLE for partitioning Added action code for recovery
2006-02-14 11:08:58 +01:00
Close the DDL log from 2 places => put into separate method Needed to close ddl log in recovery before deleting it
2006-09-19 17:40:21 +02:00
/*
Close the ddl log
SYNOPSIS
close_ddl_log()
RETURN VALUES
NONE
*/
static void close_ddl_log()
{
DBUG_ENTER("close_ddl_log");
if (global_ddl_log.file_id >= 0)
{
VOID(my_close(global_ddl_log.file_id, MYF(MY_WME)));
global_ddl_log.file_id= (File) -1;
}
DBUG_VOID_RETURN;
}
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Execute the ddl log at recovery of MySQL Server
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
execute_ddl_log_recovery()
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
RETURN VALUES
NONE
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
void execute_ddl_log_recovery()
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
uint num_entries, i;
Fixes
2006-04-03 18:26:35 +02:00
THD *thd;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DDL_LOG_ENTRY ddl_log_entry;
Fixes
2006-04-03 18:26:35 +02:00
char file_name[FN_REFLEN];
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DBUG_ENTER("execute_ddl_log_recovery");
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
BUG#19313: Crash after mysqld starts and stops immediately afterwards.
2006-04-25 23:33:54 +02:00
/*
Initialise global_ddl_log struct
*/
bzero(global_ddl_log.file_entry_buf, sizeof(global_ddl_log.file_entry_buf));
global_ddl_log.inited= FALSE;
global_ddl_log.recovery_phase= TRUE;
global_ddl_log.io_size= IO_SIZE;
After merge fixes Remove compiler warnings Fix some broken tests Workaround for syncronization bug in NDB (Bug #16445)
2006-05-04 05:28:24 +02:00
global_ddl_log.file_id= (File) -1;
BUG#19313: Crash after mysqld starts and stops immediately afterwards.
2006-04-25 23:33:54 +02:00
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
/*
To be able to run this from boot, we allocate a temporary THD
*/
if (!(thd=new THD))
DBUG_VOID_RETURN;
thd->thread_stack= (char*) &thd;
thd->store_globals();
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
num_entries= read_ddl_log_header();
WL 2826: Error handling of ALTER TABLE for partitioning Various bug fixes (mostly mixed to and from in replace methods)
2006-04-20 02:04:00 +02:00
for (i= 1; i < num_entries + 1; i++)
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (read_ddl_log_entry(i, &ddl_log_entry))
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
sql_print_error("Failed to read entry no = %u from ddl log",
i);
continue;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
if (ddl_log_entry.entry_type == DDL_LOG_EXECUTE_CODE)
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
{
Fixes
2006-04-03 18:26:35 +02:00
if (execute_ddl_log_entry(thd, ddl_log_entry.next_entry))
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
/* Real unpleasant scenario but we continue anyways. */
continue;
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
}
}
}
Close the DDL log from 2 places => put into separate method Needed to close ddl log in recovery before deleting it
2006-09-19 17:40:21 +02:00
close_ddl_log();
WL 2826: Error handling of ALTER TABLE for partitioning More review fixes
2006-04-01 16:37:36 +02:00
create_ddl_log_file_name(file_name);
VOID(my_delete(file_name, MYF(0)));
global_ddl_log.recovery_phase= FALSE;
delete thd;
/* Remember that we don't have a THD */
my_pthread_setspecific_ptr(THR_THD, 0);
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
DBUG_VOID_RETURN;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
/*
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
Release all memory allocated to the ddl log
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
SYNOPSIS
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
release_ddl_log()
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
RETURN VALUES
NONE
*/
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
void release_ddl_log()
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DDL_LOG_MEMORY_ENTRY *free_list= global_ddl_log.first_free;
DDL_LOG_MEMORY_ENTRY *used_list= global_ddl_log.first_used;
DBUG_ENTER("release_ddl_log");
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
WL 2826: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
pthread_mutex_lock(&LOCK_gdl);
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
while (used_list)
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DDL_LOG_MEMORY_ENTRY *tmp= used_list->next_log_entry;
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
my_free((char*)used_list, MYF(0));
Don't use free'd memory, bad idea Free all memory with mutex and destroy mutex as final step
2006-02-10 19:33:08 +01:00
used_list= tmp;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
while (free_list)
{
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
DDL_LOG_MEMORY_ENTRY *tmp= free_list->next_log_entry;
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
my_free((char*)free_list, MYF(0));
Don't use free'd memory, bad idea Free all memory with mutex and destroy mutex as final step
2006-02-10 19:33:08 +01:00
free_list= tmp;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
Close the DDL log from 2 places => put into separate method Needed to close ddl log in recovery before deleting it
2006-09-19 17:40:21 +02:00
close_ddl_log();
After merge fixes Remove compiler warnings Fix some broken tests Workaround for syncronization bug in NDB (Bug #16445)
2006-05-04 05:28:24 +02:00
global_ddl_log.inited= 0;
WL 2826: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
pthread_mutex_unlock(&LOCK_gdl);
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
VOID(pthread_mutex_destroy(&LOCK_gdl));
WL 2826: Error handling for ALTER TABLE for partitioning Step 13 Lots of compilation fixes
2006-02-09 00:04:58 +01:00
DBUG_VOID_RETURN;
WL 2826: Error handling for ALTER TABLE for partitioning Most of the code for handling the table log is in place now, except the action part at recovery and proper error handling in some places.
2006-02-08 15:14:15 +01:00
}
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
/*
---------------------------------------------------------------------------
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
END MODULE DDL log
WL 2826: Step 11 Lots of new code for table log
2006-02-07 10:45:07 +01:00
--------------------
---------------------------------------------------------------------------
*/
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
/*
SYNOPSIS
mysql_write_frm()
lpt Struct carrying many parameters needed for this
method
flags Flags as defined below
WFRM_INITIAL_WRITE If set we need to prepare table before
creating the frm file
WFRM_CREATE_HANDLER_FILES If set we need to create the handler file as
part of the creation of the frm file
WFRM_PACK_FRM If set we should pack the frm file and delete
the frm file
RETURN VALUES
TRUE Error
FALSE Success
DESCRIPTION
A support method that creates a new frm file and in this process it
regenerates the partition data. It works fine also for non-partitioned
tables since it only handles partitioned data if it exists.
*/
bool mysql_write_frm(ALTER_PARTITION_PARAM_TYPE *lpt, uint flags)
{
/*
Prepare table to prepare for writing a new frm file where the
partitions in add/drop state have temporarily changed their state
We set tmp_table to avoid get errors on naming of primary key index.
*/
int error= 0;
char path[FN_REFLEN+1];
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
char shadow_path[FN_REFLEN+1];
char shadow_frm_name[FN_REFLEN+1];
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
char frm_name[FN_REFLEN+1];
DBUG_ENTER("mysql_write_frm");
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
/*
Build shadow frm file name
*/
WL #2826: Sixth step, made it work with test cases
2006-02-03 18:05:29 +01:00
build_table_filename(shadow_path, sizeof(shadow_path), lpt->db,
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
lpt->table_name, "#", 0);
WL #2826: Sixth step, made it work with test cases
2006-02-03 18:05:29 +01:00
strxmov(shadow_frm_name, shadow_path, reg_ext, NullS);
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
if (flags & WFRM_WRITE_SHADOW)
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
{
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
if (mysql_copy_create_list(lpt->create_list,
&lpt->new_create_list) ||
mysql_copy_key_list(lpt->key_list,
&lpt->new_key_list) ||
mysql_prepare_table(lpt->thd, lpt->create_info,
&lpt->new_create_list,
&lpt->new_key_list,
/*tmp_table*/ 1,
&lpt->db_options,
lpt->table->file,
&lpt->key_info_buffer,
&lpt->key_count,
WL #2826: Sixth step, made it work with test cases
2006-02-03 18:05:29 +01:00
/*select_field_count*/ 0))
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
{
DBUG_RETURN(TRUE);
}
#ifdef WITH_PARTITION_STORAGE_ENGINE
{
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
partition_info *part_info= lpt->table->part_info;
char *part_syntax_buf;
uint syntax_len;
if (part_info)
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
{
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
if (!(part_syntax_buf= generate_partition_syntax(part_info,
&syntax_len,
Merge c-0409e253.1238-1-64736c10.cust.bredbandsbolaget.se:/home/pappa/clean-mysql-5.1 into c-0409e253.1238-1-64736c10.cust.bredbandsbolaget.se:/home/pappa/bug19695
2006-06-12 20:42:07 +02:00
TRUE, TRUE)))
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
{
DBUG_RETURN(TRUE);
}
part_info->part_info_string= part_syntax_buf;
part_info->part_info_len= syntax_len;
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
}
}
#endif
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
/* Write shadow frm file */
lpt->create_info->table_options= lpt->db_options;
if ((mysql_create_frm(lpt->thd, shadow_frm_name, lpt->db,
lpt->table_name, lpt->create_info,
lpt->new_create_list, lpt->key_count,
lpt->key_info_buffer, lpt->table->file)) ||
WL 2826: Error handling of ALTER TABLE for partitioning Fixed up create_handler_files call
2006-04-20 03:43:30 +02:00
lpt->table->file->create_handler_files(shadow_path, NULL,
CHF_CREATE_FLAG,
Merge c-870ae253.1238-1-64736c10.cust.bredbandsbolaget.se:/home/pappa/clean-mysql-5.1-new into c-870ae253.1238-1-64736c10.cust.bredbandsbolaget.se:/home/pappa/wl2826
2006-04-20 03:22:35 +02:00
lpt->create_info))
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
{
my_delete(shadow_frm_name, MYF(0));
error= 1;
goto end;
}
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
}
if (flags & WFRM_PACK_FRM)
{
/*
We need to pack the frm file and after packing it we delete the
frm file to ensure it doesn't get used. This is only used for
handlers that have the main version of the frm file stored in the
handler.
*/
const void *data= 0;
uint length= 0;
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
if (readfrm(shadow_path, &data, &length) ||
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
packfrm(data, length, &lpt->pack_frm_data, &lpt->pack_frm_len))
{
my_free((char*)data, MYF(MY_ALLOW_ZERO_PTR));
my_free((char*)lpt->pack_frm_data, MYF(MY_ALLOW_ZERO_PTR));
mem_alloc_error(length);
error= 1;
goto end;
}
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
error= my_delete(shadow_frm_name, MYF(MY_WME));
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
}
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
if (flags & WFRM_INSTALL_SHADOW)
{
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
partition_info *part_info= lpt->part_info;
#endif
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
/*
Build frm file name
*/
build_table_filename(path, sizeof(path), lpt->db,
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
lpt->table_name, "", 0);
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
strxmov(frm_name, path, reg_ext, NullS);
/*
When we are changing to use new frm file we need to ensure that we
don't collide with another thread in process to open the frm file.
WL 2826: Error handling of ALTER TABLE for partitioning Use \0 to indicate frm-file Add more comments Add more #ifdef code missing previously
2006-04-18 16:58:12 +02:00
We start by deleting the .frm file and possible .par file. Then we
write to the DDL log that we have completed the delete phase by
increasing the phase of the log entry. Next step is to rename the
new .frm file and the new .par file to the real name. After
completing this we write a new phase to the log entry that will
deactivate it.
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
*/
VOID(pthread_mutex_lock(&LOCK_open));
if (my_delete(frm_name, MYF(MY_WME)) ||
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
WL 2826: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
lpt->table->file->create_handler_files(path, shadow_path,
WL 2826: Error handling of ALTER TABLE for partitioning Fixed up create_handler_files call
2006-04-20 03:43:30 +02:00
CHF_DELETE_FLAG, NULL) ||
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
deactivate_ddl_log_entry(part_info->frm_log_entry->entry_pos) ||
(sync_ddl_log(), FALSE) ||
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
#endif
WL 2826: Error handling of ALTER TABLE for partitioning Use \0 to indicate frm-file Add more comments Add more #ifdef code missing previously
2006-04-18 16:58:12 +02:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
my_rename(shadow_frm_name, frm_name, MYF(MY_WME)) ||
WL 2826: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
lpt->table->file->create_handler_files(path, shadow_path,
WL 2826: Error handling of ALTER TABLE for partitioning Fixed up create_handler_files call
2006-04-20 03:43:30 +02:00
CHF_RENAME_FLAG, NULL))
WL 2826: Error handling of ALTER TABLE for partitioning Use \0 to indicate frm-file Add more comments Add more #ifdef code missing previously
2006-04-18 16:58:12 +02:00
#else
my_rename(shadow_frm_name, frm_name, MYF(MY_WME)))
#endif
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
{
error= 1;
}
VOID(pthread_mutex_unlock(&LOCK_open));
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
deactivate_ddl_log_entry(part_info->frm_log_entry->entry_pos);
WL 2826: Error handling of ALTER TABLE for partitioning Introduce log entry to handle replace of frm file
2006-02-13 13:52:23 +01:00
part_info->frm_log_entry= NULL;
WL 2826: Error handling of ALTER TABLE for partitioning Loads of review comments fixed inactivate => deactivate table log => ddl log Commented on Error Inject Module added Put various #defines into enums Fixed abort_and_upgrade_lock, removed unnecessary parameter Fixed mysqlish method intro's Fixed warning statements 5.1.7 was released still with partition states in clear text Fixed io_size bug Fixed bug in open that TRUNCATED before reading :) file_entry => file_entry_buf Don't open DDL log until first write call to DDL log handler_type => handler_name no => num
2006-03-25 00:19:13 +01:00
VOID(sync_ddl_log());
WL 2826: Error handling of ALTER TABLE for partitioning Fix compiler issues
2006-02-20 21:07:03 +01:00
#endif
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
}
WL 2826: First step in error handling of ALTER TABLE for partitioning
2006-02-01 10:06:07 +01:00
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
end:
DBUG_RETURN(error);
}
/*
SYNOPSIS
write_bin_log()
thd Thread object
clear_error is clear_error to be called
query Query to log
query_length Length of query
RETURN VALUES
NONE
DESCRIPTION
Write the binlog if open, routine used in multiple places in this
file
*/
void write_bin_log(THD *thd, bool clear_error,
char const *query, ulong query_length)
{
if (mysql_bin_log.is_open())
{
if (clear_error)
thd->clear_error();
thd->binlog_query(THD::STMT_QUERY_TYPE,
query, query_length, FALSE, FALSE);
}
}
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
Fixed compilation error on windows
2003-01-29 18:10:33 +01:00
/*
delete (drop) tables.
SYNOPSIS
mysql_rm_table()
thd Thread handle
tables List of tables to delete
if_exists If 1, don't give error if one table doesn't exists
NOTES
Will delete all tables that can be deleted and give a compact error
messages for tables that could not be deleted.
If a table is in use, we will wait for all users to free the table
before dropping it
Wait if global_read_lock (FLUSH TABLES WITH READ LOCK) is set.
RETURN
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
FALSE OK. In this case ok packet is sent to user
TRUE Error
Fixed compilation error on windows
2003-01-29 18:10:33 +01:00
*/
Import changeset
2000-07-31 21:29:14 +02:00
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool mysql_rm_table(THD *thd,TABLE_LIST *tables, my_bool if_exists,
my_bool drop_temporary)
Import changeset
2000-07-31 21:29:14 +02:00
{
global read lock code now uses a dedicated mutex (otherwise a deadlock when ALTER writes to binlog holding LOCK_open, it causes binlog rotation, binlog waits for prepared transactions to commit, and commit needs LOCK_open to check for global read lock)
2005-03-16 08:40:19 +01:00
bool error= FALSE, need_start_waiters= FALSE;
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_ENTER("mysql_rm_table");
/* mark for close and remove all cached entries */
global read lock code now uses a dedicated mutex (otherwise a deadlock when ALTER writes to binlog holding LOCK_open, it causes binlog rotation, binlog waits for prepared transactions to commit, and commit needs LOCK_open to check for global read lock)
2005-03-16 08:40:19 +01:00
if (!drop_temporary)
Portability fixes
2000-11-18 01:15:06 +01:00
{
global read lock code now uses a dedicated mutex (otherwise a deadlock when ALTER writes to binlog holding LOCK_open, it causes binlog rotation, binlog waits for prepared transactions to commit, and commit needs LOCK_open to check for global read lock)
2005-03-16 08:40:19 +01:00
if ((error= wait_if_global_read_lock(thd, 0, 1)))
fix for FLUSH TABLES WITH READ LOCK
2000-11-04 21:58:07 +01:00
{
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
my_error(ER_TABLE_NOT_LOCKED_FOR_WRITE, MYF(0), tables->table_name);
Bug#14397 - OPTIMIZE TABLE with an open HANDLER causes a crash Version for 5.0. It fixes three problems: 1. The cause of the bug was that we did not check the table version for the HANDLER ... READ commands. We did not notice when a table was replaced by a new one. This can happen during ALTER TABLE, REPAIR TABLE, and OPTIMIZE TABLE (there might be more cases). I call the fix for this problem "the primary bug fix". 2. mysql_ha_flush() was not always called with a locked LOCK_open. Though the function comment clearly said it must. I changed the code so that the locking is done when required. I call the fix for this problem "the secondary fix". 3. In 5.0 (not in 4.1 or 4.0) DROP TABLE had a possible deadlock flaw in concur with FLUSH TABLES WITH READ LOCK. I call the fix for this problem "the 5.0 addendum fix".
2005-11-15 21:57:02 +01:00
DBUG_RETURN(TRUE);
Portability fixes
2000-11-18 01:15:06 +01:00
}
global read lock code now uses a dedicated mutex (otherwise a deadlock when ALTER writes to binlog holding LOCK_open, it causes binlog rotation, binlog waits for prepared transactions to commit, and commit needs LOCK_open to check for global read lock)
2005-03-16 08:40:19 +01:00
else
need_start_waiters= TRUE;
Portability fixes
2000-11-18 01:15:06 +01:00
}
Bug#14397 - OPTIMIZE TABLE with an open HANDLER causes a crash Version for 5.0. It fixes three problems: 1. The cause of the bug was that we did not check the table version for the HANDLER ... READ commands. We did not notice when a table was replaced by a new one. This can happen during ALTER TABLE, REPAIR TABLE, and OPTIMIZE TABLE (there might be more cases). I call the fix for this problem "the primary bug fix". 2. mysql_ha_flush() was not always called with a locked LOCK_open. Though the function comment clearly said it must. I changed the code so that the locking is done when required. I call the fix for this problem "the secondary fix". 3. In 5.0 (not in 4.1 or 4.0) DROP TABLE had a possible deadlock flaw in concur with FLUSH TABLES WITH READ LOCK. I call the fix for this problem "the 5.0 addendum fix".
2005-11-15 21:57:02 +01:00
/*
Acquire LOCK_open after wait_if_global_read_lock(). If we would hold
LOCK_open during wait_if_global_read_lock(), other threads could not
close their tables. This would make a pretty deadlock.
*/
thd->mysys_var->current_mutex= &LOCK_open;
thd->mysys_var->current_cond= &COND_refresh;
VOID(pthread_mutex_lock(&LOCK_open));
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
error= mysql_rm_table_part2(thd, tables, if_exists, drop_temporary, 0, 0);
Changed DB_TYPE_INNOBASE to DB_TYPE_INNODB Fix that DROP DATABASE works with all table types Use BULK_INSERT when inserting more than one row Better TRUNCATE TABLE
2001-09-02 12:47:00 +02:00
pthread_mutex_unlock(&LOCK_open);
pthread_mutex_lock(&thd->mysys_var->mutex);
thd->mysys_var->current_mutex= 0;
thd->mysys_var->current_cond= 0;
pthread_mutex_unlock(&thd->mysys_var->mutex);
global read lock code now uses a dedicated mutex (otherwise a deadlock when ALTER writes to binlog holding LOCK_open, it causes binlog rotation, binlog waits for prepared transactions to commit, and commit needs LOCK_open to check for global read lock)
2005-03-16 08:40:19 +01:00
if (need_start_waiters)
start_waiting_global_read_lock(thd);
Changed DB_TYPE_INNOBASE to DB_TYPE_INNODB Fix that DROP DATABASE works with all table types Use BULK_INSERT when inserting more than one row Better TRUNCATE TABLE
2001-09-02 12:47:00 +02:00
if (error)
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
Fixes and code cleanups after merge with 4.0.3 Warning handling and initial prepared statement handling (last not complete yet) Changed a lot of functions that returned 0/1 to my_bool type. GRANT handling now uses read/write locks instead of mutex Change basic net functions to use THD instead of NET (needed for 4.1 protocol) Use my_sprintf instead of sprintf() + strlen() Added alloc_query() to be able to chare query initialization code with prepared statements. Cleanup handling of SHOW COUNT(*) WARNINGS and SELECT LAST_INSERT_ID() Note that the following test fails (will be fixed ASAP): sub_select, union, rpl_rotate_logs and rpl_mystery22
2002-10-02 12:33:08 +02:00
send_ok(thd);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(FALSE);
Changed DB_TYPE_INNOBASE to DB_TYPE_INNODB Fix that DROP DATABASE works with all table types Use BULK_INSERT when inserting more than one row Better TRUNCATE TABLE
2001-09-02 12:47:00 +02:00
}
Fixed compilation error on windows
2003-01-29 18:10:33 +01:00
/*
delete (drop) tables.
SYNOPSIS
Generate warning in ha_delete_table() if files is missing in handler
2005-02-21 19:41:48 +01:00
mysql_rm_table_part2_with_lock()
thd Thread handle
tables List of tables to delete
if_exists If 1, don't give error if one table doesn't exists
dont_log_query Don't write query to log files. This will also not
global read lock code now uses a dedicated mutex (otherwise a deadlock when ALTER writes to binlog holding LOCK_open, it causes binlog rotation, binlog waits for prepared transactions to commit, and commit needs LOCK_open to check for global read lock)
2005-03-16 08:40:19 +01:00
generate warnings if the handler files doesn't exists
Fixed compilation error on windows
2003-01-29 18:10:33 +01:00
NOTES
Works like documented in mysql_rm_table(), but don't check
global_read_lock and don't send_ok packet to server.
RETURN
0 ok
1 error
*/
New SET syntax & system variables. Made a some new buffers thread specific and changeable. Resize of key_buffer. AUTO_COMMIT -> AUTOCOMMIT Fixed mutex bug in DROP DATABASE Fixed bug when using auto_increment as second part of a key where first part could include NULL. Split handler->extra() to extra() and extra_opt() to be able to support thread specific buffers. Don't write message to error log when slave reconnects becasue of timeout. Fixed possible update problem when using DELETE/UPDATE on small tables (In some cases we used index even if table scanning would be better) A lot of minior code cleanups
2002-07-23 17:31:22 +02:00
int mysql_rm_table_part2_with_lock(THD *thd,
TABLE_LIST *tables, bool if_exists,
Added support for DROP TEMPORARY TABLE Removed mysql_warnings() API function. Post merge fixes.
2003-01-04 14:37:20 +01:00
bool drop_temporary, bool dont_log_query)
New SET syntax & system variables. Made a some new buffers thread specific and changeable. Resize of key_buffer. AUTO_COMMIT -> AUTOCOMMIT Fixed mutex bug in DROP DATABASE Fixed bug when using auto_increment as second part of a key where first part could include NULL. Split handler->extra() to extra() and extra_opt() to be able to support thread specific buffers. Don't write message to error log when slave reconnects becasue of timeout. Fixed possible update problem when using DELETE/UPDATE on small tables (In some cases we used index even if table scanning would be better) A lot of minior code cleanups
2002-07-23 17:31:22 +02:00
{
int error;
thd->mysys_var->current_mutex= &LOCK_open;
thd->mysys_var->current_cond= &COND_refresh;
VOID(pthread_mutex_lock(&LOCK_open));
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
error= mysql_rm_table_part2(thd, tables, if_exists, drop_temporary, 1,
dont_log_query);
New SET syntax & system variables. Made a some new buffers thread specific and changeable. Resize of key_buffer. AUTO_COMMIT -> AUTOCOMMIT Fixed mutex bug in DROP DATABASE Fixed bug when using auto_increment as second part of a key where first part could include NULL. Split handler->extra() to extra() and extra_opt() to be able to support thread specific buffers. Don't write message to error log when slave reconnects becasue of timeout. Fixed possible update problem when using DELETE/UPDATE on small tables (In some cases we used index even if table scanning would be better) A lot of minior code cleanups
2002-07-23 17:31:22 +02:00
pthread_mutex_unlock(&LOCK_open);
pthread_mutex_lock(&thd->mysys_var->mutex);
thd->mysys_var->current_mutex= 0;
thd->mysys_var->current_cond= 0;
pthread_mutex_unlock(&thd->mysys_var->mutex);
return error;
}
Fixed compilation error on windows
2003-01-29 18:10:33 +01:00
Put temporary files in binlog cache when using BEGIN/COMMIT Let MySQL 4.0 read 4.1 .frm files without 4.1 specific extensions New variables @@rand_seed1 and @@rand_seed2 (used by replication) DROP TEMPORARY TABLE
2002-11-07 03:02:37 +01:00
/*
Added support for DROP TEMPORARY TABLE Removed mysql_warnings() API function. Post merge fixes.
2003-01-04 14:37:20 +01:00
Execute the drop of a normal or temporary table
SYNOPSIS
mysql_rm_table_part2()
thd Thread handler
tables Tables to drop
if_exists If set, don't give an error if table doesn't exists.
In this case we give an warning of level 'NOTE'
drop_temporary Only drop temporary tables
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
drop_view Allow to delete VIEW .frm
Generate warning in ha_delete_table() if files is missing in handler
2005-02-21 19:41:48 +01:00
dont_log_query Don't write query to log files. This will also not
generate warnings if the handler files doesn't exists
Added support for DROP TEMPORARY TABLE Removed mysql_warnings() API function. Post merge fixes.
2003-01-04 14:37:20 +01:00
Put temporary files in binlog cache when using BEGIN/COMMIT Let MySQL 4.0 read 4.1 .frm files without 4.1 specific extensions New variables @@rand_seed1 and @@rand_seed2 (used by replication) DROP TEMPORARY TABLE
2002-11-07 03:02:37 +01:00
TODO:
When logging to the binary log, we should log
tmp_tables and transactional tables as separate statements if we
are in a transaction; This is needed to get these tables into the
cached binary log that is only written on COMMIT.
The current code only writes DROP statements that only uses temporary
tables to the cache binary log. This should be ok on most cases, but
not all.
Added support for DROP TEMPORARY TABLE Removed mysql_warnings() API function. Post merge fixes.
2003-01-04 14:37:20 +01:00
RETURN
0 ok
1 Error
-1 Thread was killed
Put temporary files in binlog cache when using BEGIN/COMMIT Let MySQL 4.0 read 4.1 .frm files without 4.1 specific extensions New variables @@rand_seed1 and @@rand_seed2 (used by replication) DROP TEMPORARY TABLE
2002-11-07 03:02:37 +01:00
*/
Changed DB_TYPE_INNOBASE to DB_TYPE_INNODB Fix that DROP DATABASE works with all table types Use BULK_INSERT when inserting more than one row Better TRUNCATE TABLE
2001-09-02 12:47:00 +02:00
int mysql_rm_table_part2(THD *thd, TABLE_LIST *tables, bool if_exists,
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
bool drop_temporary, bool drop_view,
bool dont_log_query)
Changed DB_TYPE_INNOBASE to DB_TYPE_INNODB Fix that DROP DATABASE works with all table types Use BULK_INSERT when inserting more than one row Better TRUNCATE TABLE
2001-09-02 12:47:00 +02:00
{
TABLE_LIST *table;
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
char path[FN_REFLEN], *alias;
uint path_length;
Changed DB_TYPE_INNOBASE to DB_TYPE_INNODB Fix that DROP DATABASE works with all table types Use BULK_INSERT when inserting more than one row Better TRUNCATE TABLE
2001-09-02 12:47:00 +02:00
String wrong_tables;
int error;
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
int non_temp_tables_count= 0;
dict0dict.c: Remove redundant code; parse both the database name and the table name in a FOREIGN KEY constraint with quotes in mind row0mysql.c, ha_innodb.cc, sql_table.cc: Return error message Cannot delete or update a parent row... if we try to drop a table which is referenced by a FOREIGN KEY constraint, and the user has not set foreign_key_checks=0
2004-02-01 23:10:45 +01:00
bool some_tables_deleted=0, tmp_table_deleted=0, foreign_key_error=0;
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
String built_query;
Changed DB_TYPE_INNOBASE to DB_TYPE_INNODB Fix that DROP DATABASE works with all table types Use BULK_INSERT when inserting more than one row Better TRUNCATE TABLE
2001-09-02 12:47:00 +02:00
DBUG_ENTER("mysql_rm_table_part2");
Fixed compiler and valgrind warnings Added missing DBUG_xxx_RETURN statements Fixed some usage of not initialized variables (as found by valgrind) Ensure that we don't remove locked tables used as name locks from open table cache until unlock_table_names() are called. This was fixed by having drop_locked_name() returning any table used as a name lock so that we can free it in unlock_table_names() This will allow Tomas to continue with his work to use namelocks to syncronize things. Note: valgrind still produces a lot of warnings about using not initialized code and shows memory loss errors when running the ndb tests
2006-03-29 13:27:36 +02:00
LINT_INIT(alias);
LINT_INIT(path_length);
safe_mutex_assert_owner(&LOCK_open);
WL#2977 and WL#2712 global and session-level variable to set the binlog format (row/statement), and new binlog format called "mixed" (which is statement-based except if only row-based is correct, in this cset it means if UDF or UUID is used; more cases could be added in later 5.1 release): SET GLOBAL|SESSION BINLOG_FORMAT=row|statement|mixed|default; the global default is statement unless cluster is enabled (then it's row) as in 5.1-alpha. It's not possible to use SET on this variable if a session is currently in row-based mode and has open temporary tables (because CREATE TEMPORARY TABLE was not binlogged so temp table is not known on slave), or if NDB is enabled (because NDB does not support such change on-the-fly, though it will later), of if in a stored function (see below). The added tests test the possibility or impossibility to SET, their effects, and the mixed mode, including in prepared statements and in stored procedures and functions. Caveats: a) The mixed mode will not work for stored functions: in mixed mode, a stored function will always be binlogged as one call and in a statement-based way (e.g. INSERT VALUES(myfunc()) or SELECT myfunc()). b) for the same reason, changing the thread's binlog format inside a stored function is refused with an error message. c) the same problems apply to triggers; implementing b) for triggers will be done later (will ask Dmitri). Additionally, as the binlog format is now changeable by each user for his session, I remove the implication which was done at startup, where row-based automatically set log-bin-trust-routine-creators to 1 (not possible anymore as a user can now switch to stmt-based and do nasty things again), and automatically set --innodb-locks-unsafe-for-binlog to 1 (was anyway theoretically incorrect as it disabled phantom protection). Plus fixes for compiler warnings.
2006-02-25 22:21:03 +01:00
if (thd->current_stmt_binlog_row_based && !dont_log_query)
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
{
built_query.set_charset(system_charset_info);
if (if_exists)
built_query.append("DROP TABLE IF EXISTS ");
else
built_query.append("DROP TABLE ");
}
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
/*
If we have the table in the definition cache, we don't have to check the
.frm file to find if the table is a normal table (not view) and what
engine to use.
*/
for (table= tables; table; table= table->next_local)
{
TABLE_SHARE *share;
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
table->db_type= NULL;
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
if ((share= get_cached_table_share(table->db, table->table_name)))
table->db_type= share->db_type;
Fix Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on" Log tables rely on concurrent insert machinery to add data. This means that log tables are always opened and locked by special (artificial) logger threads. Because of this, the thread which tries to drop a log table starts to wait for the table to be unlocked. Which will happen only if the log table is disabled. Alike situation happens if one tries to alter a log table. However in addition to the problem above, alter table calls check_if_locking_is_allowed() routine for the engine. The routine does not allow alter for the log tables. So, alter doesn't start waiting forever for logs to be disabled, but returns with an error. Another problem is that not all engines could be used for the log tables. That's because they need concurrent insert. In this patch we: (1) Explicitly disallow to drop/alter a log table if it is currently used by the logger. (2) Update MyISAM to support log tables (3) Allow to drop log tables/alter log tables if log is disabled At the same time we (4) Disallow to alter log tables to unsupported engine (after this patch CSV and MyISAM are alowed) Recommit with review fixes.
2006-08-03 19:28:15 +02:00
/* Disable drop of enabled log tables */
if (share && share->log_table &&
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
check_if_log_table(table->db_length, table->db,
table->table_name_length, table->table_name, 1))
Fix Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on" Log tables rely on concurrent insert machinery to add data. This means that log tables are always opened and locked by special (artificial) logger threads. Because of this, the thread which tries to drop a log table starts to wait for the table to be unlocked. Which will happen only if the log table is disabled. Alike situation happens if one tries to alter a log table. However in addition to the problem above, alter table calls check_if_locking_is_allowed() routine for the engine. The routine does not allow alter for the log tables. So, alter doesn't start waiting forever for logs to be disabled, but returns with an error. Another problem is that not all engines could be used for the log tables. That's because they need concurrent insert. In this patch we: (1) Explicitly disallow to drop/alter a log table if it is currently used by the logger. (2) Update MyISAM to support log tables (3) Allow to drop log tables/alter log tables if log is disabled At the same time we (4) Disallow to alter log tables to unsupported engine (after this patch CSV and MyISAM are alowed) Recommit with review fixes.
2006-08-03 19:28:15 +02:00
{
Remove unused error messages (no release contains them at the moment, so we can safely do that). Update an error mesage to make it translateable.
2006-09-27 15:48:00 +02:00
my_error(ER_BAD_LOG_STATEMENT, MYF(0), "DROP");
Fix Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on" Log tables rely on concurrent insert machinery to add data. This means that log tables are always opened and locked by special (artificial) logger threads. Because of this, the thread which tries to drop a log table starts to wait for the table to be unlocked. Which will happen only if the log table is disabled. Alike situation happens if one tries to alter a log table. However in addition to the problem above, alter table calls check_if_locking_is_allowed() routine for the engine. The routine does not allow alter for the log tables. So, alter doesn't start waiting forever for logs to be disabled, but returns with an error. Another problem is that not all engines could be used for the log tables. That's because they need concurrent insert. In this patch we: (1) Explicitly disallow to drop/alter a log table if it is currently used by the logger. (2) Update MyISAM to support log tables (3) Allow to drop log tables/alter log tables if log is disabled At the same time we (4) Disallow to alter log tables to unsupported engine (after this patch CSV and MyISAM are alowed) Recommit with review fixes.
2006-08-03 19:28:15 +02:00
DBUG_RETURN(1);
}
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
}
BUG#21096: locking issue ; temporary table conflicts. The problem was that during DROP TEMPORARY TABLE we tried to acquire the name lock, though temporary tables belongs to one connection, and no race is possible. The solution is to not use table name locking while executing DROP TEMPORARY TABLE.
2006-08-29 14:59:20 +02:00
if (!drop_temporary && lock_table_names(thd, tables))
Merge to get fix for LOCK TABLES + DROP TABLE in another thread
2003-03-07 00:20:56 +01:00
DBUG_RETURN(1);
Fixed a deadlock problem when using LOCK TABLE in one thread and DROP TABLE in another
2003-03-03 19:42:49 +01:00
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
/* Don't give warnings for not found errors, as we already generate notes */
thd->no_warnings_for_error= 1;
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
for (table= tables; table; table= table->next_local)
Import changeset
2000-07-31 21:29:14 +02:00
{
Merge with 3.23 to get fix for deadlock when doing LOCK TABLES in one thread and DROP TABLE in another thread
2003-03-04 12:36:59 +01:00
char *db=table->db;
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
handlerton *table_type;
enum legacy_db_type frm_db_type;
Changes in get_table_type() and mysql_frm_type(). The main problem was that in mysql_rm_table_part2_with_lock() previously we needed to open same file twice. Now once is enough.
2005-11-03 15:10:11 +01:00
Bug#14397 - OPTIMIZE TABLE with an open HANDLER causes a crash Version for 5.0. It fixes three problems: 1. The cause of the bug was that we did not check the table version for the HANDLER ... READ commands. We did not notice when a table was replaced by a new one. This can happen during ALTER TABLE, REPAIR TABLE, and OPTIMIZE TABLE (there might be more cases). I call the fix for this problem "the primary bug fix". 2. mysql_ha_flush() was not always called with a locked LOCK_open. Though the function comment clearly said it must. I changed the code so that the locking is done when required. I call the fix for this problem "the secondary fix". 3. In 5.0 (not in 4.1 or 4.0) DROP TABLE had a possible deadlock flaw in concur with FLUSH TABLES WITH READ LOCK. I call the fix for this problem "the 5.0 addendum fix".
2005-11-15 21:57:02 +01:00
mysql_ha_flush(thd, table, MYSQL_HA_CLOSE_FINAL, TRUE);
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
if (!close_temporary_table(thd, table))
Fixed some reported bugs
2000-11-18 22:13:48 +01:00
{
Put temporary files in binlog cache when using BEGIN/COMMIT Let MySQL 4.0 read 4.1 .frm files without 4.1 specific extensions New variables @@rand_seed1 and @@rand_seed2 (used by replication) DROP TEMPORARY TABLE
2002-11-07 03:02:37 +01:00
tmp_table_deleted=1;
Import changeset
2000-07-31 21:29:14 +02:00
continue; // removed temporary table
Fixed some reported bugs
2000-11-18 22:13:48 +01:00
}
Import changeset
2000-07-31 21:29:14 +02:00
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
/*
If row-based replication is used and the table is not a
temporary table, we add the table name to the drop statement
being built. The string always end in a comma and the comma
will be chopped off before being written to the binary log.
*/
WL#2977 and WL#2712 global and session-level variable to set the binlog format (row/statement), and new binlog format called "mixed" (which is statement-based except if only row-based is correct, in this cset it means if UDF or UUID is used; more cases could be added in later 5.1 release): SET GLOBAL|SESSION BINLOG_FORMAT=row|statement|mixed|default; the global default is statement unless cluster is enabled (then it's row) as in 5.1-alpha. It's not possible to use SET on this variable if a session is currently in row-based mode and has open temporary tables (because CREATE TEMPORARY TABLE was not binlogged so temp table is not known on slave), or if NDB is enabled (because NDB does not support such change on-the-fly, though it will later), of if in a stored function (see below). The added tests test the possibility or impossibility to SET, their effects, and the mixed mode, including in prepared statements and in stored procedures and functions. Caveats: a) The mixed mode will not work for stored functions: in mixed mode, a stored function will always be binlogged as one call and in a statement-based way (e.g. INSERT VALUES(myfunc()) or SELECT myfunc()). b) for the same reason, changing the thread's binlog format inside a stored function is refused with an error message. c) the same problems apply to triggers; implementing b) for triggers will be done later (will ask Dmitri). Additionally, as the binlog format is now changeable by each user for his session, I remove the implication which was done at startup, where row-based automatically set log-bin-trust-routine-creators to 1 (not possible anymore as a user can now switch to stmt-based and do nasty things again), and automatically set --innodb-locks-unsafe-for-binlog to 1 (was anyway theoretically incorrect as it disabled phantom protection). Plus fixes for compiler warnings.
2006-02-25 22:21:03 +01:00
if (thd->current_stmt_binlog_row_based && !dont_log_query)
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
{
Fixed compiler and valgrind warnings Added missing DBUG_xxx_RETURN statements Fixed some usage of not initialized variables (as found by valgrind) Ensure that we don't remove locked tables used as name locks from open table cache until unlock_table_names() are called. This was fixed by having drop_locked_name() returning any table used as a name lock so that we can free it in unlock_table_names() This will allow Tomas to continue with his work to use namelocks to syncronize things. Note: valgrind still produces a lot of warnings about using not initialized code and shows memory loss errors when running the ndb tests
2006-03-29 13:27:36 +02:00
non_temp_tables_count++;
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
/*
Don't write the database name if it is the current one (or if
thd->db is NULL).
*/
built_query.append("`");
if (thd->db == NULL || strcmp(db,thd->db) != 0)
{
built_query.append(db);
built_query.append("`.`");
}
built_query.append(table->table_name);
built_query.append("`,");
}
Import changeset
2000-07-31 21:29:14 +02:00
error=0;
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
table_type= table->db_type;
Added support for DROP TEMPORARY TABLE Removed mysql_warnings() API function. Post merge fixes.
2003-01-04 14:37:20 +01:00
if (!drop_temporary)
Import changeset
2000-07-31 21:29:14 +02:00
{
Fixed compiler and valgrind warnings Added missing DBUG_xxx_RETURN statements Fixed some usage of not initialized variables (as found by valgrind) Ensure that we don't remove locked tables used as name locks from open table cache until unlock_table_names() are called. This was fixed by having drop_locked_name() returning any table used as a name lock so that we can free it in unlock_table_names() This will allow Tomas to continue with his work to use namelocks to syncronize things. Note: valgrind still produces a lot of warnings about using not initialized code and shows memory loss errors when running the ndb tests
2006-03-29 13:27:36 +02:00
TABLE *locked_table;
Merge mysql.com:/home/my/mysql-4.1 into mysql.com:/home/my/mysql-5.0
2005-07-28 16:09:54 +02:00
abort_locked_tables(thd, db, table->table_name);
remove_table_from_cache(thd, db, table->table_name,
RTFC_WAIT_OTHER_THREAD_FLAG |
RTFC_CHECK_KILLED_FLAG);
Fixed compiler and valgrind warnings Added missing DBUG_xxx_RETURN statements Fixed some usage of not initialized variables (as found by valgrind) Ensure that we don't remove locked tables used as name locks from open table cache until unlock_table_names() are called. This was fixed by having drop_locked_name() returning any table used as a name lock so that we can free it in unlock_table_names() This will allow Tomas to continue with his work to use namelocks to syncronize things. Note: valgrind still produces a lot of warnings about using not initialized code and shows memory loss errors when running the ndb tests
2006-03-29 13:27:36 +02:00
/*
If the table was used in lock tables, remember it so that
unlock_table_names can free it
*/
if ((locked_table= drop_locked_tables(thd, db, table->table_name)))
table->table= locked_table;
Added support for DROP TEMPORARY TABLE Removed mysql_warnings() API function. Post merge fixes.
2003-01-04 14:37:20 +01:00
if (thd->killed)
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
{
thd->no_warnings_for_error= 0;
Added support for DROP TEMPORARY TABLE Removed mysql_warnings() API function. Post merge fixes.
2003-01-04 14:37:20 +01:00
DBUG_RETURN(-1);
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
}
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
alias= (lower_case_table_names == 2) ? table->alias : table->table_name;
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
/* remove .frm file and engine files */
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
path_length= build_table_filename(path, sizeof(path),
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
db, alias, reg_ext, 0);
Import changeset
2000-07-31 21:29:14 +02:00
}
Fixed compiler and valgrind warnings Added missing DBUG_xxx_RETURN statements Fixed some usage of not initialized variables (as found by valgrind) Ensure that we don't remove locked tables used as name locks from open table cache until unlock_table_names() are called. This was fixed by having drop_locked_name() returning any table used as a name lock so that we can free it in unlock_table_names() This will allow Tomas to continue with his work to use namelocks to syncronize things. Note: valgrind still produces a lot of warnings about using not initialized code and shows memory loss errors when running the ndb tests
2006-03-29 13:27:36 +02:00
if (drop_temporary ||
(table_type == NULL &&
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
(access(path, F_OK) &&
ha_create_table_from_engine(thd, db, alias)) ||
(!drop_view &&
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
mysql_frm_type(thd, path, &frm_db_type) != FRMTYPE_TABLE)))
Import changeset
2000-07-31 21:29:14 +02:00
{
BUG#10365 Cluster handler uses non-standard error codes - Added better error messages when trying to open a table that can't be discovered or unpacked. The most likely cause of this is that it does not have any frm data, probably since it has been created from NdbApi or is a NDB system table. - Separated functionality that was in ha_create_table_from_engine into two functions. One that checks if the table exists and another one that tries to create the table from the engine.
2005-06-08 13:31:59 +02:00
// Table was not found on disk and table can't be created from engine
Return a warning for DROP DATABASE/TABLE IF EXISTS <non_existing_db/table(s)>
2002-12-04 12:19:08 +01:00
if (if_exists)
Changed mysql-test to print warnings for not existing table to DROP TABLE Cleaned up test; Removed wrong DROP TABLE commands and use standard table and database names. changed store_warning() -> push_warning_print()
2003-01-06 00:48:59 +01:00
push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_NOTE,
ER_BAD_TABLE_ERROR, ER(ER_BAD_TABLE_ERROR),
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
table->table_name);
Return a warning for DROP DATABASE/TABLE IF EXISTS <non_existing_db/table(s)>
2002-12-04 12:19:08 +01:00
else
BUG#10365 Cluster handler uses non-standard error codes - Added better error messages when trying to open a table that can't be discovered or unpacked. The most likely cause of this is that it does not have any frm data, probably since it has been created from NdbApi or is a NDB system table. - Separated functionality that was in ha_create_table_from_engine into two functions. One that checks if the table exists and another one that tries to create the table from the engine.
2005-06-08 13:31:59 +02:00
error= 1;
Import changeset
2000-07-31 21:29:14 +02:00
}
else
{
Lots of fixes for BDB tables Change DROP TABLE to first drop the data, then the .frm file
2000-12-08 16:04:57 +01:00
char *end;
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
if (table_type == NULL)
{
mysql_frm_type(thd, path, &frm_db_type);
table_type= ha_resolve_by_legacy_type(thd, frm_db_type);
}
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
// Remove extension for delete
*(end= path + path_length - reg_ext_length)= '\0';
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
error= ha_delete_table(thd, table_type, path, db, table->table_name,
Generate warning in ha_delete_table() if files is missing in handler
2005-02-21 19:41:48 +01:00
!dont_log_query);
Fix for dropping a table with frm file consisting of "junk"
2005-10-24 15:05:06 +02:00
if ((error == ENOENT || error == HA_ERR_NO_SUCH_TABLE) &&
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
(if_exists || table_type == NULL))
Review of new pushed code (XA & other) Portability fixes and cleanups Fixed setting of 'res' in mysql_execute_command()
2005-02-21 13:47:57 +01:00
error= 0;
dict0dict.c: Remove redundant code; parse both the database name and the table name in a FOREIGN KEY constraint with quotes in mind row0mysql.c, ha_innodb.cc, sql_table.cc: Return error message Cannot delete or update a parent row... if we try to drop a table which is referenced by a FOREIGN KEY constraint, and the user has not set foreign_key_checks=0
2004-02-01 23:10:45 +01:00
if (error == HA_ERR_ROW_IS_REFERENCED)
Merge with 4.0.18
2004-02-11 00:06:46 +01:00
{
/* the table is referenced by a foreign key constraint */
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
foreign_key_error=1;
Merge with 4.0.18
2004-02-11 00:06:46 +01:00
}
Review of new pushed code (XA & other) Portability fixes and cleanups Fixed setting of 'res' in mysql_execute_command()
2005-02-21 13:47:57 +01:00
if (!error || error == ENOENT || error == HA_ERR_NO_SUCH_TABLE)
Lots of fixes for BDB tables Change DROP TABLE to first drop the data, then the .frm file
2000-12-08 16:04:57 +01:00
{
Review of new pushed code (XA & other) Portability fixes and cleanups Fixed setting of 'res' in mysql_execute_command()
2005-02-21 13:47:57 +01:00
int new_error;
Lots of fixes for BDB tables Change DROP TABLE to first drop the data, then the .frm file
2000-12-08 16:04:57 +01:00
/* Delete the table definition file */
strmov(end,reg_ext);
Review of new pushed code (XA & other) Portability fixes and cleanups Fixed setting of 'res' in mysql_execute_command()
2005-02-21 13:47:57 +01:00
if (!(new_error=my_delete(path,MYF(MY_WME))))
Fix for bugs #5859 "DROP TABLE does not drop triggers" and #6559 "DROP DATABASE forgets to drop triggers". If we drop table we should also drop all triggers associated with it. To do this we have to check for existence of .TRG file when we are dropping table and delete it too.
2005-03-27 14:15:21 +02:00
{
Lots of fixes for BDB tables Change DROP TABLE to first drop the data, then the .frm file
2000-12-08 16:04:57 +01:00
some_tables_deleted=1;
Fix for bugs #5892/6182/8751/8758/10994 (based on Antony's patch) "Triggers have the wrong namespace" "Triggers: duplicate names allowed" "Triggers: CREATE TRIGGER does not accept fully qualified names" "SHOW TRIGGERS"
2005-07-19 18:06:49 +02:00
new_error= Table_triggers_list::drop_all_triggers(thd, db,
table->table_name);
Fix for bugs #5859 "DROP TABLE does not drop triggers" and #6559 "DROP DATABASE forgets to drop triggers". If we drop table we should also drop all triggers associated with it. To do this we have to check for existence of .TRG file when we are dropping table and delete it too.
2005-03-27 14:15:21 +02:00
}
Review of new pushed code (XA & other) Portability fixes and cleanups Fixed setting of 'res' in mysql_execute_command()
2005-02-21 13:47:57 +01:00
error|= new_error;
BUG#8547: Wrong errormessage when dropping table that does not exist in NDB - Added tescases - Added handling of HA_ERR_NO_SUCH_TABLE in mysql_rm_table_part2 so that error messages now are consistent
2005-02-17 14:22:44 +01:00
}
Import changeset
2000-07-31 21:29:14 +02:00
}
if (error)
{
if (wrong_tables.length())
wrong_tables.append(',');
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
wrong_tables.append(String(table->table_name,system_charset_info));
Import changeset
2000-07-31 21:29:14 +02:00
}
}
Move tmp_table_used to THD Optimize depending sub querys Remove valgrind warnings
2003-02-10 16:59:16 +01:00
thd->tmp_table_used= tmp_table_deleted;
Making DROP TABLE IF EXISTS, DROP DATABASE IF EXISTS, DELETE FROM, UPDATE be logged to binlog even if they changed nothing, and a test for this. This is useful when users use these commands to clean up their master and slave by issuing one command on master (assume master and slave have slightly different data for some reason and you want to clean up both). Note that I have not changed multi-table DELETE and multi-table UPDATE because their error-reporting mechanism is more complicated.
2004-06-09 16:07:01 +02:00
error= 0;
if (wrong_tables.length())
{
if (!foreign_key_error)
bug#3891 - DROP TABLE many-unexistent-tables, was printing an error with %s instead of table names sql/sql_table.cc: print an error with a function that respects width modifiers (%.64s)
2005-04-02 20:13:19 +02:00
my_printf_error(ER_BAD_TABLE_ERROR, ER(ER_BAD_TABLE_ERROR), MYF(0),
Merge of patch for bug #3891 into 5.0 tree.
2005-04-03 08:36:18 +02:00
wrong_tables.c_ptr());
Making DROP TABLE IF EXISTS, DROP DATABASE IF EXISTS, DELETE FROM, UPDATE be logged to binlog even if they changed nothing, and a test for this. This is useful when users use these commands to clean up their master and slave by issuing one command on master (assume master and slave have slightly different data for some reason and you want to clean up both). Note that I have not changed multi-table DELETE and multi-table UPDATE because their error-reporting mechanism is more complicated.
2004-06-09 16:07:01 +02:00
else
post-review fixes
2004-11-12 13:34:00 +01:00
my_message(ER_ROW_IS_REFERENCED, ER(ER_ROW_IS_REFERENCED), MYF(0));
Making DROP TABLE IF EXISTS, DROP DATABASE IF EXISTS, DELETE FROM, UPDATE be logged to binlog even if they changed nothing, and a test for this. This is useful when users use these commands to clean up their master and slave by issuing one command on master (assume master and slave have slightly different data for some reason and you want to clean up both). Note that I have not changed multi-table DELETE and multi-table UPDATE because their error-reporting mechanism is more complicated.
2004-06-09 16:07:01 +02:00
error= 1;
}
if (some_tables_deleted || tmp_table_deleted || !error)
Import changeset
2000-07-31 21:29:14 +02:00
{
added building without query cache
2002-03-22 21:55:08 +01:00
query_cache_invalidate3(thd, tables, 0);
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
if (!dont_log_query)
Fixed back some bin log writes to avoid merge problems Added missing #ifdef for partitioning
2005-08-20 22:56:54 +02:00
{
WL#2977 and WL#2712 global and session-level variable to set the binlog format (row/statement), and new binlog format called "mixed" (which is statement-based except if only row-based is correct, in this cset it means if UDF or UUID is used; more cases could be added in later 5.1 release): SET GLOBAL|SESSION BINLOG_FORMAT=row|statement|mixed|default; the global default is statement unless cluster is enabled (then it's row) as in 5.1-alpha. It's not possible to use SET on this variable if a session is currently in row-based mode and has open temporary tables (because CREATE TEMPORARY TABLE was not binlogged so temp table is not known on slave), or if NDB is enabled (because NDB does not support such change on-the-fly, though it will later), of if in a stored function (see below). The added tests test the possibility or impossibility to SET, their effects, and the mixed mode, including in prepared statements and in stored procedures and functions. Caveats: a) The mixed mode will not work for stored functions: in mixed mode, a stored function will always be binlogged as one call and in a statement-based way (e.g. INSERT VALUES(myfunc()) or SELECT myfunc()). b) for the same reason, changing the thread's binlog format inside a stored function is refused with an error message. c) the same problems apply to triggers; implementing b) for triggers will be done later (will ask Dmitri). Additionally, as the binlog format is now changeable by each user for his session, I remove the implication which was done at startup, where row-based automatically set log-bin-trust-routine-creators to 1 (not possible anymore as a user can now switch to stmt-based and do nasty things again), and automatically set --innodb-locks-unsafe-for-binlog to 1 (was anyway theoretically incorrect as it disabled phantom protection). Plus fixes for compiler warnings.
2006-02-25 22:21:03 +01:00
if (!thd->current_stmt_binlog_row_based ||
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
non_temp_tables_count > 0 && !tmp_table_deleted)
{
/*
In this case, we are either using statement-based
replication or using row-based replication but have only
deleted one or more non-temporary tables (and no temporary
tables). In this case, we can write the original query into
the binary log.
*/
write_bin_log(thd, !error, thd->query, thd->query_length);
}
WL#2977 and WL#2712 global and session-level variable to set the binlog format (row/statement), and new binlog format called "mixed" (which is statement-based except if only row-based is correct, in this cset it means if UDF or UUID is used; more cases could be added in later 5.1 release): SET GLOBAL|SESSION BINLOG_FORMAT=row|statement|mixed|default; the global default is statement unless cluster is enabled (then it's row) as in 5.1-alpha. It's not possible to use SET on this variable if a session is currently in row-based mode and has open temporary tables (because CREATE TEMPORARY TABLE was not binlogged so temp table is not known on slave), or if NDB is enabled (because NDB does not support such change on-the-fly, though it will later), of if in a stored function (see below). The added tests test the possibility or impossibility to SET, their effects, and the mixed mode, including in prepared statements and in stored procedures and functions. Caveats: a) The mixed mode will not work for stored functions: in mixed mode, a stored function will always be binlogged as one call and in a statement-based way (e.g. INSERT VALUES(myfunc()) or SELECT myfunc()). b) for the same reason, changing the thread's binlog format inside a stored function is refused with an error message. c) the same problems apply to triggers; implementing b) for triggers will be done later (will ask Dmitri). Additionally, as the binlog format is now changeable by each user for his session, I remove the implication which was done at startup, where row-based automatically set log-bin-trust-routine-creators to 1 (not possible anymore as a user can now switch to stmt-based and do nasty things again), and automatically set --innodb-locks-unsafe-for-binlog to 1 (was anyway theoretically incorrect as it disabled phantom protection). Plus fixes for compiler warnings.
2006-02-25 22:21:03 +01:00
else if (thd->current_stmt_binlog_row_based &&
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
non_temp_tables_count > 0 &&
tmp_table_deleted)
{
/*
In this case we have deleted both temporary and
non-temporary tables, so:
- since we have deleted a non-temporary table we have to
binlog the statement, but
- since we have deleted a temporary table we cannot binlog
the statement (since the table has not been created on the
slave, this might cause the slave to stop).
Instead, we write a built statement, only containing the
non-temporary tables, to the binary log
*/
built_query.chop(); // Chop of the last comma
built_query.append(" /* generated by server */");
write_bin_log(thd, !error, built_query.ptr(), built_query.length());
}
/*
The remaining cases are:
- no tables where deleted and
- only temporary tables where deleted and row-based
replication is used.
In both these cases, nothing should be written to the binary
log.
*/
Fixed back some bin log writes to avoid merge problems Added missing #ifdef for partitioning
2005-08-20 22:56:54 +02:00
}
Import changeset
2000-07-31 21:29:14 +02:00
}
Update copyright Fixed memory leak on shutdown (Affects the embedded version & MyODBC)
2001-12-06 13:10:51 +01:00
BUG#21096: locking issue ; temporary table conflicts. The problem was that during DROP TEMPORARY TABLE we tried to acquire the name lock, though temporary tables belongs to one connection, and no race is possible. The solution is to not use table name locking while executing DROP TEMPORARY TABLE.
2006-08-29 14:59:20 +02:00
if (!drop_temporary)
Merge mockturtle.local:/home/dlenev/src/mysql-4.1-bg22338-2 into mockturtle.local:/home/dlenev/src/mysql-5.0-rt-merge
2006-09-29 10:36:12 +02:00
unlock_table_names(thd, tables, (TABLE_LIST*) 0);
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
thd->no_warnings_for_error= 0;
Changed DB_TYPE_INNOBASE to DB_TYPE_INNODB Fix that DROP DATABASE works with all table types Use BULK_INSERT when inserting more than one row Better TRUNCATE TABLE
2001-09-02 12:47:00 +02:00
DBUG_RETURN(error);
Import changeset
2000-07-31 21:29:14 +02:00
}
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
/*
Quickly remove a table.
SYNOPSIS
quick_rm_table()
base The handlerton handle.
db The database name.
table_name The table name.
flags flags for build_table_filename().
RETURN
0 OK
!= 0 Error
*/
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
bool quick_rm_table(handlerton *base,const char *db,
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
const char *table_name, uint flags)
Import changeset
2000-07-31 21:29:14 +02:00
{
char path[FN_REFLEN];
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
bool error= 0;
DBUG_ENTER("quick_rm_table");
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
uint path_length= build_table_filename(path, sizeof(path),
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
db, table_name, reg_ext, flags);
Import changeset
2000-07-31 21:29:14 +02:00
if (my_delete(path,MYF(0)))
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
error= 1; /* purecov: inspected */
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
path[path_length - reg_ext_length]= '\0'; // Remove reg_ext
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
DBUG_RETURN(ha_delete_table(current_thd, base, path, db, table_name, 0) ||
error);
Import changeset
2000-07-31 21:29:14 +02:00
}
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
/*
Sort keys in the following order:
- PRIMARY KEY
- UNIQUE keyws where all column are NOT NULL
- Other UNIQUE keys
- Normal keys
- Fulltext keys
This will make checking for duplicated keys faster and ensure that
PRIMARY keys are prioritized.
*/
static int sort_keys(KEY *a, KEY *b)
{
if (a->flags & HA_NOSAME)
{
if (!(b->flags & HA_NOSAME))
return -1;
Fix autoincrement for signed columns (Bug #1366) Fixed problem with char > 128 in QUOTE() function. (Bug #1868) Disable creation of symlinks if my_disable_symlink is set Fixed searching of TEXT with end space. (Bug #1651) Fixed caching bug in multi-table-update where same table was used twice. (Bug #1711) Fixed problem with UNIX_TIMESTAMP() for timestamps close to 0. (Bug #1998) Fixed timestamp.test
2003-12-12 21:26:58 +01:00
if ((a->flags ^ b->flags) & (HA_NULL_PART_KEY | HA_END_SPACE_KEY))
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
{
/* Sort NOT NULL keys before other keys */
Fix autoincrement for signed columns (Bug #1366) Fixed problem with char > 128 in QUOTE() function. (Bug #1868) Disable creation of symlinks if my_disable_symlink is set Fixed searching of TEXT with end space. (Bug #1651) Fixed caching bug in multi-table-update where same table was used twice. (Bug #1711) Fixed problem with UNIX_TIMESTAMP() for timestamps close to 0. (Bug #1998) Fixed timestamp.test
2003-12-12 21:26:58 +01:00
return (a->flags & (HA_NULL_PART_KEY | HA_END_SPACE_KEY)) ? 1 : -1;
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
}
if (a->name == primary_key_name)
return -1;
if (b->name == primary_key_name)
return 1;
}
else if (b->flags & HA_NOSAME)
return 1; // Prefer b
if ((a->flags ^ b->flags) & HA_FULLTEXT)
{
return (a->flags & HA_FULLTEXT) ? 1 : -1;
}
Fix for FLUSH QUERY CACHE Fix for new bug in CREATE TABLE when sorting keys.
2002-01-05 21:51:42 +01:00
/*
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
Prefer original key order. usable_key_parts contains here
Fix for FLUSH QUERY CACHE Fix for new bug in CREATE TABLE when sorting keys.
2002-01-05 21:51:42 +01:00
the original key position.
*/
return ((a->usable_key_parts < b->usable_key_parts) ? -1 :
(a->usable_key_parts > b->usable_key_parts) ? 1 :
0);
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
}
fixed bug #1427 "enum allows duplicate values in the list"
2004-03-13 20:13:31 +01:00
/*
Check TYPELIB (set or enum) for duplicates
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
fixed bug #1427 "enum allows duplicate values in the list"
2004-03-13 20:13:31 +01:00
SYNOPSIS
check_duplicates_in_interval()
set_or_name "SET" or "ENUM" string for warning message
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
name name of the checked column
typelib list of values for the column
fixed bug #1427 "enum allows duplicate values in the list"
2004-03-13 20:13:31 +01:00
DESCRIPTION
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
This function prints an warning for each value in list
fixed bug #1427 "enum allows duplicate values in the list"
2004-03-13 20:13:31 +01:00
which has some duplicates on its right
RETURN VALUES
void
*/
void check_duplicates_in_interval(const char *set_or_name,
- Check ENUM/SET duplicate values taking in account charset/collation.
2004-10-26 07:41:14 +02:00
const char *name, TYPELIB *typelib,
CHARSET_INFO *cs)
fixed bug #1427 "enum allows duplicate values in the list"
2004-03-13 20:13:31 +01:00
{
- Check ENUM/SET duplicate values taking in account charset/collation.
2004-10-26 07:41:14 +02:00
TYPELIB tmp= *typelib;
fixed bug #1427 "enum allows duplicate values in the list"
2004-03-13 20:13:31 +01:00
const char **cur_value= typelib->type_names;
- Check ENUM/SET duplicate values taking in account charset/collation.
2004-10-26 07:41:14 +02:00
unsigned int *cur_length= typelib->type_lengths;
for ( ; tmp.count > 1; cur_value++, cur_length++)
fixed bug #1427 "enum allows duplicate values in the list"
2004-03-13 20:13:31 +01:00
{
- Check ENUM/SET duplicate values taking in account charset/collation.
2004-10-26 07:41:14 +02:00
tmp.type_names++;
tmp.type_lengths++;
tmp.count--;
if (find_type2(&tmp, (const char*)*cur_value, *cur_length, cs))
fixed bug #1427 "enum allows duplicate values in the list"
2004-03-13 20:13:31 +01:00
{
Optimizations
2004-03-15 11:53:27 +01:00
push_warning_printf(current_thd,MYSQL_ERROR::WARN_LEVEL_NOTE,
fixed bug #1427 "enum allows duplicate values in the list"
2004-03-13 20:13:31 +01:00
ER_DUPLICATED_VALUE_IN_TYPE,
ER(ER_DUPLICATED_VALUE_IN_TYPE),
name,*cur_value,set_or_name);
}
}
}
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
/*
Check TYPELIB (set or enum) max and total lengths
SYNOPSIS
calculate_interval_lengths()
cs charset+collation pair of the interval
typelib list of values for the column
max_length length of the longest item
tot_length sum of the item lengths
DESCRIPTION
After this function call:
- ENUM uses max_length
- SET uses tot_length.
RETURN VALUES
void
*/
void calculate_interval_lengths(CHARSET_INFO *cs, TYPELIB *interval,
uint32 *max_length, uint32 *tot_length)
{
const char **pos;
uint *len;
*max_length= *tot_length= 0;
for (pos= interval->type_names, len= interval->type_lengths;
*pos ; pos++, len++)
{
uint length= cs->cset->numchars(cs, *pos, *pos + *len);
*tot_length+= length;
set_if_bigger(*max_length, (uint32)length);
}
}
Bug#3788 Crashes with stored procedure return non-string values Also fixes Bug#2773
2005-03-04 22:14:35 +01:00
/*
Prepare a create_table instance for packing
SYNOPSIS
prepare_create_field()
sql_field field to prepare for packing
blob_columns count for BLOBs
timestamps count for timestamps
table_flags table flags
DESCRIPTION
This function prepares a create_field instance.
Fields such as pack_flag are valid after this call.
RETURN VALUES
0 ok
1 Error
*/
int prepare_create_field(create_field *sql_field,
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
uint *blob_columns,
int *timestamps, int *timestamps_with_niladic,
Fixed compiler warnings (Mostly VC++): - Removed not used variables - Changed some ulong parameters/variables to ulonglong (possible serious bug) - Added casts to get rid of safe assignment from longlong to long (and similar) - Added casts to function parameters - Fixed signed/unsigned compares - Added some constructores to structures - Removed some not portable constructs Better fix for bug Bug #21428 "skipped 9 bytes from file: socket (3)" on "mysqladmin shutdown" (Added new parameter to net_clear() to define when we want the communication buffer to be emptied)
2006-11-30 02:40:42 +01:00
longlong table_flags)
Bug#3788 Crashes with stored procedure return non-string values Also fixes Bug#2773
2005-03-04 22:14:35 +01:00
{
DBUG_ENTER("prepare_field");
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
/*
This code came from mysql_prepare_table.
Indent preserved to make patching easier
*/
DBUG_ASSERT(sql_field->charset);
switch (sql_field->sql_type) {
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
case MYSQL_TYPE_BLOB:
case MYSQL_TYPE_MEDIUM_BLOB:
case MYSQL_TYPE_TINY_BLOB:
case MYSQL_TYPE_LONG_BLOB:
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
sql_field->pack_flag=FIELDFLAG_BLOB |
pack_length_to_packflag(sql_field->pack_length -
portable_sizeof_char_ptr);
if (sql_field->charset->state & MY_CS_BINSORT)
sql_field->pack_flag|=FIELDFLAG_BINARY;
sql_field->length=8; // Unireg field length
sql_field->unireg_check=Field::BLOB_FIELD;
Move handler dependent tests to the specific handler (myisam, bdb, innodb) Enabled VARCHAR testing for innodb NOTE: innodb.test currently fails becasue of a bug in InnoDB. I have informed Heikki about this and expect him to fix this ASAP
2005-03-17 00:22:12 +01:00
(*blob_columns)++;
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
break;
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
case MYSQL_TYPE_GEOMETRY:
Bug#3788 Crashes with stored procedure return non-string values Also fixes Bug#2773
2005-03-04 22:14:35 +01:00
#ifdef HAVE_SPATIAL
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
if (!(table_flags & HA_CAN_GEOMETRY))
{
my_printf_error(ER_CHECK_NOT_IMPLEMENTED, ER(ER_CHECK_NOT_IMPLEMENTED),
MYF(0), "GEOMETRY");
Bug#3788 Crashes with stored procedure return non-string values Also fixes Bug#2773
2005-03-04 22:14:35 +01:00
DBUG_RETURN(1);
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
}
sql_field->pack_flag=FIELDFLAG_GEOM |
pack_length_to_packflag(sql_field->pack_length -
portable_sizeof_char_ptr);
if (sql_field->charset->state & MY_CS_BINSORT)
sql_field->pack_flag|=FIELDFLAG_BINARY;
sql_field->length=8; // Unireg field length
sql_field->unireg_check=Field::BLOB_FIELD;
Move handler dependent tests to the specific handler (myisam, bdb, innodb) Enabled VARCHAR testing for innodb NOTE: innodb.test currently fails becasue of a bug in InnoDB. I have informed Heikki about this and expect him to fix this ASAP
2005-03-17 00:22:12 +01:00
(*blob_columns)++;
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
break;
#else
my_printf_error(ER_FEATURE_DISABLED,ER(ER_FEATURE_DISABLED), MYF(0),
sym_group_geom.name, sym_group_geom.needed_define);
DBUG_RETURN(1);
Bug#3788 Crashes with stored procedure return non-string values Also fixes Bug#2773
2005-03-04 22:14:35 +01:00
#endif /*HAVE_SPATIAL*/
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
case MYSQL_TYPE_VARCHAR:
Bug#3788 Crashes with stored procedure return non-string values Also fixes Bug#2773
2005-03-04 22:14:35 +01:00
#ifndef QQ_ALL_HANDLERS_SUPPORT_VARCHAR
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
if (table_flags & HA_NO_VARCHAR)
{
/* convert VARCHAR to CHAR because handler is not yet up to date */
sql_field->sql_type= MYSQL_TYPE_VAR_STRING;
sql_field->pack_length= calc_pack_length(sql_field->sql_type,
(uint) sql_field->length);
if ((sql_field->length / sql_field->charset->mbmaxlen) >
MAX_FIELD_CHARLENGTH)
Bug#3788 Crashes with stored procedure return non-string values Also fixes Bug#2773
2005-03-04 22:14:35 +01:00
{
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
my_printf_error(ER_TOO_BIG_FIELDLENGTH, ER(ER_TOO_BIG_FIELDLENGTH),
MYF(0), sql_field->field_name, MAX_FIELD_CHARLENGTH);
Bug#3788 Crashes with stored procedure return non-string values Also fixes Bug#2773
2005-03-04 22:14:35 +01:00
DBUG_RETURN(1);
}
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
}
#endif
/* fall through */
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
case MYSQL_TYPE_STRING:
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
sql_field->pack_flag=0;
if (sql_field->charset->state & MY_CS_BINSORT)
sql_field->pack_flag|=FIELDFLAG_BINARY;
break;
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
case MYSQL_TYPE_ENUM:
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
sql_field->pack_flag=pack_length_to_packflag(sql_field->pack_length) |
FIELDFLAG_INTERVAL;
if (sql_field->charset->state & MY_CS_BINSORT)
sql_field->pack_flag|=FIELDFLAG_BINARY;
sql_field->unireg_check=Field::INTERVAL_FIELD;
check_duplicates_in_interval("ENUM",sql_field->field_name,
sql_field->interval,
sql_field->charset);
break;
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
case MYSQL_TYPE_SET:
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
sql_field->pack_flag=pack_length_to_packflag(sql_field->pack_length) |
FIELDFLAG_BITFIELD;
if (sql_field->charset->state & MY_CS_BINSORT)
sql_field->pack_flag|=FIELDFLAG_BINARY;
sql_field->unireg_check=Field::BIT_FIELD;
check_duplicates_in_interval("SET",sql_field->field_name,
sql_field->interval,
sql_field->charset);
break;
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
case MYSQL_TYPE_DATE: // Rest of string types
case MYSQL_TYPE_NEWDATE:
case MYSQL_TYPE_TIME:
case MYSQL_TYPE_DATETIME:
case MYSQL_TYPE_NULL:
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
sql_field->pack_flag=f_settype((uint) sql_field->sql_type);
break;
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
case MYSQL_TYPE_BIT:
Bit type: code clean-up.
2005-04-12 20:12:00 +02:00
/*
We have sql_field->pack_flag already set here, see mysql_prepare_table().
*/
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
break;
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
case MYSQL_TYPE_NEWDECIMAL:
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
sql_field->pack_flag=(FIELDFLAG_NUMBER |
(sql_field->flags & UNSIGNED_FLAG ? 0 :
FIELDFLAG_DECIMAL) |
(sql_field->flags & ZEROFILL_FLAG ?
FIELDFLAG_ZEROFILL : 0) |
(sql_field->decimals << FIELDFLAG_DEC_SHIFT));
break;
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
case MYSQL_TYPE_TIMESTAMP:
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
/* We should replace old TIMESTAMP fields with their newer analogs */
if (sql_field->unireg_check == Field::TIMESTAMP_OLD_FIELD)
{
Move handler dependent tests to the specific handler (myisam, bdb, innodb) Enabled VARCHAR testing for innodb NOTE: innodb.test currently fails becasue of a bug in InnoDB. I have informed Heikki about this and expect him to fix this ASAP
2005-03-17 00:22:12 +01:00
if (!*timestamps)
Bug#3788 Crashes with stored procedure return non-string values Also fixes Bug#2773
2005-03-04 22:14:35 +01:00
{
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
sql_field->unireg_check= Field::TIMESTAMP_DNUN_FIELD;
Move handler dependent tests to the specific handler (myisam, bdb, innodb) Enabled VARCHAR testing for innodb NOTE: innodb.test currently fails becasue of a bug in InnoDB. I have informed Heikki about this and expect him to fix this ASAP
2005-03-17 00:22:12 +01:00
(*timestamps_with_niladic)++;
Bug#3788 Crashes with stored procedure return non-string values Also fixes Bug#2773
2005-03-04 22:14:35 +01:00
}
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
else
sql_field->unireg_check= Field::NONE;
}
else if (sql_field->unireg_check != Field::NONE)
Move handler dependent tests to the specific handler (myisam, bdb, innodb) Enabled VARCHAR testing for innodb NOTE: innodb.test currently fails becasue of a bug in InnoDB. I have informed Heikki about this and expect him to fix this ASAP
2005-03-17 00:22:12 +01:00
(*timestamps_with_niladic)++;
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
Move handler dependent tests to the specific handler (myisam, bdb, innodb) Enabled VARCHAR testing for innodb NOTE: innodb.test currently fails becasue of a bug in InnoDB. I have informed Heikki about this and expect him to fix this ASAP
2005-03-17 00:22:12 +01:00
(*timestamps)++;
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
/* fall-through */
default:
sql_field->pack_flag=(FIELDFLAG_NUMBER |
(sql_field->flags & UNSIGNED_FLAG ? 0 :
FIELDFLAG_DECIMAL) |
(sql_field->flags & ZEROFILL_FLAG ?
FIELDFLAG_ZEROFILL : 0) |
f_settype((uint) sql_field->sql_type) |
(sql_field->decimals << FIELDFLAG_DEC_SHIFT));
break;
}
if (!(sql_field->flags & NOT_NULL_FLAG))
sql_field->pack_flag|= FIELDFLAG_MAYBE_NULL;
if (sql_field->flags & NO_DEFAULT_VALUE_FLAG)
sql_field->pack_flag|= FIELDFLAG_NO_DEFAULT;
Bug#3788 Crashes with stored procedure return non-string values Also fixes Bug#2773
2005-03-04 22:14:35 +01:00
DBUG_RETURN(0);
}
Some trivial optimzations Check if AND/OR expression can be NULL; Fixed bug in GROUP BY and-or-expression where expression could be NULL Bug fix for SHOW OPEN TABLES when user didn't have privilege to access all open tables. Better fix for ALTER TABLE on BDB tables.
2002-09-03 14:44:25 +02:00
/*
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
Preparation for table creation
Some trivial optimzations Check if AND/OR expression can be NULL; Fixed bug in GROUP BY and-or-expression where expression could be NULL Bug fix for SHOW OPEN TABLES when user didn't have privilege to access all open tables. Better fix for ALTER TABLE on BDB tables.
2002-09-03 14:44:25 +02:00
SYNOPSIS
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
mysql_prepare_table()
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
thd Thread object.
create_info Create information (like MAX_ROWS).
fields List of fields to create.
keys List of keys to create.
tmp_table If a temporary table is to be created.
db_options INOUT Table options (like HA_OPTION_PACK_RECORD).
file The handler for the new table.
key_info_buffer OUT An array of KEY structs for the indexes.
key_count OUT The number of elements in the array.
select_field_count The number of fields coming from a select table.
Some trivial optimzations Check if AND/OR expression can be NULL; Fixed bug in GROUP BY and-or-expression where expression could be NULL Bug fix for SHOW OPEN TABLES when user didn't have privilege to access all open tables. Better fix for ALTER TABLE on BDB tables.
2002-09-03 14:44:25 +02:00
CHECKSUM TABLE table1, table2, ... [ QUICK | EXTENDED ]
2003-09-03 11:34:32 +02:00
DESCRIPTION
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
Prepares the table and key structures for table creation.
Some trivial optimzations Check if AND/OR expression can be NULL; Fixed bug in GROUP BY and-or-expression where expression could be NULL Bug fix for SHOW OPEN TABLES when user didn't have privilege to access all open tables. Better fix for ALTER TABLE on BDB tables.
2002-09-03 14:44:25 +02:00
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
NOTES
Hack to ensure that CHAR's created in 5.0 are not converted to VARCHAR even if the row type is dynamic (For 4.1 tables old 'VARCHAR' fields are converted to true VARCHAR in the next ALTER TABLE) This ensures that one can use MySQL 5.0 privilege tables with MySQL 4.1
2005-03-22 14:48:06 +01:00
sets create_info->varchar if the table has a varchar
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
Some trivial optimzations Check if AND/OR expression can be NULL; Fixed bug in GROUP BY and-or-expression where expression could be NULL Bug fix for SHOW OPEN TABLES when user didn't have privilege to access all open tables. Better fix for ALTER TABLE on BDB tables.
2002-09-03 14:44:25 +02:00
RETURN VALUES
0 ok
-1 error
*/
Import changeset
2000-07-31 21:29:14 +02:00
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
static int mysql_prepare_table(THD *thd, HA_CREATE_INFO *create_info,
List<create_field> *fields,
List<Key> *keys, bool tmp_table,
uint *db_options,
handler *file, KEY **key_info_buffer,
uint *key_count, int select_field_count)
Import changeset
2000-07-31 21:29:14 +02:00
{
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
const char *key_name;
Import changeset
2000-07-31 21:29:14 +02:00
create_field *sql_field,*dup_field;
After merge fixes
2005-05-16 14:21:35 +02:00
uint field,null_fields,blob_columns,max_key_length;
Merge with 4.0
2005-05-13 23:01:40 +02:00
ulong record_offset= 0;
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
KEY *key_info;
Import changeset
2000-07-31 21:29:14 +02:00
KEY_PART_INFO *key_part_info;
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
int timestamps= 0, timestamps_with_niladic= 0;
int field_no,dup_no;
int select_field_pos,auto_increment=0;
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
List_iterator<create_field> it(*fields),it2(*fields);
bit type
2004-12-17 15:06:05 +01:00
uint total_uneven_bit_length= 0;
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
DBUG_ENTER("mysql_prepare_table");
Import changeset
2000-07-31 21:29:14 +02:00
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
select_field_pos= fields->elements - select_field_count;
Import changeset
2000-07-31 21:29:14 +02:00
null_fields=blob_columns=0;
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
create_info->varchar= 0;
After merge fixes
2005-05-16 14:21:35 +02:00
max_key_length= file->max_key_length();
sql_table.cc re-apply a lost change where we check to see if the sql_table.cc table handler supports temporary tables
2001-06-06 21:45:07 +02:00
Fixes and code cleanups after merge with 4.0.3 Warning handling and initial prepared statement handling (last not complete yet) Changed a lot of functions that returned 0/1 to my_bool type. GRANT handling now uses read/write locks instead of mutex Change basic net functions to use THD instead of NET (needed for 4.1 protocol) Use my_sprintf instead of sprintf() + strlen() Added alloc_query() to be able to chare query initialization code with prepared statements. Cleanup handling of SHOW COUNT(*) WARNINGS and SELECT LAST_INSERT_ID() Note that the following test fails (will be fixed ASAP): sub_select, union, rpl_rotate_logs and rpl_mystery22
2002-10-02 12:33:08 +02:00
for (field_no=0; (sql_field=it++) ; field_no++)
Import changeset
2000-07-31 21:29:14 +02:00
{
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
CHARSET_INFO *save_cs;
A fix and a test case for Bug#13134 "Length of VARCHAR() utf8 column is increasing when table is recreated with PS/SP": make use of create_field::char_length more consistent in the code. Reinit create_field::length from create_field::char_length for every execution of a prepared statement (actually fixes the bug).
2006-02-21 17:52:20 +01:00
/*
Initialize length from its original value (number of characters),
which was set in the parser. This is necessary if we're
executing a prepared statement for the second time.
*/
sql_field->length= sql_field->char_length;
mysql_create_table now accepts field lengths in CREATE form, not in internal form. This is to simplify fixing indexes to mean number of chars rather than bytes too.
2003-09-10 14:25:26 +02:00
if (!sql_field->charset)
CLIENT_MULTI_QUERIES -> CLIENT_MULTI_STATEMENTS New multi-key-cache handling. This was needed becasue the old one didn't work reliable with MERGE tables. ALTER TABLE table_name ... CHARACTER SET ... now changes all char/varchar/text columns to the given character set (One must use ALTER TABLE ... DEFAULT CHARACTER SET ... to change the default character set) Fixed that have_compress is detected properly (fixes problems with func_compress.test on platforms without zlib) New syntax for CACHE INDEX ('keys' is optional if no index name is given and one mentions the key cache name only ones) Removed compiler warnings Added mysql_set_server_option() to allow clients like PHP to easaily set/reset the multi-statement flag.
2003-11-18 12:47:27 +01:00
sql_field->charset= create_info->default_table_charset;
/*
table_charset is set in ALTER TABLE if we want change character set
Bug #2821 Table default character set affects LONGBLOB fields
2004-03-19 08:37:49 +01:00
for all varchar/char columns.
But the table charset must not affect the BLOB fields, so don't
allow to change my_charset_bin to somethig else.
CLIENT_MULTI_QUERIES -> CLIENT_MULTI_STATEMENTS New multi-key-cache handling. This was needed becasue the old one didn't work reliable with MERGE tables. ALTER TABLE table_name ... CHARACTER SET ... now changes all char/varchar/text columns to the given character set (One must use ALTER TABLE ... DEFAULT CHARACTER SET ... to change the default character set) Fixed that have_compress is detected properly (fixes problems with func_compress.test on platforms without zlib) New syntax for CACHE INDEX ('keys' is optional if no index name is given and one mentions the key cache name only ones) Removed compiler warnings Added mysql_set_server_option() to allow clients like PHP to easaily set/reset the multi-statement flag.
2003-11-18 12:47:27 +01:00
*/
Bug #2821 Table default character set affects LONGBLOB fields
2004-03-19 08:37:49 +01:00
if (create_info->table_charset && sql_field->charset != &my_charset_bin)
mysql_create_table now accepts field lengths in CREATE form, not in internal form. This is to simplify fixing indexes to mean number of chars rather than bytes too.
2003-09-10 14:25:26 +02:00
sql_field->charset= create_info->table_charset;
1. New data types, from the user point of view: BINARY(N) and VARBIBARY(N) 2. More 4.0 compatibility and more BINARY keyword consistency: 2a. CREATE TABLE a (a CHAR(N) BINARY) is now synonym for CREATE TABLE a (a CHAR(N) COLLATE xxxx_bin) 2b. SELECT BINARY x is still synonin for SELECT x COLLATE xxxxx_bin.
2004-03-26 13:11:46 +01:00
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
save_cs= sql_field->charset;
1. New data types, from the user point of view: BINARY(N) and VARBIBARY(N) 2. More 4.0 compatibility and more BINARY keyword consistency: 2a. CREATE TABLE a (a CHAR(N) BINARY) is now synonym for CREATE TABLE a (a CHAR(N) COLLATE xxxx_bin) 2b. SELECT BINARY x is still synonin for SELECT x COLLATE xxxxx_bin.
2004-03-26 13:11:46 +01:00
if ((sql_field->flags & BINCMP_FLAG) &&
!(sql_field->charset= get_charset_by_csname(sql_field->charset->csname,
MY_CS_BINSORT,MYF(0))))
{
char tmp[64];
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
strmake(strmake(tmp, save_cs->csname, sizeof(tmp)-4),
STRING_WITH_LEN("_bin"));
1. New data types, from the user point of view: BINARY(N) and VARBIBARY(N) 2. More 4.0 compatibility and more BINARY keyword consistency: 2a. CREATE TABLE a (a CHAR(N) BINARY) is now synonym for CREATE TABLE a (a CHAR(N) COLLATE xxxx_bin) 2b. SELECT BINARY x is still synonin for SELECT x COLLATE xxxxx_bin.
2004-03-26 13:11:46 +01:00
my_error(ER_UNKNOWN_COLLATION, MYF(0), tmp);
DBUG_RETURN(-1);
}
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
Fix for bug #20695: Charset introducer overrides charset definition for column. - if there are two character set definitions in the column declaration, we replace the first one with the second one as we store both in the LEX->charset slot. Add a separate slot to the LEX structure to store underscore charset. - convert default values to the column charset of STRING, VARSTRING fields if necessary as well.
2006-08-15 12:24:07 +02:00
/*
Merge maint2.mysql.com:/data/localhome/tsmith/bk/41 into maint2.mysql.com:/data/localhome/tsmith/bk/50
2006-09-01 08:53:56 +02:00
Convert the default value from client character
Fix for bug #20695: Charset introducer overrides charset definition for column. - if there are two character set definitions in the column declaration, we replace the first one with the second one as we store both in the LEX->charset slot. Add a separate slot to the LEX structure to store underscore charset. - convert default values to the column charset of STRING, VARSTRING fields if necessary as well.
2006-08-15 12:24:07 +02:00
set into the column character set if necessary.
*/
if (sql_field->def &&
Merge maint2.mysql.com:/data/localhome/tsmith/bk/41 into maint2.mysql.com:/data/localhome/tsmith/bk/50
2006-09-01 08:53:56 +02:00
save_cs != sql_field->def->collation.collation &&
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
(sql_field->sql_type == MYSQL_TYPE_VAR_STRING ||
sql_field->sql_type == MYSQL_TYPE_STRING ||
sql_field->sql_type == MYSQL_TYPE_SET ||
sql_field->sql_type == MYSQL_TYPE_ENUM))
Fix for bug #20695: Charset introducer overrides charset definition for column. - if there are two character set definitions in the column declaration, we replace the first one with the second one as we store both in the LEX->charset slot. Add a separate slot to the LEX structure to store underscore charset. - convert default values to the column charset of STRING, VARSTRING fields if necessary as well.
2006-08-15 12:24:07 +02:00
{
Merge maint2.mysql.com:/data/localhome/tsmith/bk/41 into maint2.mysql.com:/data/localhome/tsmith/bk/50
2006-09-01 08:53:56 +02:00
Query_arena backup_arena;
bool need_to_change_arena= !thd->stmt_arena->is_conventional();
Fix for bug #20695: Charset introducer overrides charset definition for column. - if there are two character set definitions in the column declaration, we replace the first one with the second one as we store both in the LEX->charset slot. Add a separate slot to the LEX structure to store underscore charset. - convert default values to the column charset of STRING, VARSTRING fields if necessary as well.
2006-08-15 12:24:07 +02:00
if (need_to_change_arena)
{
Merge maint2.mysql.com:/data/localhome/tsmith/bk/41 into maint2.mysql.com:/data/localhome/tsmith/bk/50
2006-09-01 08:53:56 +02:00
/* Asser that we don't do that at every PS execute */
DBUG_ASSERT(thd->stmt_arena->is_first_stmt_execute() ||
thd->stmt_arena->is_first_sp_execute());
thd->set_n_backup_active_arena(thd->stmt_arena, &backup_arena);
Fix for bug #20695: Charset introducer overrides charset definition for column. - if there are two character set definitions in the column declaration, we replace the first one with the second one as we store both in the LEX->charset slot. Add a separate slot to the LEX structure to store underscore charset. - convert default values to the column charset of STRING, VARSTRING fields if necessary as well.
2006-08-15 12:24:07 +02:00
}
Merge maint2.mysql.com:/data/localhome/tsmith/bk/41 into maint2.mysql.com:/data/localhome/tsmith/bk/50
2006-09-01 08:53:56 +02:00
sql_field->def= sql_field->def->safe_charset_converter(save_cs);
Fix for bug #20695: Charset introducer overrides charset definition for column. - if there are two character set definitions in the column declaration, we replace the first one with the second one as we store both in the LEX->charset slot. Add a separate slot to the LEX structure to store underscore charset. - convert default values to the column charset of STRING, VARSTRING fields if necessary as well.
2006-08-15 12:24:07 +02:00
if (need_to_change_arena)
Merge maint2.mysql.com:/data/localhome/tsmith/bk/41 into maint2.mysql.com:/data/localhome/tsmith/bk/50
2006-09-01 08:53:56 +02:00
thd->restore_active_arena(thd->stmt_arena, &backup_arena);
Fix for bug #20695: Charset introducer overrides charset definition for column. - if there are two character set definitions in the column declaration, we replace the first one with the second one as we store both in the LEX->charset slot. Add a separate slot to the LEX structure to store underscore charset. - convert default values to the column charset of STRING, VARSTRING fields if necessary as well.
2006-08-15 12:24:07 +02:00
if (sql_field->def == NULL)
{
/* Could not convert */
my_error(ER_INVALID_DEFAULT, MYF(0), sql_field->field_name);
DBUG_RETURN(-1);
}
}
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
if (sql_field->sql_type == MYSQL_TYPE_SET ||
sql_field->sql_type == MYSQL_TYPE_ENUM)
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
{
uint32 dummy;
CHARSET_INFO *cs= sql_field->charset;
Fix for bug #6840: Default value is not checked in ALTER column SET DEFAULT 'x'
2004-12-07 12:08:56 +01:00
TYPELIB *interval= sql_field->interval;
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
/*
Create typelib from interval_list, and if necessary
convert strings from client character set to the
column character set.
*/
Fix for bug #6840: Default value is not checked in ALTER column SET DEFAULT 'x'
2004-12-07 12:08:56 +01:00
if (!interval)
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
{
A fix and a test case for Bug#14410 "Crash in Enum or Set type in CREATE TABLE and PS/SP": make sure that 'typelib' object for ENUM values and 'Item_string' object for DEFAULT clause are created in the statement memory root.
2005-11-25 11:25:31 +01:00
/*
Create the typelib in prepared statement memory if we're
executing one.
*/
Post-merge fixes.
2005-11-25 16:53:55 +01:00
MEM_ROOT *stmt_root= thd->stmt_arena->mem_root;
A fix and a test case for Bug#14410 "Crash in Enum or Set type in CREATE TABLE and PS/SP": make sure that 'typelib' object for ENUM values and 'Item_string' object for DEFAULT clause are created in the statement memory root.
2005-11-25 11:25:31 +01:00
interval= sql_field->interval= typelib(stmt_root,
sql_field->interval_list);
Fixed compiler warnings detected by option -Wshadow and -Wunused: - Removed not used variables and functions - Added #ifdef around code that is not used - Renamed variables and functions to avoid conflicts - Removed some not used arguments Fixed some class/struct warnings in ndb Added define IS_LONGDATA() to simplify code in libmysql.c I did run gcov on the changes and added 'purecov' comments on almost all lines that was not just variable name changes
2006-12-14 23:51:37 +01:00
List_iterator<String> int_it(sql_field->interval_list);
Fix for bug #6840: Default value is not checked in ALTER column SET DEFAULT 'x'
2004-12-07 12:08:56 +01:00
String conv, *tmp;
Fix for bug#15316 SET value having comma not correctly handled disallow the use of comma in SET members
2006-03-29 16:52:26 +02:00
char comma_buf[2];
int comma_length= cs->cset->wc_mb(cs, ',', (uchar*) comma_buf,
(uchar*) comma_buf +
sizeof(comma_buf));
DBUG_ASSERT(comma_length > 0);
Fixed compiler warnings detected by option -Wshadow and -Wunused: - Removed not used variables and functions - Added #ifdef around code that is not used - Renamed variables and functions to avoid conflicts - Removed some not used arguments Fixed some class/struct warnings in ndb Added define IS_LONGDATA() to simplify code in libmysql.c I did run gcov on the changes and added 'purecov' comments on almost all lines that was not just variable name changes
2006-12-14 23:51:37 +01:00
for (uint i= 0; (tmp= int_it++); i++)
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
{
Fixes while reviewing new code Added option --count to mysqlshow (to show number of rows) Fixed possible core dump in information schema
2005-05-06 10:39:30 +02:00
uint lengthsp;
Fix for bug #6840: Default value is not checked in ALTER column SET DEFAULT 'x'
2004-12-07 12:08:56 +01:00
if (String::needs_conversion(tmp->length(), tmp->charset(),
cs, &dummy))
{
uint cnv_errs;
conv.copy(tmp->ptr(), tmp->length(), tmp->charset(), cs, &cnv_errs);
Merge mysql.com:/opt/local/work/mysql-4.1-root into mysql.com:/opt/local/work/mysql-5.0-root
2005-11-25 11:57:13 +01:00
interval->type_names[i]= strmake_root(stmt_root, conv.ptr(),
Fixes while reviewing new code Added option --count to mysqlshow (to show number of rows) Fixed possible core dump in information schema
2005-05-06 10:39:30 +02:00
conv.length());
Fix for bug #6840: Default value is not checked in ALTER column SET DEFAULT 'x'
2004-12-07 12:08:56 +01:00
interval->type_lengths[i]= conv.length();
}
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
Fix for bug #6840: Default value is not checked in ALTER column SET DEFAULT 'x'
2004-12-07 12:08:56 +01:00
// Strip trailing spaces.
Fixes while reviewing new code Added option --count to mysqlshow (to show number of rows) Fixed possible core dump in information schema
2005-05-06 10:39:30 +02:00
lengthsp= cs->cset->lengthsp(cs, interval->type_names[i],
interval->type_lengths[i]);
Fix for bug #6840: Default value is not checked in ALTER column SET DEFAULT 'x'
2004-12-07 12:08:56 +01:00
interval->type_lengths[i]= lengthsp;
((uchar *)interval->type_names[i])[lengthsp]= '\0';
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
if (sql_field->sql_type == MYSQL_TYPE_SET)
Fix for bug#15316 SET value having comma not correctly handled disallow the use of comma in SET members
2006-03-29 16:52:26 +02:00
{
if (cs->coll->instr(cs, interval->type_names[i],
interval->type_lengths[i],
comma_buf, comma_length, NULL, 0))
{
post-merge fix
2006-03-30 06:13:25 +02:00
my_error(ER_ILLEGAL_VALUE_FOR_TYPE, MYF(0), "set", tmp->ptr());
Fix for bug#15316 SET value having comma not correctly handled disallow the use of comma in SET members
2006-03-29 16:52:26 +02:00
DBUG_RETURN(-1);
}
}
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
}
Fix for bug #6840: Default value is not checked in ALTER column SET DEFAULT 'x'
2004-12-07 12:08:56 +01:00
sql_field->interval_list.empty(); // Don't need interval_list anymore
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
}
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
if (sql_field->sql_type == MYSQL_TYPE_SET)
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
{
Update results for new varchar handling Fixed compiler warnings String results in CREATE ... SELECT are now created as CHAR(0), VARCHAR(X) or TEXT() depending on item->max_length
2004-12-07 14:47:00 +01:00
uint32 field_length;
BUG#19145: mysqld crashes if you set the default value of an enum field to NULL Now test for NULLness the pointers returned from objects created from the default value. Pushing patch on behalf of cmiller.
2006-04-28 18:15:29 +02:00
if (sql_field->def != NULL)
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
{
char *not_used;
uint not_used2;
bool not_found= 0;
String str, *def= sql_field->def->val_str(&str);
BUG#19145: mysqld crashes if you set the default value of an enum field to NULL Now test for NULLness the pointers returned from objects created from the default value. Pushing patch on behalf of cmiller.
2006-04-28 18:15:29 +02:00
if (def == NULL) /* SQL "NULL" maps to NULL */
{
if ((sql_field->flags & NOT_NULL_FLAG) != 0)
{
my_error(ER_INVALID_DEFAULT, MYF(0), sql_field->field_name);
DBUG_RETURN(-1);
}
/* else, NULL is an allowed value */
(void) find_set(interval, NULL, 0,
cs, &not_used, &not_used2, &not_found);
}
else /* not NULL */
{
(void) find_set(interval, def->ptr(), def->length(),
cs, &not_used, &not_used2, &not_found);
}
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
if (not_found)
{
my_error(ER_INVALID_DEFAULT, MYF(0), sql_field->field_name);
DBUG_RETURN(-1);
}
}
Update results for new varchar handling Fixed compiler warnings String results in CREATE ... SELECT are now created as CHAR(0), VARCHAR(X) or TEXT() depending on item->max_length
2004-12-07 14:47:00 +01:00
calculate_interval_lengths(cs, interval, &dummy, &field_length);
sql_field->length= field_length + (interval->count - 1);
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
}
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
else /* MYSQL_TYPE_ENUM */
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
{
Update results for new varchar handling Fixed compiler warnings String results in CREATE ... SELECT are now created as CHAR(0), VARCHAR(X) or TEXT() depending on item->max_length
2004-12-07 14:47:00 +01:00
uint32 field_length;
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
DBUG_ASSERT(sql_field->sql_type == MYSQL_TYPE_ENUM);
BUG#19145: mysqld crashes if you set the default value of an enum field to NULL Now test for NULLness the pointers returned from objects created from the default value. Pushing patch on behalf of cmiller.
2006-04-28 18:15:29 +02:00
if (sql_field->def != NULL)
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
{
String str, *def= sql_field->def->val_str(&str);
BUG#19145: mysqld crashes if you set the default value of an enum field to NULL Now test for NULLness the pointers returned from objects created from the default value. Pushing patch on behalf of cmiller.
2006-04-28 18:15:29 +02:00
if (def == NULL) /* SQL "NULL" maps to NULL */
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
{
BUG#19145: mysqld crashes if you set the default value of an enum field to NULL Now test for NULLness the pointers returned from objects created from the default value. Pushing patch on behalf of cmiller.
2006-04-28 18:15:29 +02:00
if ((sql_field->flags & NOT_NULL_FLAG) != 0)
{
my_error(ER_INVALID_DEFAULT, MYF(0), sql_field->field_name);
DBUG_RETURN(-1);
}
/* else, the defaults yield the correct length for NULLs. */
}
else /* not NULL */
{
def->length(cs->cset->lengthsp(cs, def->ptr(), def->length()));
if (find_type2(interval, def->ptr(), def->length(), cs) == 0) /* not found */
{
my_error(ER_INVALID_DEFAULT, MYF(0), sql_field->field_name);
DBUG_RETURN(-1);
}
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
}
}
Update results for new varchar handling Fixed compiler warnings String results in CREATE ... SELECT are now created as CHAR(0), VARCHAR(X) or TEXT() depending on item->max_length
2004-12-07 14:47:00 +01:00
calculate_interval_lengths(cs, interval, &field_length, &dummy);
sql_field->length= field_length;
Bug #6379: ENUM values are incorrectly converted - add_field_to_list() now uses <List>String instead of TYPELIB to be able to distinguish literals 'aaa' and hex literals 0xaabbcc. - move some code from add_field_to_list() where we don't know column charset yet, to mysql_prepare_table(), where we do.
2004-12-02 09:48:43 +01:00
}
set_if_smaller(sql_field->length, MAX_FIELD_WIDTH-1);
}
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
if (sql_field->sql_type == MYSQL_TYPE_BIT)
Bit type support for non-MyISAM tables.
2005-04-12 09:27:43 +02:00
{
Bit type: code clean-up.
2005-04-12 20:12:00 +02:00
sql_field->pack_flag= FIELDFLAG_NUMBER;
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
if (file->ha_table_flags() & HA_CAN_BIT_FIELD)
Bit type support for non-MyISAM tables.
2005-04-12 09:27:43 +02:00
total_uneven_bit_length+= sql_field->length & 7;
else
sql_field->pack_flag|= FIELDFLAG_TREAT_BIT_AS_CHAR;
}
mysql_create_table now accepts field lengths in CREATE form, not in internal form. This is to simplify fixing indexes to mean number of chars rather than bytes too.
2003-09-10 14:25:26 +02:00
sql_field->create_length_to_internal_length();
Fixes while reviewing new code Added option --count to mysqlshow (to show number of rows) Fixed possible core dump in information schema
2005-05-06 10:39:30 +02:00
if (prepare_blob_field(thd, sql_field))
DBUG_RETURN(-1);
stricter checks on FULLTEXT index creating
2003-10-15 13:40:20 +02:00
Import changeset
2000-07-31 21:29:14 +02:00
if (!(sql_field->flags & NOT_NULL_FLAG))
null_fields++;
bit type
2004-12-17 15:06:05 +01:00
Check for empty table/column names
2003-03-14 16:08:42 +01:00
if (check_column_name(sql_field->field_name))
{
Delete duplicate error message names
2003-11-18 16:28:00 +01:00
my_error(ER_WRONG_COLUMN_NAME, MYF(0), sql_field->field_name);
Check for empty table/column names
2003-03-14 16:08:42 +01:00
DBUG_RETURN(-1);
}
Merge with 4.0
2003-03-16 18:17:54 +01:00
Fixes and code cleanups after merge with 4.0.3 Warning handling and initial prepared statement handling (last not complete yet) Changed a lot of functions that returned 0/1 to my_bool type. GRANT handling now uses read/write locks instead of mutex Change basic net functions to use THD instead of NET (needed for 4.1 protocol) Use my_sprintf instead of sprintf() + strlen() Added alloc_query() to be able to chare query initialization code with prepared statements. Cleanup handling of SHOW COUNT(*) WARNINGS and SELECT LAST_INSERT_ID() Note that the following test fails (will be fixed ASAP): sub_select, union, rpl_rotate_logs and rpl_mystery22
2002-10-02 12:33:08 +02:00
/* Check if we have used the same field name before */
for (dup_no=0; (dup_field=it2++) != sql_field; dup_no++)
Import changeset
2000-07-31 21:29:14 +02:00
{
CREATE ... SELECT extension (WL 317)
2002-07-20 16:36:56 +02:00
if (my_strcasecmp(system_charset_info,
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
sql_field->field_name,
dup_field->field_name) == 0)
Import changeset
2000-07-31 21:29:14 +02:00
{
Fixes and code cleanups after merge with 4.0.3 Warning handling and initial prepared statement handling (last not complete yet) Changed a lot of functions that returned 0/1 to my_bool type. GRANT handling now uses read/write locks instead of mutex Change basic net functions to use THD instead of NET (needed for 4.1 protocol) Use my_sprintf instead of sprintf() + strlen() Added alloc_query() to be able to chare query initialization code with prepared statements. Cleanup handling of SHOW COUNT(*) WARNINGS and SELECT LAST_INSERT_ID() Note that the following test fails (will be fixed ASAP): sub_select, union, rpl_rotate_logs and rpl_mystery22
2002-10-02 12:33:08 +02:00
/*
If this was a CREATE ... SELECT statement, accept a field
redefinition if we are changing a field in the SELECT part
*/
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
if (field_no < select_field_pos || dup_no >= select_field_pos)
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_DUP_FIELDNAME, MYF(0), sql_field->field_name);
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
DBUG_RETURN(-1);
}
else
{
Fixes and code cleanups after merge with 4.0.3 Warning handling and initial prepared statement handling (last not complete yet) Changed a lot of functions that returned 0/1 to my_bool type. GRANT handling now uses read/write locks instead of mutex Change basic net functions to use THD instead of NET (needed for 4.1 protocol) Use my_sprintf instead of sprintf() + strlen() Added alloc_query() to be able to chare query initialization code with prepared statements. Cleanup handling of SHOW COUNT(*) WARNINGS and SELECT LAST_INSERT_ID() Note that the following test fails (will be fixed ASAP): sub_select, union, rpl_rotate_logs and rpl_mystery22
2002-10-02 12:33:08 +02:00
/* Field redefined */
BUG#14480: post-fix: use the default field value from CREATE list too.
2005-11-09 07:34:46 +01:00
sql_field->def= dup_field->def;
Fix: create table t1 (a char(10) character set cp1251) SELECT _koi8r'blabla' as a The above query created a field of koi8r charset, not cp1251 Change: CREATE TABLE a (a CHAR(1) CHARACTER SET utf8) Length now means character length, not byte length. The above creates a field that guarantees can store a multibyte value 1 character long. For utf8 the above creates a field that can store 3 bytes.
2003-08-11 15:18:34 +02:00
sql_field->sql_type= dup_field->sql_type;
CLIENT_MULTI_QUERIES -> CLIENT_MULTI_STATEMENTS New multi-key-cache handling. This was needed becasue the old one didn't work reliable with MERGE tables. ALTER TABLE table_name ... CHARACTER SET ... now changes all char/varchar/text columns to the given character set (One must use ALTER TABLE ... DEFAULT CHARACTER SET ... to change the default character set) Fixed that have_compress is detected properly (fixes problems with func_compress.test on platforms without zlib) New syntax for CACHE INDEX ('keys' is optional if no index name is given and one mentions the key cache name only ones) Removed compiler warnings Added mysql_set_server_option() to allow clients like PHP to easaily set/reset the multi-statement flag.
2003-11-18 12:47:27 +01:00
sql_field->charset= (dup_field->charset ?
dup_field->charset :
create_info->default_table_charset);
A fix and a test case for Bug#13134 "Length of VARCHAR() utf8 column is increasing when table is recreated with PS/SP": make use of create_field::char_length more consistent in the code. Reinit create_field::length from create_field::char_length for every execution of a prepared statement (actually fixes the bug).
2006-02-21 17:52:20 +01:00
sql_field->length= dup_field->char_length;
BUG#14139: When handling "CREATE TABLE(field_X type_spec,...) SELECT smth AS field_X, ...." avoid multiplying length of field_X by charset->mbmaxlen twice when calculating space required for field_X in the new table.
2005-10-25 22:56:17 +02:00
sql_field->pack_length= dup_field->pack_length;
Update results for new varchar handling Fixed compiler warnings String results in CREATE ... SELECT are now created as CHAR(0), VARCHAR(X) or TEXT() depending on item->max_length
2004-12-07 14:47:00 +01:00
sql_field->key_length= dup_field->key_length;
Fix: create table t1 (a char(10) character set cp1251) SELECT _koi8r'blabla' as a The above query created a field of koi8r charset, not cp1251 Change: CREATE TABLE a (a CHAR(1) CHARACTER SET utf8) Length now means character length, not byte length. The above creates a field that guarantees can store a multibyte value 1 character long. For utf8 the above creates a field that can store 3 bytes.
2003-08-11 15:18:34 +02:00
sql_field->create_length_to_internal_length();
Fixes and code cleanups after merge with 4.0.3 Warning handling and initial prepared statement handling (last not complete yet) Changed a lot of functions that returned 0/1 to my_bool type. GRANT handling now uses read/write locks instead of mutex Change basic net functions to use THD instead of NET (needed for 4.1 protocol) Use my_sprintf instead of sprintf() + strlen() Added alloc_query() to be able to chare query initialization code with prepared statements. Cleanup handling of SHOW COUNT(*) WARNINGS and SELECT LAST_INSERT_ID() Note that the following test fails (will be fixed ASAP): sub_select, union, rpl_rotate_logs and rpl_mystery22
2002-10-02 12:33:08 +02:00
sql_field->decimals= dup_field->decimals;
sql_field->unireg_check= dup_field->unireg_check;
BUG#14480, attempt2: In CREATE ... SELECT ..., don't count the same field twice when calculating table->null_fields.
2005-11-07 07:23:43 +01:00
/*
We're making one field from two, the result field will have
dup_field->flags as flags. If we've incremented null_fields
because of sql_field->flags, decrement it back.
*/
if (!(sql_field->flags & NOT_NULL_FLAG))
null_fields--;
sql_field->flags= dup_field->flags;
fix for bug#12913 (Simple SQL can crash server or connection) (not initialized member leads to server crash)
2005-09-01 00:13:02 +02:00
sql_field->interval= dup_field->interval;
Fixes and code cleanups after merge with 4.0.3 Warning handling and initial prepared statement handling (last not complete yet) Changed a lot of functions that returned 0/1 to my_bool type. GRANT handling now uses read/write locks instead of mutex Change basic net functions to use THD instead of NET (needed for 4.1 protocol) Use my_sprintf instead of sprintf() + strlen() Added alloc_query() to be able to chare query initialization code with prepared statements. Cleanup handling of SHOW COUNT(*) WARNINGS and SELECT LAST_INSERT_ID() Note that the following test fails (will be fixed ASAP): sub_select, union, rpl_rotate_logs and rpl_mystery22
2002-10-02 12:33:08 +02:00
it2.remove(); // Remove first (create) definition
select_field_pos--;
break;
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
}
Import changeset
2000-07-31 21:29:14 +02:00
}
}
Update results for new varchar handling Fixed compiler warnings String results in CREATE ... SELECT are now created as CHAR(0), VARCHAR(X) or TEXT() depending on item->max_length
2004-12-07 14:47:00 +01:00
/* Don't pack rows in old tables if the user has requested this */
if ((sql_field->flags & BLOB_FLAG) ||
sql_field->sql_type == MYSQL_TYPE_VARCHAR &&
create_info->row_type != ROW_TYPE_FIXED)
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
(*db_options)|= HA_OPTION_PACK_RECORD;
Import changeset
2000-07-31 21:29:14 +02:00
it2.rewind();
}
Change create_field->offset to store offset from start of fields, independent of null bits. Count null_bits separately from field offsets and adjust them in case of primary key parts. (Previously a CREATE TABLE with a lot of null fields that was part of a primary key caused MySQL to wrongly count the number of bytes needed to store null bits) This is a more complete bug fix for #6236
2005-05-13 10:11:50 +02:00
/* record_offset will be increased with 'length-of-null-bits' later */
record_offset= 0;
Merge with 4.1
2005-05-14 17:31:22 +02:00
null_fields+= total_uneven_bit_length;
Import changeset
2000-07-31 21:29:14 +02:00
it.rewind();
while ((sql_field=it++))
{
Remove compiler warnings and remove not used variables (Found during build process)
2005-02-25 15:53:22 +01:00
DBUG_ASSERT(sql_field->charset != 0);
CLIENT_MULTI_QUERIES -> CLIENT_MULTI_STATEMENTS New multi-key-cache handling. This was needed becasue the old one didn't work reliable with MERGE tables. ALTER TABLE table_name ... CHARACTER SET ... now changes all char/varchar/text columns to the given character set (One must use ALTER TABLE ... DEFAULT CHARACTER SET ... to change the default character set) Fixed that have_compress is detected properly (fixes problems with func_compress.test on platforms without zlib) New syntax for CACHE INDEX ('keys' is optional if no index name is given and one mentions the key cache name only ones) Removed compiler warnings Added mysql_set_server_option() to allow clients like PHP to easaily set/reset the multi-statement flag.
2003-11-18 12:47:27 +01:00
Cleanup during reviews Removed some optional arguments Fixed portability problem in federated tests
2005-03-16 15:11:01 +01:00
if (prepare_create_field(sql_field, &blob_columns,
&timestamps, &timestamps_with_niladic,
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
file->ha_table_flags()))
SCRUM: WL#1163 (Making spatial code optional) Pack of changes to do in sql/ code.
2004-01-15 18:06:22 +01:00
DBUG_RETURN(-1);
Hack to ensure that CHAR's created in 5.0 are not converted to VARCHAR even if the row type is dynamic (For 4.1 tables old 'VARCHAR' fields are converted to true VARCHAR in the next ALTER TABLE) This ensures that one can use MySQL 5.0 privilege tables with MySQL 4.1
2005-03-22 14:48:06 +01:00
if (sql_field->sql_type == MYSQL_TYPE_VARCHAR)
Bug#3788 Crashes with stored procedure return non-string values Also fixes Bug#2773
2005-03-04 22:14:35 +01:00
create_info->varchar= 1;
Change create_field->offset to store offset from start of fields, independent of null bits. Count null_bits separately from field offsets and adjust them in case of primary key parts. (Previously a CREATE TABLE with a lot of null fields that was part of a primary key caused MySQL to wrongly count the number of bytes needed to store null bits) This is a more complete bug fix for #6236
2005-05-13 10:11:50 +02:00
sql_field->offset= record_offset;
Import changeset
2000-07-31 21:29:14 +02:00
if (MTYP_TYPENR(sql_field->unireg_check) == Field::NEXT_NUMBER)
auto_increment++;
Change create_field->offset to store offset from start of fields, independent of null bits. Count null_bits separately from field offsets and adjust them in case of primary key parts. (Previously a CREATE TABLE with a lot of null fields that was part of a primary key caused MySQL to wrongly count the number of bytes needed to store null bits) This is a more complete bug fix for #6236
2005-05-13 10:11:50 +02:00
record_offset+= sql_field->pack_length;
Import changeset
2000-07-31 21:29:14 +02:00
}
WL#1266 "Separate auto-set logic from TIMESTAMP type." Final version of patch. Adds support for specifying of DEFAULT NOW() and/or ON UPDATE NOW() clauses for TIMESTAMP field definition. Current implementation allows only one such field per table and uses several unireg types for storing info about this properties of field. It should be replaced with better implementation when new .frm format is introduced.
2004-04-02 08:12:53 +02:00
if (timestamps_with_niladic > 1)
{
post-review fixes
2004-11-12 13:34:00 +01:00
my_message(ER_TOO_MUCH_AUTO_TIMESTAMP_COLS,
ER(ER_TOO_MUCH_AUTO_TIMESTAMP_COLS), MYF(0));
WL#1266 "Separate auto-set logic from TIMESTAMP type." Final version of patch. Adds support for specifying of DEFAULT NOW() and/or ON UPDATE NOW() clauses for TIMESTAMP field definition. Current implementation allows only one such field per table and uses several unireg types for storing info about this properties of field. It should be replaced with better implementation when new .frm format is introduced.
2004-04-02 08:12:53 +02:00
DBUG_RETURN(-1);
}
Import changeset
2000-07-31 21:29:14 +02:00
if (auto_increment > 1)
{
post-review fixes
2004-11-12 13:34:00 +01:00
my_message(ER_WRONG_AUTO_KEY, ER(ER_WRONG_AUTO_KEY), MYF(0));
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(-1);
}
if (auto_increment &&
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
(file->ha_table_flags() & HA_NO_AUTO_INCREMENT))
Import changeset
2000-07-31 21:29:14 +02:00
{
post-review fixes
2004-11-12 13:34:00 +01:00
my_message(ER_TABLE_CANT_HANDLE_AUTO_INCREMENT,
ER(ER_TABLE_CANT_HANDLE_AUTO_INCREMENT), MYF(0));
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(-1);
}
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
if (blob_columns && (file->ha_table_flags() & HA_NO_BLOBS))
Import changeset
2000-07-31 21:29:14 +02:00
{
post-review fixes
2004-11-12 13:34:00 +01:00
my_message(ER_TABLE_CANT_HANDLE_BLOB, ER(ER_TABLE_CANT_HANDLE_BLOB),
MYF(0));
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(-1);
}
/* Create keys */
Bugfix when using a multi-part unique key in the SET and WHERE part Put PRIMARY KEY and UNIQUE keys before other keys Fixes for auto_increment keys for BDB tables
2000-12-10 20:10:03 +01:00
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
List_iterator<Key> key_iterator(*keys), key_iterator2(*keys);
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
uint key_parts=0, fk_key_count=0;
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
bool primary_key=0,unique_key=0;
Introduce keys in child tables corresponding to FOREIGN KEYs Remove redundant keys in CREATE TABLE and ALTER TABLE
2004-04-21 12:15:43 +02:00
Key *key, *key2;
Fix for FLUSH QUERY CACHE Fix for new bug in CREATE TABLE when sorting keys.
2002-01-05 21:51:42 +01:00
uint tmp, key_number;
Introduce keys in child tables corresponding to FOREIGN KEYs Remove redundant keys in CREATE TABLE and ALTER TABLE
2004-04-21 12:15:43 +02:00
/* special marker for keys to be ignored */
static char ignore_key[1];
Bugfix when using a multi-part unique key in the SET and WHERE part Put PRIMARY KEY and UNIQUE keys before other keys Fixes for auto_increment keys for BDB tables
2000-12-10 20:10:03 +01:00
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
/* Calculate number of key segements */
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
*key_count= 0;
Bugfix when using a multi-part unique key in the SET and WHERE part Put PRIMARY KEY and UNIQUE keys before other keys Fixes for auto_increment keys for BDB tables
2000-12-10 20:10:03 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
while ((key=key_iterator++))
{
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
DBUG_PRINT("info", ("key name: '%s' type: %d", key->name ? key->name :
"(none)" , key->type));
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
if (key->type == Key::FOREIGN_KEY)
{
fk_key_count++;
foreign_key *fk_key= (foreign_key*) key;
if (fk_key->ref_columns.elements &&
fk_key->ref_columns.elements != fk_key->columns.elements)
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_WRONG_FK_DEF, MYF(0),
(fk_key->name ? fk_key->name : "foreign key without name"),
ER(ER_KEY_REF_DO_NOT_MATCH_TABLE_REF));
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
DBUG_RETURN(-1);
}
continue;
}
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
(*key_count)++;
handler interface cleanups: more logical table/index_flags return HA_ERR_WRONG_COMMAND instead of abstract methods where appropriate max_keys and other limits renamed to max_supported_keys/etc max_keys/etc are now wrappers to max_supported_keys/etc ha_index_init/ha_rnd_init/ha_index_end/ha_rnd_end are now wrappers to real {index,rnd}_{init,end} to enforce strict pairing
2004-06-23 12:29:05 +02:00
tmp=file->max_key_parts();
Import changeset
2000-07-31 21:29:14 +02:00
if (key->columns.elements > tmp)
{
my_error(ER_TOO_MANY_KEY_PARTS,MYF(0),tmp);
DBUG_RETURN(-1);
}
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
if (key->name && strlen(key->name) > NAME_LEN)
Import changeset
2000-07-31 21:29:14 +02:00
{
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
my_error(ER_TOO_LONG_IDENT, MYF(0), key->name);
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(-1);
}
Introduce keys in child tables corresponding to FOREIGN KEYs Remove redundant keys in CREATE TABLE and ALTER TABLE
2004-04-21 12:15:43 +02:00
key_iterator2.rewind ();
Don't automaticly generate a new key for a foreign key constraint if there is already a usable key. Prefer not automatic keys before automatic keys. If there is two conf
2004-05-11 23:29:52 +02:00
if (key->type != Key::FOREIGN_KEY)
Introduce keys in child tables corresponding to FOREIGN KEYs Remove redundant keys in CREATE TABLE and ALTER TABLE
2004-04-21 12:15:43 +02:00
{
Don't automaticly generate a new key for a foreign key constraint if there is already a usable key. Prefer not automatic keys before automatic keys. If there is two conf
2004-05-11 23:29:52 +02:00
while ((key2 = key_iterator2++) != key)
Introduce keys in child tables corresponding to FOREIGN KEYs Remove redundant keys in CREATE TABLE and ALTER TABLE
2004-04-21 12:15:43 +02:00
{
sql_table.cc: Fix bug #3749: if there were several indexs that made an automatically generated FOREIGN KEY index redundant, then MySQL miscalculated the number of keys, and the AUTO_INCREMENT flag was forgotten by MySQL from a PRIMARY KEY. There were probably a multitude of other errors caused by this.
2004-05-14 08:02:06 +02:00
/*
Better fix for bug #3749 (bug in deleting automatic generated foreign keys)
2004-05-15 10:57:40 +02:00
foreign_key_prefix(key, key2) returns 0 if key or key2, or both, is
'generated', and a generated key is a prefix of the other key.
Then we do not need the generated shorter key.
sql_table.cc: Fix bug #3749: if there were several indexs that made an automatically generated FOREIGN KEY index redundant, then MySQL miscalculated the number of keys, and the AUTO_INCREMENT flag was forgotten by MySQL from a PRIMARY KEY. There were probably a multitude of other errors caused by this.
2004-05-14 08:02:06 +02:00
*/
Better fix for bug #3749 (bug in deleting automatic generated foreign keys)
2004-05-15 10:57:40 +02:00
if ((key2->type != Key::FOREIGN_KEY &&
key2->name != ignore_key &&
!foreign_key_prefix(key, key2)))
Don't automaticly generate a new key for a foreign key constraint if there is already a usable key. Prefer not automatic keys before automatic keys. If there is two conf
2004-05-11 23:29:52 +02:00
{
better fix for bug#3749 - do not consider already removed keys in key removal process
2004-05-14 12:49:18 +02:00
/* TODO: issue warning message */
Don't automaticly generate a new key for a foreign key constraint if there is already a usable key. Prefer not automatic keys before automatic keys. If there is two conf
2004-05-11 23:29:52 +02:00
/* mark that the generated key should be ignored */
if (!key2->generated ||
(key->generated && key->columns.elements <
key2->columns.elements))
key->name= ignore_key;
else
{
better fix for bug#3749 - do not consider already removed keys in key removal process
2004-05-14 12:49:18 +02:00
key2->name= ignore_key;
key_parts-= key2->columns.elements;
(*key_count)--;
Don't automaticly generate a new key for a foreign key constraint if there is already a usable key. Prefer not automatic keys before automatic keys. If there is two conf
2004-05-11 23:29:52 +02:00
}
break;
}
Introduce keys in child tables corresponding to FOREIGN KEYs Remove redundant keys in CREATE TABLE and ALTER TABLE
2004-04-21 12:15:43 +02:00
}
}
if (key->name != ignore_key)
key_parts+=key->columns.elements;
else
(*key_count)--;
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
if (key->name && !tmp_table && (key->type != Key::PRIMARY) &&
fixed bug #856 'Naming a key "Primary" causes trouble'
2004-03-13 22:31:30 +01:00
!my_strcasecmp(system_charset_info,key->name,primary_key_name))
{
my_error(ER_WRONG_NAME_FOR_INDEX, MYF(0), key->name);
DBUG_RETURN(-1);
}
Bugfix when using a multi-part unique key in the SET and WHERE part Put PRIMARY KEY and UNIQUE keys before other keys Fixes for auto_increment keys for BDB tables
2000-12-10 20:10:03 +01:00
}
handler interface cleanups: more logical table/index_flags return HA_ERR_WRONG_COMMAND instead of abstract methods where appropriate max_keys and other limits renamed to max_supported_keys/etc max_keys/etc are now wrappers to max_supported_keys/etc ha_index_init/ha_rnd_init/ha_index_end/ha_rnd_end are now wrappers to real {index,rnd}_{init,end} to enforce strict pairing
2004-06-23 12:29:05 +02:00
tmp=file->max_keys();
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
if (*key_count > tmp)
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
{
my_error(ER_TOO_MANY_KEYS,MYF(0),tmp);
DBUG_RETURN(-1);
}
Bugfix when using a multi-part unique key in the SET and WHERE part Put PRIMARY KEY and UNIQUE keys before other keys Fixes for auto_increment keys for BDB tables
2000-12-10 20:10:03 +01:00
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
(*key_info_buffer)= key_info= (KEY*) sql_calloc(sizeof(KEY) * (*key_count));
Bugfix when using a multi-part unique key in the SET and WHERE part Put PRIMARY KEY and UNIQUE keys before other keys Fixes for auto_increment keys for BDB tables
2000-12-10 20:10:03 +01:00
key_part_info=(KEY_PART_INFO*) sql_calloc(sizeof(KEY_PART_INFO)*key_parts);
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
if (!*key_info_buffer || ! key_part_info)
Bugfix when using a multi-part unique key in the SET and WHERE part Put PRIMARY KEY and UNIQUE keys before other keys Fixes for auto_increment keys for BDB tables
2000-12-10 20:10:03 +01:00
DBUG_RETURN(-1); // Out of memory
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
key_iterator.rewind();
Fix for FLUSH QUERY CACHE Fix for new bug in CREATE TABLE when sorting keys.
2002-01-05 21:51:42 +01:00
key_number=0;
few small bug fixes ...
2002-11-15 15:37:44 +01:00
for (; (key=key_iterator++) ; key_number++)
Bugfix when using a multi-part unique key in the SET and WHERE part Put PRIMARY KEY and UNIQUE keys before other keys Fixes for auto_increment keys for BDB tables
2000-12-10 20:10:03 +01:00
{
uint key_length=0;
key_part_spec *column;
Introduce keys in child tables corresponding to FOREIGN KEYs Remove redundant keys in CREATE TABLE and ALTER TABLE
2004-04-21 12:15:43 +02:00
if (key->name == ignore_key)
{
/* ignore redundant keys */
do
key=key_iterator++;
while (key && key->name == ignore_key);
if (!key)
break;
}
Added support for key_block_size for key and table level (WL#602) Added support for key_block_size to MyISAM. Simplify interface to 'new Key' to make it easier to add new key options. mysqld option --new is used to define where key options are printed. (In 5.3 we should move all key options to after key part definition to avoid problem with reserved names) Fixed some compiler warnings and a memory leak in ssl
2006-05-03 14:59:17 +02:00
switch (key->type) {
few small bug fixes ...
2002-11-15 15:37:44 +01:00
case Key::MULTIPLE:
Don't automaticly generate a new key for a foreign key constraint if there is already a usable key. Prefer not automatic keys before automatic keys. If there is two conf
2004-05-11 23:29:52 +02:00
key_info->flags= 0;
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
break;
few small bug fixes ...
2002-11-15 15:37:44 +01:00
case Key::FULLTEXT:
Don't automaticly generate a new key for a foreign key constraint if there is already a usable key. Prefer not automatic keys before automatic keys. If there is two conf
2004-05-11 23:29:52 +02:00
key_info->flags= HA_FULLTEXT;
Cleanups after review of WL#602 Fixed warnings from test suite Some fixes in mysql-test-run script to catch more warnings
2006-05-03 18:40:52 +02:00
if ((key_info->parser_name= &key->key_create_info.parser_name)->str)
WL#2575 - Fulltext: Parser plugin for FTS WL#2763 - MySQL plugin interface: step 1 Manual merge from CNET tree.
2005-11-06 13:13:06 +01:00
key_info->flags|= HA_USES_PARSER;
Added support for key_block_size for key and table level (WL#602) Added support for key_block_size to MyISAM. Simplify interface to 'new Key' to make it easier to add new key options. mysqld option --new is used to define where key options are printed. (In 5.3 we should move all key options to after key part definition to avoid problem with reserved names) Fixed some compiler warnings and a memory leak in ssl
2006-05-03 14:59:17 +02:00
else
key_info->parser_name= 0;
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
break;
few small bug fixes ...
2002-11-15 15:37:44 +01:00
case Key::SPATIAL:
SCRUM: WL#1163 (Making spatial code optional) Pack of changes to do in sql/ code.
2004-01-15 18:06:22 +01:00
#ifdef HAVE_SPATIAL
Don't automaticly generate a new key for a foreign key constraint if there is already a usable key. Prefer not automatic keys before automatic keys. If there is two conf
2004-05-11 23:29:52 +02:00
key_info->flags= HA_SPATIAL;
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
break;
SCRUM: WL#1163 (Making spatial code optional) Pack of changes to do in sql/ code.
2004-01-15 18:06:22 +01:00
#else
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_FEATURE_DISABLED, MYF(0),
sym_group_geom.name, sym_group_geom.needed_define);
SCRUM: WL#1163 (Making spatial code optional) Pack of changes to do in sql/ code.
2004-01-15 18:06:22 +01:00
DBUG_RETURN(-1);
#endif
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
case Key::FOREIGN_KEY:
key_number--; // Skip this key
continue;
default:
Don't automaticly generate a new key for a foreign key constraint if there is already a usable key. Prefer not automatic keys before automatic keys. If there is two conf
2004-05-11 23:29:52 +02:00
key_info->flags = HA_NOSAME;
break;
Added GIS extension
2002-02-22 12:24:42 +01:00
}
Don't automaticly generate a new key for a foreign key constraint if there is already a usable key. Prefer not automatic keys before automatic keys. If there is two conf
2004-05-11 23:29:52 +02:00
if (key->generated)
key_info->flags|= HA_GENERATED_KEY;
Added GIS extension
2002-02-22 12:24:42 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
key_info->key_parts=(uint8) key->columns.elements;
key_info->key_part=key_part_info;
Fix for FLUSH QUERY CACHE Fix for new bug in CREATE TABLE when sorting keys.
2002-01-05 21:51:42 +01:00
key_info->usable_key_parts= key_number;
Cleanups after review of WL#602 Fixed warnings from test suite Some fixes in mysql-test-run script to catch more warnings
2006-05-03 18:40:52 +02:00
key_info->algorithm= key->key_create_info.algorithm;
Import changeset
2000-07-31 21:29:14 +02:00
HA_NO_FULLTEXT_KEY and an appropriate error message
2001-10-23 10:52:51 +02:00
if (key->type == Key::FULLTEXT)
{
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
if (!(file->ha_table_flags() & HA_CAN_FULLTEXT))
HA_NO_FULLTEXT_KEY and an appropriate error message
2001-10-23 10:52:51 +02:00
{
post-review fixes
2004-11-12 13:34:00 +01:00
my_message(ER_TABLE_CANT_HANDLE_FT, ER(ER_TABLE_CANT_HANDLE_FT),
MYF(0));
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
DBUG_RETURN(-1);
HA_NO_FULLTEXT_KEY and an appropriate error message
2001-10-23 10:52:51 +02:00
}
}
Added GIS extension
2002-02-22 12:24:42 +01:00
/*
Make SPATIAL to be RTREE by default
SPATIAL only on BLOB or at least BINARY, this
CHECKSUM TABLE table1, table2, ... [ QUICK | EXTENDED ]
2003-09-03 11:34:32 +02:00
actually should be replaced by special GEOM type
Added GIS extension
2002-02-22 12:24:42 +01:00
in near future when new frm file is ready
checking for proper key parts number:
*/
CHECKSUM TABLE table1, table2, ... [ QUICK | EXTENDED ]
2003-09-03 11:34:32 +02:00
stricter checks on FULLTEXT index creating
2003-10-15 13:40:20 +02:00
/* TODO: Add proper checks if handler supports key_type and algorithm */
Don't automaticly generate a new key for a foreign key constraint if there is already a usable key. Prefer not automatic keys before automatic keys. If there is two conf
2004-05-11 23:29:52 +02:00
if (key_info->flags & HA_SPATIAL)
Fixed problem in fulltest testcase
2002-05-22 17:51:21 +02:00
{
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
if (!(file->ha_table_flags() & HA_CAN_RTREEKEYS))
bug #15860 (SPATIAL keys in INNODB)
2006-04-12 19:05:23 +02:00
{
my_message(ER_TABLE_CANT_HANDLE_SPKEYS, ER(ER_TABLE_CANT_HANDLE_SPKEYS),
MYF(0));
DBUG_RETURN(-1);
}
Fixed problem in fulltest testcase
2002-05-22 17:51:21 +02:00
if (key_info->key_parts != 1)
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_WRONG_ARGUMENTS, MYF(0), "SPATIAL INDEX");
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
DBUG_RETURN(-1);
Added GIS extension
2002-02-22 12:24:42 +01:00
}
Fixed problem in fulltest testcase
2002-05-22 17:51:21 +02:00
}
Cleanups during review stage Added auto-correct of field length for enum/set tables for ALTER TABLE This is becasue of a bug in previous MySQL 4.1 versions where the length for enum/set was set incorrectly after ALTER TABLE
2004-12-02 13:43:51 +01:00
else if (key_info->algorithm == HA_KEY_ALG_RTREE)
Added GIS extension
2002-02-22 12:24:42 +01:00
{
SCRUM: WL#1163 (Making spatial code optional) Pack of changes to do in sql/ code.
2004-01-15 18:06:22 +01:00
#ifdef HAVE_RTREE_KEYS
Fixed problem in fulltest testcase
2002-05-22 17:51:21 +02:00
if ((key_info->key_parts & 1) == 1)
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_WRONG_ARGUMENTS, MYF(0), "RTREE INDEX");
Fixed problem in fulltest testcase
2002-05-22 17:51:21 +02:00
DBUG_RETURN(-1);
Added GIS extension
2002-02-22 12:24:42 +01:00
}
Fixed a lot of wrong memory references as reported by valgrind Portability fixes Added new client function: mysql_get_server_version() New server help code (From Victor Vagin) Fixed wrong usage of binary() Disabled RTREE usage for now.
2003-02-12 20:55:37 +01:00
/* TODO: To be deleted */
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_NOT_SUPPORTED_YET, MYF(0), "RTREE INDEX");
Fixed a lot of wrong memory references as reported by valgrind Portability fixes Added new client function: mysql_get_server_version() New server help code (From Victor Vagin) Fixed wrong usage of binary() Disabled RTREE usage for now.
2003-02-12 20:55:37 +01:00
DBUG_RETURN(-1);
SCRUM: WL#1163 (Making spatial code optional) Pack of changes to do in sql/ code.
2004-01-15 18:06:22 +01:00
#else
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_FEATURE_DISABLED, MYF(0),
sym_group_rtree.name, sym_group_rtree.needed_define);
SCRUM: WL#1163 (Making spatial code optional) Pack of changes to do in sql/ code.
2004-01-15 18:06:22 +01:00
DBUG_RETURN(-1);
#endif
Added GIS extension
2002-02-22 12:24:42 +01:00
}
CHECKSUM TABLE table1, table2, ... [ QUICK | EXTENDED ]
2003-09-03 11:34:32 +02:00
Added support for key_block_size for key and table level (WL#602) Added support for key_block_size to MyISAM. Simplify interface to 'new Key' to make it easier to add new key options. mysqld option --new is used to define where key options are printed. (In 5.3 we should move all key options to after key part definition to avoid problem with reserved names) Fixed some compiler warnings and a memory leak in ssl
2006-05-03 14:59:17 +02:00
/* Take block size from key part or table part */
/*
TODO: Add warning if block size changes. We can't do it here, as
this may depend on the size of the key
*/
Cleanups after review of WL#602 Fixed warnings from test suite Some fixes in mysql-test-run script to catch more warnings
2006-05-03 18:40:52 +02:00
key_info->block_size= (key->key_create_info.block_size ?
key->key_create_info.block_size :
Added support for key_block_size for key and table level (WL#602) Added support for key_block_size to MyISAM. Simplify interface to 'new Key' to make it easier to add new key options. mysqld option --new is used to define where key options are printed. (In 5.3 we should move all key options to after key part definition to avoid problem with reserved names) Fixed some compiler warnings and a memory leak in ssl
2006-05-03 14:59:17 +02:00
create_info->key_block_size);
if (key_info->block_size)
key_info->flags|= HA_USES_BLOCK_SIZE;
Bug#6252 - Duplicate columns in keys should fail Added check for duplicate column in key Added tests and fixed tests which exploit bug
2004-11-22 19:07:04 +01:00
List_iterator<key_part_spec> cols(key->columns), cols2(key->columns);
stricter checks on FULLTEXT index creating
2003-10-15 13:40:20 +02:00
CHARSET_INFO *ft_key_charset=0; // for FULLTEXT
Import changeset
2000-07-31 21:29:14 +02:00
for (uint column_nr=0 ; (column=cols++) ; column_nr++)
{
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
uint length;
Cleanups during review stage Added auto-correct of field length for enum/set tables for ALTER TABLE This is becasue of a bug in previous MySQL 4.1 versions where the length for enum/set was set incorrectly after ALTER TABLE
2004-12-02 13:43:51 +01:00
key_part_spec *dup_column;
Import changeset
2000-07-31 21:29:14 +02:00
it.rewind();
field=0;
while ((sql_field=it++) &&
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
my_strcasecmp(system_charset_info,
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
column->field_name,
sql_field->field_name))
Import changeset
2000-07-31 21:29:14 +02:00
field++;
if (!sql_field)
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_KEY_COLUMN_DOES_NOT_EXITS, MYF(0), column->field_name);
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(-1);
}
Cleanups during review stage Added auto-correct of field length for enum/set tables for ALTER TABLE This is becasue of a bug in previous MySQL 4.1 versions where the length for enum/set was set incorrectly after ALTER TABLE
2004-12-02 13:43:51 +01:00
while ((dup_column= cols2++) != column)
Bug#6252 - Duplicate columns in keys should fail Added check for duplicate column in key Added tests and fixed tests which exploit bug
2004-11-22 19:07:04 +01:00
{
if (!my_strcasecmp(system_charset_info,
column->field_name, dup_column->field_name))
{
my_printf_error(ER_DUP_FIELDNAME,
ER(ER_DUP_FIELDNAME),MYF(0),
column->field_name);
DBUG_RETURN(-1);
}
}
cols2.rewind();
bug 283: FULLTEXT index on a TEXT filed converted to a CHAR field doesn't work anymore
2003-04-23 13:44:39 +02:00
if (key->type == Key::FULLTEXT)
stricter checks on FULLTEXT index creating
2003-10-15 13:40:20 +02:00
{
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
if ((sql_field->sql_type != MYSQL_TYPE_STRING &&
sql_field->sql_type != MYSQL_TYPE_VARCHAR &&
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
!f_is_blob(sql_field->pack_flag)) ||
sql_field->charset == &my_charset_bin ||
Allocate memory when a character set is requested: - For simple character sets: from_uni convertion table. - For UCA: alternative weight arrays. Use mbminlen instead of MY_CS_NONTEXT
2004-06-11 13:29:16 +02:00
sql_field->charset->mbminlen > 1 || // ucs2 doesn't work yet
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
(ft_key_charset && sql_field->charset != ft_key_charset))
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_BAD_FT_COLUMN, MYF(0), column->field_name);
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
DBUG_RETURN(-1);
}
ft_key_charset=sql_field->charset;
/*
for fulltext keys keyseg length is 1 for blobs (it's ignored in ft
code anyway, and 0 (set to column width later) for char's. it has
to be correct col width for char's, as char data are not prefixed
with length (unlike blobs, where ft code takes data length from a
data prefix, ignoring column->length).
*/
column->length=test(f_is_blob(sql_field->pack_flag));
stricter checks on FULLTEXT index creating
2003-10-15 13:40:20 +02:00
}
In CREATE syntax: KEY a (field_name(10)) 10 now means "number of characters", not number of bytes.
2003-09-15 11:45:42 +02:00
else
stricter checks on FULLTEXT index creating
2003-10-15 13:40:20 +02:00
{
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
column->length*= sql_field->charset->mbmaxlen;
Fix for bug #12267 (primary key over GEOMETRY field)
2005-09-26 11:55:52 +02:00
if (f_is_blob(sql_field->pack_flag) ||
(f_is_geom(sql_field->pack_flag) && key->type != Key::SPATIAL))
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
{
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
if (!(file->ha_table_flags() & HA_CAN_INDEX_BLOBS))
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_BLOB_USED_AS_KEY, MYF(0), column->field_name);
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
DBUG_RETURN(-1);
}
Additional fix for bug #12267 (Can't use POINT for primary key)
2005-10-15 19:23:13 +02:00
if (f_is_geom(sql_field->pack_flag) && sql_field->geom_type ==
Field::GEOM_POINT)
column->length= 21;
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
if (!column->length)
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_BLOB_KEY_WITHOUT_LENGTH, MYF(0), column->field_name);
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
DBUG_RETURN(-1);
}
}
SCRUM: WL#1163 (Making spatial code optional) Pack of changes to do in sql/ code.
2004-01-15 18:06:22 +01:00
#ifdef HAVE_SPATIAL
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
if (key->type == Key::SPATIAL)
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
{
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
if (!column->length)
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
{
/*
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
4 is: (Xmin,Xmax,Ymin,Ymax), this is for 2D case
Lately we'll extend this code to support more dimensions
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
*/
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
column->length= 4*sizeof(double);
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
}
}
SCRUM: WL#1163 (Making spatial code optional) Pack of changes to do in sql/ code.
2004-01-15 18:06:22 +01:00
#endif
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
if (!(sql_field->flags & NOT_NULL_FLAG))
{
if (key->type == Key::PRIMARY)
{
/* Implicitly set primary key fields to NOT NULL for ISO conf. */
sql_field->flags|= NOT_NULL_FLAG;
sql_field->pack_flag&= ~FIELDFLAG_MAYBE_NULL;
Merge with 4.0
2005-05-13 23:01:40 +02:00
null_fields--;
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
}
else
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
{
key_info->flags|= HA_NULL_PART_KEY;
if (!(file->ha_table_flags() & HA_NULL_IN_KEY))
{
my_error(ER_NULL_COLUMN_IN_INDEX, MYF(0), column->field_name);
DBUG_RETURN(-1);
}
if (key->type == Key::SPATIAL)
{
my_message(ER_SPATIAL_CANT_HAVE_NULL,
ER(ER_SPATIAL_CANT_HAVE_NULL), MYF(0));
DBUG_RETURN(-1);
}
}
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
}
if (MTYP_TYPENR(sql_field->unireg_check) == Field::NEXT_NUMBER)
{
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
if (column_nr == 0 || (file->ha_table_flags() & HA_AUTO_PART_KEY))
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
auto_increment--; // Field is used
}
Import changeset
2000-07-31 21:29:14 +02:00
}
stricter checks on FULLTEXT index creating
2003-10-15 13:40:20 +02:00
Import changeset
2000-07-31 21:29:14 +02:00
key_part_info->fieldnr= field;
key_part_info->offset= (uint16) sql_field->offset;
key_part_info->key_type=sql_field->pack_flag;
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
length= sql_field->key_length;
Import changeset
2000-07-31 21:29:14 +02:00
if (column->length)
{
if (f_is_blob(sql_field->pack_flag))
{
After merge fixes
2005-05-16 14:21:35 +02:00
if ((length=column->length) > max_key_length ||
Give an error if you use a BLOB(X) where X > 255 Fixes for MIT pthreads
2000-10-22 00:19:05 +02:00
length > file->max_key_part_length())
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
{
After merge fixes
2005-05-16 14:21:35 +02:00
length=min(max_key_length, file->max_key_part_length());
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
if (key->type == Key::MULTIPLE)
{
/* not a critical problem */
char warn_buff[MYSQL_ERRMSG_SIZE];
my_snprintf(warn_buff, sizeof(warn_buff), ER(ER_TOO_LONG_KEY),
length);
push_warning(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
ER_TOO_LONG_KEY, warn_buff);
}
else
{
my_error(ER_TOO_LONG_KEY,MYF(0),length);
DBUG_RETURN(-1);
}
}
Import changeset
2000-07-31 21:29:14 +02:00
}
if a key length exceeds the supported maximum and it is safe to auto-decrease it, do it.
2004-02-17 16:57:39 +01:00
else if (!f_is_geom(sql_field->pack_flag) &&
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
(column->length > length ||
((f_is_packed(sql_field->pack_flag) ||
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
((file->ha_table_flags() & HA_NO_PREFIX_CHAR_KEYS) &&
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
(key_info->flags & HA_NOSAME))) &&
column->length != length)))
{
post-review fixes
2004-11-12 13:34:00 +01:00
my_message(ER_WRONG_SUB_KEY, ER(ER_WRONG_SUB_KEY), MYF(0));
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
DBUG_RETURN(-1);
}
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
else if (!(file->ha_table_flags() & HA_NO_PREFIX_CHAR_KEYS))
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
length=column->length;
Import changeset
2000-07-31 21:29:14 +02:00
}
else if (length == 0)
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_WRONG_KEY_COLUMN, MYF(0), column->field_name);
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(-1);
}
BUG#13835: max key length is 1000 bytes when trying to create a fulltext index Allow fulltext index on VARCHAR columns longer than max key length.
2006-01-13 18:25:12 +01:00
if (length > file->max_key_part_length() && key->type != Key::FULLTEXT)
Bug #2699 UTF8 breaks primary keys for cols > 85 characters
2004-02-13 14:58:02 +01:00
{
Fixed bug#17530: Incorrect key truncation on table creation caused server crash. When a too long field is used for a key, only a prefix part of the field is used. Length is reduced to the max key length allowed for storage. But if the field have a multibyte charset it is possible to break multibyte char sequence. This leads to the failed assertion in the innodb code and server crash when a record is inserted. The make_prepare_table() now aligns truncated key length to the boundary of multibyte char.
2006-02-21 16:09:32 +01:00
length= file->max_key_part_length();
/* Align key length to multibyte char boundary */
length-= length % sql_field->charset->mbmaxlen;
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
if (key->type == Key::MULTIPLE)
{
/* not a critical problem */
char warn_buff[MYSQL_ERRMSG_SIZE];
my_snprintf(warn_buff, sizeof(warn_buff), ER(ER_TOO_LONG_KEY),
length);
push_warning(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
ER_TOO_LONG_KEY, warn_buff);
}
else
{
my_error(ER_TOO_LONG_KEY,MYF(0),length);
DBUG_RETURN(-1);
}
Bug #2699 UTF8 breaks primary keys for cols > 85 characters
2004-02-13 14:58:02 +01:00
}
key_part_info->length=(uint16) length;
Import changeset
2000-07-31 21:29:14 +02:00
/* Use packed keys for long strings on the first column */
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
if (!((*db_options) & HA_OPTION_NO_PACK_KEYS) &&
Import changeset
2000-07-31 21:29:14 +02:00
(length >= KEY_DEFAULT_PACK_LENGTH &&
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
(sql_field->sql_type == MYSQL_TYPE_STRING ||
sql_field->sql_type == MYSQL_TYPE_VARCHAR ||
Import changeset
2000-07-31 21:29:14 +02:00
sql_field->pack_flag & FIELDFLAG_BLOB)))
{
Add support for up to VARCHAR (size up to 65535) Renamed HA_VAR_LENGTH to HA_VAR_LENGTH_PART Renamed in all files FIELD_TYPE_STRING and FIELD_TYPE_VAR_STRING to MYSQL_TYPE_STRING and MYSQL_TYPE_VAR_STRING to make it easy to catch all possible errors Added support for VARCHAR KEYS to heap Removed support for ISAM Now only long VARCHAR columns are changed to TEXT on demand (not CHAR) Internal temporary files can now use fixed length tables if the used VARCHAR columns are short
2004-12-06 01:00:37 +01:00
if (column_nr == 0 && (sql_field->pack_flag & FIELDFLAG_BLOB) ||
sql_field->sql_type == MYSQL_TYPE_VARCHAR)
key_info->flags|= HA_BINARY_PACK_KEY | HA_VAR_LENGTH_KEY;
Import changeset
2000-07-31 21:29:14 +02:00
else
key_info->flags|= HA_PACK_KEY;
}
key_length+=length;
key_part_info++;
/* Create the key name based on the first column (if not given) */
if (column_nr == 0)
{
if (key->type == Key::PRIMARY)
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
{
if (primary_key)
{
post-review fixes
2004-11-12 13:34:00 +01:00
my_message(ER_MULTIPLE_PRI_KEY, ER(ER_MULTIPLE_PRI_KEY),
MYF(0));
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
DBUG_RETURN(-1);
}
key_name=primary_key_name;
primary_key=1;
}
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
else if (!(key_name = key->name))
Import changeset
2000-07-31 21:29:14 +02:00
key_name=make_unique_key_name(sql_field->field_name,
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
*key_info_buffer, key_info);
if (check_if_keyname_exists(key_name, *key_info_buffer, key_info))
Import changeset
2000-07-31 21:29:14 +02:00
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_DUP_KEYNAME, MYF(0), key_name);
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(-1);
}
key_info->name=(char*) key_name;
}
}
added checking for empty index name (fixed bug #1419)
2003-10-16 17:33:44 +02:00
if (!key_info->name || check_column_name(key_info->name))
{
word substitution in error messages removed - can create problems for i18n
2003-11-18 10:23:49 +01:00
my_error(ER_WRONG_NAME_FOR_INDEX, MYF(0), key_info->name);
added checking for empty index name (fixed bug #1419)
2003-10-16 17:33:44 +02:00
DBUG_RETURN(-1);
}
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
if (!(key_info->flags & HA_NULL_PART_KEY))
unique_key=1;
Import changeset
2000-07-31 21:29:14 +02:00
key_info->key_length=(uint16) key_length;
Limit created keys to MAX_KEY_LENGTH. Fix problem with query cache and database names
2001-12-23 01:43:46 +01:00
if (key_length > max_key_length && key->type != Key::FULLTEXT)
Import changeset
2000-07-31 21:29:14 +02:00
{
Limit created keys to MAX_KEY_LENGTH. Fix problem with query cache and database names
2001-12-23 01:43:46 +01:00
my_error(ER_TOO_LONG_KEY,MYF(0),max_key_length);
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(-1);
}
few small bug fixes ...
2002-11-15 15:37:44 +01:00
key_info++;
Import changeset
2000-07-31 21:29:14 +02:00
}
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
if (!unique_key && !primary_key &&
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
(file->ha_table_flags() & HA_REQUIRE_PRIMARY_KEY))
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
{
post-review fixes
2004-11-12 13:34:00 +01:00
my_message(ER_REQUIRES_PRIMARY_KEY, ER(ER_REQUIRES_PRIMARY_KEY), MYF(0));
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
DBUG_RETURN(-1);
}
Import changeset
2000-07-31 21:29:14 +02:00
if (auto_increment > 0)
{
post-review fixes
2004-11-12 13:34:00 +01:00
my_message(ER_WRONG_AUTO_KEY, ER(ER_WRONG_AUTO_KEY), MYF(0));
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(-1);
}
Added macros for nice TIMESPEC usage. Fixes for building MySQL with gcc 3.0 Added SIGNED / UNSIGNED casts Fixed core dump bug in net_clear() with libmysqld. Back to using semaphores in query cache. Added 'Null' and 'Index_type' to SHOW INDEX.
2002-01-02 20:29:41 +01:00
/* Sort keys in optimized order */
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
qsort((gptr) *key_info_buffer, *key_count, sizeof(KEY),
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
(qsort_cmp) sort_keys);
After merge fixes
2005-05-14 15:24:36 +02:00
create_info->null_bits= null_fields;
Import changeset
2000-07-31 21:29:14 +02:00
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
DBUG_RETURN(0);
}
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
/*
Set table default charset, if not set
SYNOPSIS
set_table_default_charset()
create_info Table create information
DESCRIPTION
If the table character set was not given explicitely,
let's fetch the database default character set and
apply it to the table.
*/
static void set_table_default_charset(THD *thd,
HA_CREATE_INFO *create_info, char *db)
{
Merge bodhi.local:/opt/local/work/tmp_merge into bodhi.local:/opt/local/work/mysql-5.1-runtime-merge
2006-08-12 19:06:51 +02:00
/*
If the table character set was not given explicitly,
let's fetch the database default character set and
apply it to the table.
*/
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
if (!create_info->default_table_charset)
{
HA_CREATE_INFO db_info;
Merge bodhi.local:/opt/local/work/tmp_merge into bodhi.local:/opt/local/work/mysql-5.1-runtime-merge
2006-08-12 19:06:51 +02:00
load_db_opt_by_name(thd, db, &db_info);
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
create_info->default_table_charset= db_info.default_table_charset;
}
}
Fixes while reviewing new code Added option --count to mysqlshow (to show number of rows) Fixed possible core dump in information schema
2005-05-06 10:39:30 +02:00
/*
Extend long VARCHAR fields to blob & prepare field if it's a blob
SYNOPSIS
prepare_blob_field()
sql_field Field to check
RETURN
0 ok
1 Error (sql_field can't be converted to blob)
In this case the error is given
*/
static bool prepare_blob_field(THD *thd, create_field *sql_field)
{
DBUG_ENTER("prepare_blob_field");
if (sql_field->length > MAX_FIELD_VARCHARLENGTH &&
!(sql_field->flags & BLOB_FLAG))
{
/* Convert long VARCHAR columns to TEXT or BLOB */
char warn_buff[MYSQL_ERRMSG_SIZE];
fix (Bug #8295 and #8296: varchar and varbinary conversion) select, gis & gis-tree tests fails at the moment, but I will push this CS because it was tested before and I'm absolutely sure it's right.
2005-05-12 12:56:04 +02:00
if (sql_field->def || (thd->variables.sql_mode & (MODE_STRICT_TRANS_TABLES |
MODE_STRICT_ALL_TABLES)))
Fixes while reviewing new code Added option --count to mysqlshow (to show number of rows) Fixed possible core dump in information schema
2005-05-06 10:39:30 +02:00
{
my_error(ER_TOO_BIG_FIELDLENGTH, MYF(0), sql_field->field_name,
MAX_FIELD_VARCHARLENGTH / sql_field->charset->mbmaxlen);
DBUG_RETURN(1);
}
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
sql_field->sql_type= MYSQL_TYPE_BLOB;
Fixes while reviewing new code Added option --count to mysqlshow (to show number of rows) Fixed possible core dump in information schema
2005-05-06 10:39:30 +02:00
sql_field->flags|= BLOB_FLAG;
sprintf(warn_buff, ER(ER_AUTO_CONVERT), sql_field->field_name,
fix (Bug #8295 and #8296: varchar and varbinary conversion) select, gis & gis-tree tests fails at the moment, but I will push this CS because it was tested before and I'm absolutely sure it's right.
2005-05-12 12:56:04 +02:00
(sql_field->charset == &my_charset_bin) ? "VARBINARY" : "VARCHAR",
Fixes while reviewing new code Added option --count to mysqlshow (to show number of rows) Fixed possible core dump in information schema
2005-05-06 10:39:30 +02:00
(sql_field->charset == &my_charset_bin) ? "BLOB" : "TEXT");
push_warning(thd, MYSQL_ERROR::WARN_LEVEL_NOTE, ER_AUTO_CONVERT,
warn_buff);
}
if ((sql_field->flags & BLOB_FLAG) && sql_field->length)
{
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
if (sql_field->sql_type == MYSQL_TYPE_BLOB)
Fixes while reviewing new code Added option --count to mysqlshow (to show number of rows) Fixed possible core dump in information schema
2005-05-06 10:39:30 +02:00
{
/* The user has given a length to the blob column */
sql_field->sql_type= get_blob_type_from_length(sql_field->length);
sql_field->pack_length= calc_pack_length(sql_field->sql_type, 0);
}
sql_field->length= 0;
}
DBUG_RETURN(0);
}
Bug#9102 - Stored proccedures: function which returns blob causes crash Initialization of fields for sp return type was not complete.
2005-04-19 10:09:25 +02:00
/*
Preparation of create_field for SP function return values.
Based on code used in the inner loop of mysql_prepare_table() above
SYNOPSIS
sp_prepare_create_field()
thd Thread object
sql_field Field to prepare
DESCRIPTION
Prepares the field structures for field creation.
*/
void sp_prepare_create_field(THD *thd, create_field *sql_field)
{
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
if (sql_field->sql_type == MYSQL_TYPE_SET ||
sql_field->sql_type == MYSQL_TYPE_ENUM)
Bug#9102 - Stored proccedures: function which returns blob causes crash Initialization of fields for sp return type was not complete.
2005-04-19 10:09:25 +02:00
{
uint32 field_length, dummy;
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
if (sql_field->sql_type == MYSQL_TYPE_SET)
Bug#9102 - Stored proccedures: function which returns blob causes crash Initialization of fields for sp return type was not complete.
2005-04-19 10:09:25 +02:00
{
calculate_interval_lengths(sql_field->charset,
sql_field->interval, &dummy,
&field_length);
sql_field->length= field_length +
(sql_field->interval->count - 1);
}
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
else /* MYSQL_TYPE_ENUM */
Bug#9102 - Stored proccedures: function which returns blob causes crash Initialization of fields for sp return type was not complete.
2005-04-19 10:09:25 +02:00
{
calculate_interval_lengths(sql_field->charset,
sql_field->interval,
&field_length, &dummy);
sql_field->length= field_length;
}
set_if_smaller(sql_field->length, MAX_FIELD_WIDTH-1);
}
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
if (sql_field->sql_type == MYSQL_TYPE_BIT)
Bug#9102 - Stored proccedures: function which returns blob causes crash Initialization of fields for sp return type was not complete.
2005-04-19 10:09:25 +02:00
{
sql_field->pack_flag= FIELDFLAG_NUMBER |
FIELDFLAG_TREAT_BIT_AS_CHAR;
}
sql_field->create_length_to_internal_length();
Fixes while reviewing new code Added option --count to mysqlshow (to show number of rows) Fixed possible core dump in information schema
2005-05-06 10:39:30 +02:00
DBUG_ASSERT(sql_field->def == 0);
/* Can't go wrong as sql_field->def is not defined */
(void) prepare_blob_field(thd, sql_field);
}
Bug#9102 - Stored proccedures: function which returns blob causes crash Initialization of fields for sp return type was not complete.
2005-04-19 10:09:25 +02:00
BUG#19309: Crash if double procedural alter
2006-06-15 20:03:17 +02:00
/*
Copy HA_CREATE_INFO struct
SYNOPSIS
copy_create_info()
lex_create_info The create_info struct setup by parser
RETURN VALUES
> 0 A pointer to a copy of the lex_create_info
0 Memory allocation error
DESCRIPTION
Allocate memory for copy of HA_CREATE_INFO structure from parser
to ensure we can reuse the parser struct in stored procedures
and prepared statements.
*/
static HA_CREATE_INFO *copy_create_info(HA_CREATE_INFO *lex_create_info)
{
HA_CREATE_INFO *create_info;
if (!(create_info= (HA_CREATE_INFO*)sql_alloc(sizeof(HA_CREATE_INFO))))
mem_alloc_error(sizeof(HA_CREATE_INFO));
else
memcpy((void*)create_info, (void*)lex_create_info, sizeof(HA_CREATE_INFO));
return create_info;
}
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
/*
Create a table
SYNOPSIS
renamedb.test, renamedb.result: new file Many files: WL#757 RENAME DATABASE
2006-02-13 08:49:28 +01:00
mysql_create_table_internal()
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
thd Thread object
db Database
table_name Table name
BUG#19309: Problem with calling proecdures twice Need to flag when a copy is needed to not overwrite a create_info object connected to the lex structure
2006-06-21 16:57:30 +02:00
lex_create_info Create information (like MAX_ROWS)
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
fields List of fields to create
keys List of keys to create
query_id and my_xid -> ulonglong fix for binlog+autocommit+tclog comments, style fixes
2005-01-27 22:38:56 +01:00
internal_tmp_table Set to 1 if this is an internal temporary table
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
(From ALTER TABLE)
BUG#19309: Problem with calling proecdures twice Need to flag when a copy is needed to not overwrite a create_info object connected to the lex structure
2006-06-21 16:57:30 +02:00
select_field_count
use_copy_create_info Should we make a copy of create info (we do this
when this is called from sql_parse.cc where we
want to ensure lex object isn't manipulated.
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
DESCRIPTION
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
If one creates a temporary table, this is automatically opened
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
no_log is needed for the case of CREATE ... SELECT,
as the logging will be done later in sql_insert.cc
select_field_count is also used for CREATE ... SELECT,
and must be zero for standard create of table.
RETURN VALUES
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
FALSE OK
TRUE error
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
*/
renamedb.test, renamedb.result: new file Many files: WL#757 RENAME DATABASE
2006-02-13 08:49:28 +01:00
bool mysql_create_table_internal(THD *thd,
const char *db, const char *table_name,
BUG#19309: Crash if double procedural alter
2006-06-15 20:03:17 +02:00
HA_CREATE_INFO *lex_create_info,
renamedb.test, renamedb.result: new file Many files: WL#757 RENAME DATABASE
2006-02-13 08:49:28 +01:00
List<create_field> &fields,
List<Key> &keys,bool internal_tmp_table,
BUG#19309: Problem with calling proecdures twice Need to flag when a copy is needed to not overwrite a create_info object connected to the lex structure
2006-06-21 16:57:30 +02:00
uint select_field_count,
bool use_copy_create_info)
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
{
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
char path[FN_REFLEN];
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
uint path_length;
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
const char *alias;
uint db_options, key_count;
KEY *key_info_buffer;
BUG#19309: Crash if double procedural alter
2006-06-15 20:03:17 +02:00
HA_CREATE_INFO *create_info;
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
handler *file;
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool error= TRUE;
renamedb.test, renamedb.result: new file Many files: WL#757 RENAME DATABASE
2006-02-13 08:49:28 +01:00
DBUG_ENTER("mysql_create_table_internal");
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
DBUG_PRINT("enter", ("db: '%s' table: '%s' tmp: %d",
db, table_name, internal_tmp_table));
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
BUG#19309: Problem with calling proecdures twice Need to flag when a copy is needed to not overwrite a create_info object connected to the lex structure
2006-06-21 16:57:30 +02:00
if (use_copy_create_info)
BUG#19309: Crash if double procedural alter
2006-06-15 20:03:17 +02:00
{
BUG#19309: Problem with calling proecdures twice Need to flag when a copy is needed to not overwrite a create_info object connected to the lex structure
2006-06-21 16:57:30 +02:00
if (!(create_info= copy_create_info(lex_create_info)))
{
DBUG_RETURN(TRUE);
}
BUG#19309: Crash if double procedural alter
2006-06-15 20:03:17 +02:00
}
BUG#19309: Problem with calling proecdures twice Need to flag when a copy is needed to not overwrite a create_info object connected to the lex structure
2006-06-21 16:57:30 +02:00
else
create_info= lex_create_info;
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
/* Check for duplicate fields and check type of table to create */
if (!fields.elements)
{
post-review fixes
2004-11-12 13:34:00 +01:00
my_message(ER_TABLE_MUST_HAVE_COLUMNS, ER(ER_TABLE_MUST_HAVE_COLUMNS),
MYF(0));
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
}
Bug #17210 Create temp table call to ha_ndbcluster::create_handler_files caused core - do not try to create ndb temp tables even if default storage is ndb
2006-02-17 17:12:35 +01:00
if (check_engine(thd, table_name, create_info))
Bug#6877 MySQL should give an error if the requested table type is not available Implement new SQL mode - NO_ENGINE_SUBSTITUTION
2005-06-17 23:14:44 +02:00
DBUG_RETURN(TRUE);
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
db_options= create_info->table_options;
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
if (create_info->row_type == ROW_TYPE_DYNAMIC)
db_options|=HA_OPTION_PACK_RECORD;
alias= table_case_name(create_info, table_name);
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
if (!(file= get_new_handler((TABLE_SHARE*) 0, thd->mem_root,
create_info->db_type)))
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
{
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
mem_alloc_error(sizeof(handler));
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
DBUG_RETURN(TRUE);
}
Make storage engines "pluggable", handlerton work
2005-11-07 16:25:06 +01:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
bug #17290 (sp with delete, create and rollback causes MySQLD core) additional fixes
2006-03-20 14:30:01 +01:00
partition_info *part_info= thd->work_part_info;
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
if (!part_info && create_info->db_type->partition_flags &&
(create_info->db_type->partition_flags() & HA_USE_AUTO_PARTITION))
{
/*
Table is not defined as a partitioned table but the engine handles
all tables as partitioned. The handler will set up the partition info
object with the default settings.
*/
bug 17290 and bug 14350 added THD::work_part_info member where we now store modified partition_info structure. It allows no solve problem when different parts of the part_info get into different mem_roots
2006-03-18 15:48:21 +01:00
thd->work_part_info= part_info= new partition_info();
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
if (!part_info)
{
mem_alloc_error(sizeof(partition_info));
DBUG_RETURN(TRUE);
}
file->set_auto_partitions(part_info);
BUG#17754 Added new syntax ALTER TABLE t1 REMOVE PARTITIONING, changed semantics of ALTER TABLE t1 ENGINE=X; to not remove partitioning Fix a number of mix engine bugs in partitioning
2006-03-20 20:36:21 +01:00
part_info->default_engine_type= create_info->db_type;
BUG#19010: Fix issues with that ALTER TABLE from auto-partitioned NDB table doesn't work unless primary key exists on table.
2006-05-10 18:53:40 +02:00
part_info->is_auto_partitioned= TRUE;
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
}
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
if (part_info)
{
/*
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
The table has been specified as a partitioned table.
If this is part of an ALTER TABLE the handler will be the partition
handler but we need to specify the default handler to use for
partitions also in the call to check_partition_info. We transport
this information in the default_db_type variable, it is either
DB_TYPE_DEFAULT or the engine set in the ALTER TABLE command.
Check that we don't use foreign keys in the table since it won't
work even with InnoDB beneath it.
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
*/
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
List_iterator<Key> key_iterator(keys);
Key *key;
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
handlerton *part_engine_type= create_info->db_type;
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
char *part_syntax_buf;
uint syntax_len;
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
handlerton *engine_type;
Bug #17497: Partitions: crash if add partition on temporary table Temporary tables are no longer allowed to be partitioned.
2006-03-08 18:48:40 +01:00
if (create_info->options & HA_LEX_CREATE_TMP_TABLE)
{
my_error(ER_PARTITION_NO_TEMPORARY, MYF(0));
goto err;
}
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
while ((key= key_iterator++))
{
Bug #18483 Cannot create table with FK constraint - remove error set for auto partitioned tables
2006-06-01 14:51:58 +02:00
if (key->type == Key::FOREIGN_KEY &&
!part_info->is_auto_partitioned)
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
{
my_error(ER_CANNOT_ADD_FOREIGN, MYF(0));
goto err;
}
}
This changes the order of the universe, black is now the new white. In practice this means that handlerton is now created by the server and is passed to the engine. Plugin startups can now also control how plugins are inited (and can optionally pass values). Bit more flexibility to those who want to write plugin interfaces to the database.
2006-09-15 19:28:00 +02:00
if ((part_engine_type == partition_hton) &&
bug #17290 (sp with delete, create and rollback causes MySQLD core) additional fixes
2006-03-20 14:30:01 +01:00
part_info->default_engine_type)
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
{
/*
This only happens at ALTER TABLE.
default_engine_type was assigned from the engine set in the ALTER
TABLE command.
*/
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
;
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
}
Handle default engine type better for partitioned tables
2005-09-20 16:29:59 +02:00
else
{
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
if (create_info->used_fields & HA_CREATE_USED_ENGINE)
{
part_info->default_engine_type= create_info->db_type;
}
else
{
if (part_info->default_engine_type == NULL)
{
part_info->default_engine_type= ha_checktype(thd,
DB_TYPE_DEFAULT, 0, 0);
}
}
Handle default engine type better for partitioned tables
2005-09-20 16:29:59 +02:00
}
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
DBUG_PRINT("info", ("db_type = %d",
ha_legacy_type(part_info->default_engine_type)));
BUG#16002: Make partition functions that are unsigned work properly
2006-04-18 04:51:34 +02:00
if (part_info->check_partition_info(thd, &engine_type, file,
Merge mronstrom@bk-internal.mysql.com:/home/bk/bugs/bug18198 into dator5.(none):/home/pappa/bug18198
2006-07-12 17:33:22 +02:00
create_info, TRUE))
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2005-11-05 00:32:55 +01:00
goto err;
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
part_info->default_engine_type= engine_type;
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2005-11-05 00:32:55 +01:00
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
/*
We reverse the partitioning parser and generate a standard format
for syntax stored in frm file.
*/
if (!(part_syntax_buf= generate_partition_syntax(part_info,
&syntax_len,
Merge c-0409e253.1238-1-64736c10.cust.bredbandsbolaget.se:/home/pappa/clean-mysql-5.1 into c-0409e253.1238-1-64736c10.cust.bredbandsbolaget.se:/home/pappa/bug19695
2006-06-12 20:42:07 +02:00
TRUE, TRUE)))
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2005-11-05 00:32:55 +01:00
goto err;
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
part_info->part_info_string= part_syntax_buf;
part_info->part_info_len= syntax_len;
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
if ((!(engine_type->partition_flags &&
engine_type->partition_flags() & HA_CAN_PARTITION)) ||
This changes the order of the universe, black is now the new white. In practice this means that handlerton is now created by the server and is passed to the engine. Plugin startups can now also control how plugins are inited (and can optionally pass values). Bit more flexibility to those who want to write plugin interfaces to the database.
2006-09-15 19:28:00 +02:00
create_info->db_type == partition_hton)
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
{
/*
The handler assigned to the table cannot handle partitioning.
Assign the partition handler as the handler of the table.
*/
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
DBUG_PRINT("info", ("db_type: %d",
ha_legacy_type(create_info->db_type)));
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
delete file;
This changes the order of the universe, black is now the new white. In practice this means that handlerton is now created by the server and is passed to the engine. Plugin startups can now also control how plugins are inited (and can optionally pass values). Bit more flexibility to those who want to write plugin interfaces to the database.
2006-09-15 19:28:00 +02:00
create_info->db_type= partition_hton;
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
if (!(file= get_ha_partition(part_info)))
{
DBUG_RETURN(TRUE);
}
BUG #16806: ALTER TABLE fails when creating new field not last in table Moved some code to else path to avoid delete, create, delete, create scenarios. Fixed up the partition info object for some cases where we move from default partitioning in NDB to default partitioning using partitioning.
2006-03-23 22:00:58 +01:00
/*
If we have default number of partitions or subpartitions we
might require to set-up the part_info object such that it
creates a proper .par file. The current part_info object is
only used to create the frm-file and .par-file.
*/
if (part_info->use_default_no_partitions &&
part_info->no_parts &&
partition functions to pass create_info, not only max_rows
2006-06-27 22:19:27 +02:00
(int)part_info->no_parts !=
file->get_default_no_partitions(create_info))
BUG #16806: ALTER TABLE fails when creating new field not last in table Moved some code to else path to avoid delete, create, delete, create scenarios. Fixed up the partition info object for some cases where we move from default partitioning in NDB to default partitioning using partitioning.
2006-03-23 22:00:58 +01:00
{
BUG#16806: ALTER TABLE with new field in before position fails Review fixes
2006-03-25 00:21:43 +01:00
uint i;
BUG #16806: ALTER TABLE fails when creating new field not last in table Moved some code to else path to avoid delete, create, delete, create scenarios. Fixed up the partition info object for some cases where we move from default partitioning in NDB to default partitioning using partitioning.
2006-03-23 22:00:58 +01:00
List_iterator<partition_element> part_it(part_info->partitions);
BUG#16806: ALTER TABLE with new field in before position fails Review fixes
2006-03-25 00:21:43 +01:00
part_it++;
DBUG_ASSERT(thd->lex->sql_command != SQLCOM_CREATE_TABLE);
for (i= 1; i < part_info->partitions.elements; i++)
(part_it++)->part_state= PART_TO_BE_DROPPED;
BUG #16806: ALTER TABLE fails when creating new field not last in table Moved some code to else path to avoid delete, create, delete, create scenarios. Fixed up the partition info object for some cases where we move from default partitioning in NDB to default partitioning using partitioning.
2006-03-23 22:00:58 +01:00
}
else if (part_info->is_sub_partitioned() &&
part_info->use_default_no_subpartitions &&
part_info->no_subparts &&
BUG#16806: ALTER TABLE with new field in before position fails Review fixes
2006-03-25 00:21:43 +01:00
(int)part_info->no_subparts !=
partition functions to pass create_info, not only max_rows
2006-06-27 22:19:27 +02:00
file->get_default_no_partitions(create_info))
BUG #16806: ALTER TABLE fails when creating new field not last in table Moved some code to else path to avoid delete, create, delete, create scenarios. Fixed up the partition info object for some cases where we move from default partitioning in NDB to default partitioning using partitioning.
2006-03-23 22:00:58 +01:00
{
BUG#16806: ALTER TABLE with new field in before position fails Review fixes
2006-03-25 00:21:43 +01:00
DBUG_ASSERT(thd->lex->sql_command != SQLCOM_CREATE_TABLE);
partition functions to pass create_info, not only max_rows
2006-06-27 22:19:27 +02:00
part_info->no_subparts= file->get_default_no_partitions(create_info);
BUG #16806: ALTER TABLE fails when creating new field not last in table Moved some code to else path to avoid delete, create, delete, create scenarios. Fixed up the partition info object for some cases where we move from default partitioning in NDB to default partitioning using partitioning.
2006-03-23 22:00:58 +01:00
}
}
else if (create_info->db_type != engine_type)
{
BUG#19140: Create index caused crash
2006-05-09 03:41:10 +02:00
/*
We come here when we don't use a partitioned handler.
Since we use a partitioned table it must be "native partitioned".
We have switched engine from defaults, most likely only specified
engines in partition clauses.
*/
BUG #16806: ALTER TABLE fails when creating new field not last in table Moved some code to else path to avoid delete, create, delete, create scenarios. Fixed up the partition info object for some cases where we move from default partitioning in NDB to default partitioning using partitioning.
2006-03-23 22:00:58 +01:00
delete file;
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
if (!(file= get_new_handler((TABLE_SHARE*) 0, thd->mem_root,
engine_type)))
BUG #16806: ALTER TABLE fails when creating new field not last in table Moved some code to else path to avoid delete, create, delete, create scenarios. Fixed up the partition info object for some cases where we move from default partitioning in NDB to default partitioning using partitioning.
2006-03-23 22:00:58 +01:00
{
mem_alloc_error(sizeof(handler));
DBUG_RETURN(TRUE);
}
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
}
}
#endif
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
set_table_default_charset(thd, create_info, (char*) db);
WL#807 Optimize loading database options in load_db_opt also known as BUG#2326 Charset of table is determined by charset of db only if "USE db;"
2004-07-08 12:03:01 +02:00
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
if (mysql_prepare_table(thd, create_info, &fields,
&keys, internal_tmp_table, &db_options, file,
&key_info_buffer, &key_count,
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
select_field_count))
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
goto err;
Import changeset
2000-07-31 21:29:14 +02:00
/* Check if table exists */
if (create_info->options & HA_LEX_CREATE_TMP_TABLE)
{
Bug#15904 --tmpdir requires trailing slash '/' - Use same code for creating the temptable filename both from "mysql_create_table" and "mysql_create_like_table"
2006-01-23 13:28:42 +01:00
path_length= build_tmptable_filename(thd, path, sizeof(path));
Fix for sporadically failures of innodb.test on Darwin 7.3 which occured because we were not lowering case of file names for temporary tables altough handler assumes so if lower_case_table_names==2. Now we are lowering case for them.
2004-05-25 00:30:09 +02:00
if (lower_case_table_names)
Restore creation of files for temporary tables in the tmpdir, which was broken by an earlier bug fix. (Bug #11440)
2005-06-21 20:25:51 +02:00
my_casedn_str(files_charset_info, path);
Import changeset
2000-07-31 21:29:14 +02:00
create_info->table_options|=HA_CREATE_DELAY_KEY_WRITE;
}
Bug#20616: drop_temp_table test fails on Windows platform
2006-06-23 18:15:39 +02:00
else
{
Merge zippy.(none):/home/cmiller/work/mysql/merge/mysql-5.0 into zippy.(none):/home/cmiller/work/mysql/merge/mysql-5.1
2006-07-03 17:35:58 +02:00
#ifdef FN_DEVCHAR
/* check if the table name contains FN_DEVCHAR when defined */
const char *start= alias;
while (*start != '\0')
{
if (*start == FN_DEVCHAR)
{
my_error(ER_WRONG_TABLE_NAME, MYF(0), alias);
DBUG_RETURN(TRUE);
}
start++;
}
#endif
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
path_length= build_table_filename(path, sizeof(path), db, alias, reg_ext,
internal_tmp_table ? FN_IS_TMP : 0);
Bug#20616: drop_temp_table test fails on Windows platform
2006-06-23 18:15:39 +02:00
}
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
Import changeset
2000-07-31 21:29:14 +02:00
/* Check if table already exists */
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
if ((create_info->options & HA_LEX_CREATE_TMP_TABLE) &&
find_temporary_table(thd, db, table_name))
Import changeset
2000-07-31 21:29:14 +02:00
{
Automatic repair of MyISAM tables + small bug fixes
2000-10-16 23:47:15 +02:00
if (create_info->options & HA_LEX_CREATE_IF_NOT_EXISTS)
Better fix for CREATE TABLE IF NOT EXISTS ... SELECT Fixed chsize() problem on windows Extend default timeout on windows clients to 1 year (to avoid timeout problems)
2003-10-15 20:41:13 +02:00
{
create_info->table_existed= 1; // Mark that table existed
Fix for bug #6008: MySQL does not create warnings when creating database and using IF NOT EXISTS produce warning for 'create database if not exists' if database exists do not update database options in this case produce warning for 'create table if not exists' if table exists
2005-09-12 14:09:19 +02:00
push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_NOTE,
ER_TABLE_EXISTS_ERROR, ER(ER_TABLE_EXISTS_ERROR),
alias);
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2005-11-05 00:32:55 +01:00
error= 0;
goto err;
Better fix for CREATE TABLE IF NOT EXISTS ... SELECT Fixed chsize() problem on windows Extend default timeout on windows clients to 1 year (to avoid timeout problems)
2003-10-15 20:41:13 +02:00
}
Merge with 4.0.18
2004-02-11 00:06:46 +01:00
my_error(ER_TABLE_EXISTS_ERROR, MYF(0), alias);
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
goto err;
Import changeset
2000-07-31 21:29:14 +02:00
}
Merge mysql.com:/home/mydev/mysql-5.0-tmp_merge into mysql.com:/home/mydev/mysql-5.1-amerge
2006-07-04 10:02:11 +02:00
Import changeset
2000-07-31 21:29:14 +02:00
VOID(pthread_mutex_lock(&LOCK_open));
query_id and my_xid -> ulonglong fix for binlog+autocommit+tclog comments, style fixes
2005-01-27 22:38:56 +01:00
if (!internal_tmp_table && !(create_info->options & HA_LEX_CREATE_TMP_TABLE))
Import changeset
2000-07-31 21:29:14 +02:00
{
if (!access(path,F_OK))
{
if (create_info->options & HA_LEX_CREATE_IF_NOT_EXISTS)
Fix for bug #6008: MySQL does not create warnings when creating database and using IF NOT EXISTS produce warning for 'create database if not exists' if database exists do not update database options in this case produce warning for 'create table if not exists' if table exists
2005-09-12 14:09:19 +02:00
goto warn;
my_error(ER_TABLE_EXISTS_ERROR,MYF(0),table_name);
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2005-11-05 00:32:55 +01:00
goto unlock_and_end;
Import changeset
2000-07-31 21:29:14 +02:00
}
Fix for bug#16532 mysql server assert in debug if table definition is removed Emit an error instead of doing ASSERT
2006-07-25 14:52:24 +02:00
/*
We don't assert here, but check the result, because the table could be
in the table definition cache and in the same time the .frm could be
missing from the disk, in case of manual intervention which deletes
the .frm file. The user has to use FLUSH TABLES; to clear the cache.
Then she could create the table. This case is pretty obscure and
therefore we don't introduce a new error message only for it.
*/
if (get_cached_table_share(db, alias))
{
my_error(ER_TABLE_EXISTS_ERROR, MYF(0), table_name);
goto unlock_and_end;
}
Import changeset
2000-07-31 21:29:14 +02:00
}
Added NDB storage engine
2004-04-15 09:14:14 +02:00
/*
Check that table with given name does not already
exist in any storage engine. In such a case it should
be discovered and the error ER_TABLE_EXISTS_ERROR be returned
unless user specified CREATE TABLE IF EXISTS
The LOCK_open mutex has been locked to make sure no
one else is attempting to discover the table. Since
it's not on disk as a frm file, no one could be using it!
*/
if (!(create_info->options & HA_LEX_CREATE_TMP_TABLE))
{
bool create_if_not_exists =
create_info->options & HA_LEX_CREATE_IF_NOT_EXISTS;
BUG#10365 Cluster handler uses non-standard error codes - Added better error messages when trying to open a table that can't be discovered or unpacked. The most likely cause of this is that it does not have any frm data, probably since it has been created from NdbApi or is a NDB system table. - Separated functionality that was in ha_create_table_from_engine into two functions. One that checks if the table exists and another one that tries to create the table from the engine.
2005-06-08 13:31:59 +02:00
if (ha_table_exists_in_engine(thd, db, table_name))
Added NDB storage engine
2004-04-15 09:14:14 +02:00
{
BUG#10365 Cluster handler uses non-standard error codes - Added better error messages when trying to open a table that can't be discovered or unpacked. The most likely cause of this is that it does not have any frm data, probably since it has been created from NdbApi or is a NDB system table. - Separated functionality that was in ha_create_table_from_engine into two functions. One that checks if the table exists and another one that tries to create the table from the engine.
2005-06-08 13:31:59 +02:00
DBUG_PRINT("info", ("Table with same name already existed in handler"));
Added NDB storage engine
2004-04-15 09:14:14 +02:00
if (create_if_not_exists)
Fix for bug #6008: MySQL does not create warnings when creating database and using IF NOT EXISTS produce warning for 'create database if not exists' if database exists do not update database options in this case produce warning for 'create table if not exists' if table exists
2005-09-12 14:09:19 +02:00
goto warn;
my_error(ER_TABLE_EXISTS_ERROR,MYF(0),table_name);
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2005-11-05 00:32:55 +01:00
goto unlock_and_end;
Import changeset
2000-07-31 21:29:14 +02:00
}
}
thd->proc_info="creating table";
Better fix for CREATE TABLE IF NOT EXISTS ... SELECT Fixed chsize() problem on windows Extend default timeout on windows clients to 1 year (to avoid timeout problems)
2003-10-15 20:41:13 +02:00
create_info->table_existed= 0; // Mark that table is created
Import changeset
2000-07-31 21:29:14 +02:00
After merge fixes. Note that mix_innodb_myisam_binlog and union fails after this patch (Will be fixed shortly by maintaners of this code)
2003-09-11 18:06:23 +02:00
if (thd->variables.sql_mode & MODE_NO_DIR_IN_CREATE)
show symlinked directories in SHOW CREATE TABLE
2003-08-23 11:25:39 +02:00
create_info->data_file_name= create_info->index_file_name= 0;
Import changeset
2000-07-31 21:29:14 +02:00
create_info->table_options=db_options;
show symlinked directories in SHOW CREATE TABLE
2003-08-23 11:25:39 +02:00
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
path[path_length - reg_ext_length]= '\0'; // Remove .frm extension
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
if (rea_create_table(thd, path, db, table_name, create_info, fields,
key_count, key_info_buffer, file))
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2005-11-05 00:32:55 +01:00
goto unlock_and_end;
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
if (create_info->options & HA_LEX_CREATE_TMP_TABLE)
{
/* Open table and put in temporary table list */
if (!(open_temporary_table(thd, path, db, table_name, 1)))
{
(void) rm_temporary_table(create_info->db_type, path);
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2005-11-05 00:32:55 +01:00
goto unlock_and_end;
Import changeset
2000-07-31 21:29:14 +02:00
}
Move tmp_table_used to THD Optimize depending sub querys Remove valgrind warnings
2003-02-10 16:59:16 +01:00
thd->tmp_table_used= 1;
Import changeset
2000-07-31 21:29:14 +02:00
}
Merge with 4.1
2004-12-06 10:38:56 +01:00
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
/*
Don't write statement if:
- It is an internal temporary table,
- Row-based logging is used and it we are creating a temporary table, or
- The binary log is not open.
Bug#15920 (Temporary tables are not logged in binlog): Fixing faulty tests preventing some CREATE TEMPORARY TABLE statements from being binlogged under statement-based replication.
2005-12-22 11:09:50 +01:00
Otherwise, the statement shall be binlogged.
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
*/
if (!internal_tmp_table &&
WL#2977 and WL#2712 global and session-level variable to set the binlog format (row/statement), and new binlog format called "mixed" (which is statement-based except if only row-based is correct, in this cset it means if UDF or UUID is used; more cases could be added in later 5.1 release): SET GLOBAL|SESSION BINLOG_FORMAT=row|statement|mixed|default; the global default is statement unless cluster is enabled (then it's row) as in 5.1-alpha. It's not possible to use SET on this variable if a session is currently in row-based mode and has open temporary tables (because CREATE TEMPORARY TABLE was not binlogged so temp table is not known on slave), or if NDB is enabled (because NDB does not support such change on-the-fly, though it will later), of if in a stored function (see below). The added tests test the possibility or impossibility to SET, their effects, and the mixed mode, including in prepared statements and in stored procedures and functions. Caveats: a) The mixed mode will not work for stored functions: in mixed mode, a stored function will always be binlogged as one call and in a statement-based way (e.g. INSERT VALUES(myfunc()) or SELECT myfunc()). b) for the same reason, changing the thread's binlog format inside a stored function is refused with an error message. c) the same problems apply to triggers; implementing b) for triggers will be done later (will ask Dmitri). Additionally, as the binlog format is now changeable by each user for his session, I remove the implication which was done at startup, where row-based automatically set log-bin-trust-routine-creators to 1 (not possible anymore as a user can now switch to stmt-based and do nasty things again), and automatically set --innodb-locks-unsafe-for-binlog to 1 (was anyway theoretically incorrect as it disabled phantom protection). Plus fixes for compiler warnings.
2006-02-25 22:21:03 +01:00
(!thd->current_stmt_binlog_row_based ||
(thd->current_stmt_binlog_row_based &&
Bug#15920 (Temporary tables are not logged in binlog): Fixing faulty tests preventing some CREATE TEMPORARY TABLE statements from being binlogged under statement-based replication.
2005-12-22 11:09:50 +01:00
!(create_info->options & HA_LEX_CREATE_TMP_TABLE))))
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
write_bin_log(thd, TRUE, thd->query, thd->query_length);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
error= FALSE;
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2005-11-05 00:32:55 +01:00
unlock_and_end:
Import changeset
2000-07-31 21:29:14 +02:00
VOID(pthread_mutex_unlock(&LOCK_open));
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2005-11-05 00:32:55 +01:00
err:
Import changeset
2000-07-31 21:29:14 +02:00
thd->proc_info="After create";
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2005-11-05 00:32:55 +01:00
delete file;
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(error);
Fix for bug #6008: MySQL does not create warnings when creating database and using IF NOT EXISTS produce warning for 'create database if not exists' if database exists do not update database options in this case produce warning for 'create table if not exists' if table exists
2005-09-12 14:09:19 +02:00
warn:
Simple changes during review of code Added back flag that I accidently removed in last patch
2005-10-07 23:57:40 +02:00
error= FALSE;
Fix for bug #6008: MySQL does not create warnings when creating database and using IF NOT EXISTS produce warning for 'create database if not exists' if database exists do not update database options in this case produce warning for 'create table if not exists' if table exists
2005-09-12 14:09:19 +02:00
push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_NOTE,
ER_TABLE_EXISTS_ERROR, ER(ER_TABLE_EXISTS_ERROR),
alias);
create_info->table_existed= 1; // Mark that table existed
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2005-11-05 00:32:55 +01:00
goto unlock_and_end;
Import changeset
2000-07-31 21:29:14 +02:00
}
renamedb.test, renamedb.result: new file Many files: WL#757 RENAME DATABASE
2006-02-13 08:49:28 +01:00
/*
Database locking aware wrapper for mysql_create_table_internal(),
*/
bool mysql_create_table(THD *thd, const char *db, const char *table_name,
HA_CREATE_INFO *create_info,
List<create_field> &fields,
List<Key> &keys,bool internal_tmp_table,
BUG#19309: Problem with calling proecdures twice Need to flag when a copy is needed to not overwrite a create_info object connected to the lex structure
2006-06-21 16:57:30 +02:00
uint select_field_count,
bool use_copy_create_info)
renamedb.test, renamedb.result: new file Many files: WL#757 RENAME DATABASE
2006-02-13 08:49:28 +01:00
{
bool result;
DBUG_ENTER("mysql_create_table");
/* Wait for any database locks */
pthread_mutex_lock(&LOCK_lock_db);
while (!thd->killed &&
hash_search(&lock_db_cache,(byte*) db, strlen(db)))
{
wait_for_condition(thd, &LOCK_lock_db, &COND_refresh);
pthread_mutex_lock(&LOCK_lock_db);
}
if (thd->killed)
{
pthread_mutex_unlock(&LOCK_lock_db);
DBUG_RETURN(TRUE);
}
creating_table++;
pthread_mutex_unlock(&LOCK_lock_db);
result= mysql_create_table_internal(thd, db, table_name, create_info,
fields, keys, internal_tmp_table,
BUG#19309: Problem with calling proecdures twice Need to flag when a copy is needed to not overwrite a create_info object connected to the lex structure
2006-06-21 16:57:30 +02:00
select_field_count,
use_copy_create_info);
renamedb.test, renamedb.result: new file Many files: WL#757 RENAME DATABASE
2006-02-13 08:49:28 +01:00
pthread_mutex_lock(&LOCK_lock_db);
if (!--creating_table && creating_database)
pthread_cond_signal(&COND_refresh);
pthread_mutex_unlock(&LOCK_lock_db);
DBUG_RETURN(result);
}
Import changeset
2000-07-31 21:29:14 +02:00
/*
** Give the key name after the first field with an optional '_#' after
**/
static bool
check_if_keyname_exists(const char *name, KEY *start, KEY *end)
{
for (KEY *key=start ; key != end ; key++)
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
if (!my_strcasecmp(system_charset_info,name,key->name))
Import changeset
2000-07-31 21:29:14 +02:00
return 1;
return 0;
}
static char *
make_unique_key_name(const char *field_name,KEY *start,KEY *end)
{
char buff[MAX_FIELD_NAME],*buff_end;
fixed bug #856 'Naming a key "Primary" causes trouble'
2004-03-13 22:31:30 +01:00
if (!check_if_keyname_exists(field_name,start,end) &&
my_strcasecmp(system_charset_info,field_name,primary_key_name))
Import changeset
2000-07-31 21:29:14 +02:00
return (char*) field_name; // Use fieldname
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
buff_end=strmake(buff,field_name, sizeof(buff)-4);
/*
Only 3 chars + '\0' left, so need to limit to 2 digit
This is ok as we can't have more than 100 keys anyway
*/
Worklog#1563 - Support of on-line CREATE/DROP INDEX. Corrected minor problems of the preceding changeset 1.1705.
2004-04-07 13:12:05 +02:00
for (uint i=2 ; i< 100; i++)
Import changeset
2000-07-31 21:29:14 +02:00
{
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
*buff_end= '_';
int10_to_str(i, buff_end+1, 10);
Import changeset
2000-07-31 21:29:14 +02:00
if (!check_if_keyname_exists(buff,start,end))
return sql_strdup(buff);
}
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
return (char*) "not_specified"; // Should never happen
Import changeset
2000-07-31 21:29:14 +02:00
}
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
Import changeset
2000-07-31 21:29:14 +02:00
/****************************************************************************
** Alter a table definition
****************************************************************************/
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
/*
Rename a table.
SYNOPSIS
mysql_rename_table()
base The handlerton handle.
old_db The old database name.
old_name The old table name.
new_db The new database name.
new_name The new table name.
flags flags for build_table_filename().
FN_FROM_IS_TMP old_name is temporary.
FN_TO_IS_TMP new_name is temporary.
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
NO_FRM_RENAME Don't rename the FRM file
but only the table in the storage engine.
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
RETURN
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
FALSE OK
TRUE Error
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
*/
RENAME TABLE table_name TO new_table_name ; Faster Alloc Small bug fixes
2000-08-21 02:00:52 +02:00
bool
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
mysql_rename_table(handlerton *base, const char *old_db,
const char *old_name, const char *new_db,
const char *new_name, uint flags)
Import changeset
2000-07-31 21:29:14 +02:00
{
A fix and a test case for Bug#14210 "Simple query with > operator on large table gives server crash": make sure that when a MyISAM temporary table is created for a cursor, it's created in its memory root, not the memory root of the current query.
2005-11-03 12:20:13 +01:00
THD *thd= current_thd;
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
char from[FN_REFLEN], to[FN_REFLEN], lc_from[FN_REFLEN], lc_to[FN_REFLEN];
char *from_base= from, *to_base= to;
char tmp_name[NAME_LEN+1];
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
handler *file;
FLUSH TABLE table_list Fixes for RENAME TABLE Portability fixes
2000-08-21 23:18:32 +02:00
int error=0;
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_ENTER("mysql_rename_table");
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
DBUG_PRINT("enter", ("old: '%s'.'%s' new: '%s'.'%s'",
old_db, old_name, new_db, new_name));
Some small portability fixes. Added support for lower_case_table_names=2, which is to be used on case insensitive file systems. This tells MySQL to preserve the used case of filenames and database names to make it esier to move files between cases sensitive can case insensitive file systems (like Windows and Linux)
2003-12-30 12:14:21 +01:00
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
file= (base == NULL ? 0 :
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
get_new_handler((TABLE_SHARE*) 0, thd->mem_root, base));
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
build_table_filename(from, sizeof(from), old_db, old_name, "",
flags & FN_FROM_IS_TMP);
build_table_filename(to, sizeof(to), new_db, new_name, "",
flags & FN_TO_IS_TMP);
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
/*
If lower_case_table_names == 2 (case-preserving but case-insensitive
file system) and the storage is not HA_FILE_BASED, we need to provide
a lowercase file name, but we leave the .frm in mixed case.
*/
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
if (lower_case_table_names == 2 && file &&
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
!(file->ha_table_flags() & HA_FILE_BASED))
Some small portability fixes. Added support for lower_case_table_names=2, which is to be used on case insensitive file systems. This tells MySQL to preserve the used case of filenames and database names to make it esier to move files between cases sensitive can case insensitive file systems (like Windows and Linux)
2003-12-30 12:14:21 +01:00
{
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
strmov(tmp_name, old_name);
my_casedn_str(files_charset_info, tmp_name);
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
build_table_filename(lc_from, sizeof(lc_from), old_db, tmp_name, "",
flags & FN_FROM_IS_TMP);
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
from_base= lc_from;
Some small portability fixes. Added support for lower_case_table_names=2, which is to be used on case insensitive file systems. This tells MySQL to preserve the used case of filenames and database names to make it esier to move files between cases sensitive can case insensitive file systems (like Windows and Linux)
2003-12-30 12:14:21 +01:00
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
strmov(tmp_name, new_name);
my_casedn_str(files_charset_info, tmp_name);
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
build_table_filename(lc_to, sizeof(lc_to), new_db, tmp_name, "",
flags & FN_TO_IS_TMP);
Fix table renaming to not lowercase table names for all storage engines when lower_case_table_names == 2, as it did previously for InnoDB and MEMORY. (Bug #9660)
2005-05-26 05:26:40 +02:00
to_base= lc_to;
Some small portability fixes. Added support for lower_case_table_names=2, which is to be used on case insensitive file systems. This tells MySQL to preserve the used case of filenames and database names to make it esier to move files between cases sensitive can case insensitive file systems (like Windows and Linux)
2003-12-30 12:14:21 +01:00
}
Merge mysql.com:/home/jimw/my/mysql-4.1-clean into mysql.com:/home/jimw/my/mysql-5.0-clean
2005-06-02 01:13:24 +02:00
if (!file || !(error=file->rename_table(from_base, to_base)))
RENAME TABLE table_name TO new_table_name ; Faster Alloc Small bug fixes
2000-08-21 02:00:52 +02:00
{
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
if (!(flags & NO_FRM_RENAME) && rename_file_ext(from,to,reg_ext))
RENAME TABLE table_name TO new_table_name ; Faster Alloc Small bug fixes
2000-08-21 02:00:52 +02:00
{
FLUSH TABLE table_list Fixes for RENAME TABLE Portability fixes
2000-08-21 23:18:32 +02:00
error=my_errno;
RENAME TABLE table_name TO new_table_name ; Faster Alloc Small bug fixes
2000-08-21 02:00:52 +02:00
/* Restore old file name */
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
if (file)
Merge mysql.com:/home/jimw/my/mysql-4.1-clean into mysql.com:/home/jimw/my/mysql-5.0-clean
2005-06-02 01:13:24 +02:00
file->rename_table(to_base, from_base);
RENAME TABLE table_name TO new_table_name ; Faster Alloc Small bug fixes
2000-08-21 02:00:52 +02:00
}
}
Import changeset
2000-07-31 21:29:14 +02:00
delete file;
Fix for bug #7643: ALTER TABLE fails for CSV, EXAMPLE, ARCHIVE engines
2006-04-25 12:27:28 +02:00
if (error == HA_ERR_WRONG_COMMAND)
my_error(ER_NOT_SUPPORTED_YET, MYF(0), "ALTER TABLE");
else if (error)
FLUSH TABLE table_list Fixes for RENAME TABLE Portability fixes
2000-08-21 23:18:32 +02:00
my_error(ER_ERROR_ON_RENAME, MYF(0), from, to, error);
DBUG_RETURN(error != 0);
Import changeset
2000-07-31 21:29:14 +02:00
}
Some small portability fixes. Added support for lower_case_table_names=2, which is to be used on case insensitive file systems. This tells MySQL to preserve the used case of filenames and database names to make it esier to move files between cases sensitive can case insensitive file systems (like Windows and Linux)
2003-12-30 12:14:21 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
/*
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
Force all other threads to stop using the table
SYNOPSIS
wait_while_table_is_used()
thd Thread handler
table Table to remove from cache
Fixed bug in ALTER TABLE ... DISABLE/ENABLE KEYS Removed double my_thread_end() which caused fatal error on windows if mysqld died on startup
2003-07-08 22:58:04 +02:00
function HA_EXTRA_PREPARE_FOR_DELETE if table is to be deleted
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
HA_EXTRA_FORCE_REOPEN if table is not be used
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
NOTES
When returning, the table will be unusable for other threads until
the table is closed.
PREREQUISITES
Lock on LOCK_open
Win32 clients must also have a WRITE LOCK on the table !
Import changeset
2000-07-31 21:29:14 +02:00
*/
Fixed bug in ALTER TABLE ... DISABLE/ENABLE KEYS Removed double my_thread_end() which caused fatal error on windows if mysqld died on startup
2003-07-08 22:58:04 +02:00
static void wait_while_table_is_used(THD *thd,TABLE *table,
enum ha_extra_function function)
Import changeset
2000-07-31 21:29:14 +02:00
{
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
DBUG_ENTER("wait_while_table_is_used");
Fixed a LOT of compiler warnings Added missing DBUG_RETURN statements (in mysqldump.c) Added missing enums Fixed a lot of wrong DBUG_PRINT() statements, some of which could cause crashes Removed usage of %lld and %p in printf strings as these are not portable or produces different results on different systems.
2006-11-27 00:47:38 +01:00
DBUG_PRINT("enter", ("table: '%s' share: 0x%lx db_stat: %u version: %lu",
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
table->s->table_name.str, (ulong) table->s,
table->db_stat, table->s->version));
Import changeset
2000-07-31 21:29:14 +02:00
Fixed bug in ALTER TABLE ... DISABLE/ENABLE KEYS Removed double my_thread_end() which caused fatal error on windows if mysqld died on startup
2003-07-08 22:58:04 +02:00
VOID(table->file->extra(function));
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
/* Mark all tables that are in use as 'old' */
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
mysql_lock_abort(thd, table, TRUE); /* end threads waiting on lock */
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
/* Wait until all there are no other threads that has this table open */
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
remove_table_from_cache(thd, table->s->db.str,
table->s->table_name.str,
RTFC_WAIT_OTHER_THREAD_FLAG);
One test case, one bug fix and one new feature
2003-04-03 19:24:15 +02:00
DBUG_VOID_RETURN;
}
Import changeset
2000-07-31 21:29:14 +02:00
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
/*
Close a cached table
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
SYNOPSIS
Some small portability fixes. Added support for lower_case_table_names=2, which is to be used on case insensitive file systems. This tells MySQL to preserve the used case of filenames and database names to make it esier to move files between cases sensitive can case insensitive file systems (like Windows and Linux)
2003-12-30 12:14:21 +01:00
close_cached_table()
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
thd Thread handler
table Table to remove from cache
NOTES
Function ends by signaling threads waiting for the table to try to
reopen the table.
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
PREREQUISITES
Lock on LOCK_open
Win32 clients must also have a WRITE LOCK on the table !
*/
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
WL#1218 "Triggers". Some very preliminary version of patch. Mostly needed for Monty for him getting notion what needed for triggers from new .FRM format. Things to be done: - Right placement of trigger's invocations - Right handling of errors in triggers (including transaction rollback) - Support for priviliges - Right handling of DROP/RENAME table (hope that it will be handled automatically with merging of .TRG into .FRM file) - Saving/restoring some information critical for trigger creation and replication with their definitions (e.g. sql_mode, creator, ...) - Replication Already has some known bugs so probably not for general review.
2004-09-07 14:29:46 +02:00
void close_cached_table(THD *thd, TABLE *table)
One test case, one bug fix and one new feature
2003-04-03 19:24:15 +02:00
{
DBUG_ENTER("close_cached_table");
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
Fixed bug in ALTER TABLE ... DISABLE/ENABLE KEYS Removed double my_thread_end() which caused fatal error on windows if mysqld died on startup
2003-07-08 22:58:04 +02:00
wait_while_table_is_used(thd, table, HA_EXTRA_PREPARE_FOR_DELETE);
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
/* Close lock if this is not got with LOCK TABLES */
if (thd->lock)
One test case, one bug fix and one new feature
2003-04-03 19:24:15 +02:00
{
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
mysql_unlock_tables(thd, thd->lock);
thd->lock=0; // Start locked threads
Import changeset
2000-07-31 21:29:14 +02:00
}
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
/* Close all copies of 'table'. This also frees all LOCK TABLES lock */
thd->open_tables=unlink_open_table(thd,thd->open_tables,table);
/* When lock on LOCK_open is freed other threads can continue */
Bug#16986 - Deadlock condition with MyISAM tables Addendum fixes after changing the condition variable for the global read lock. The stress test suite revealed some deadlocks. Some were related to the new condition variable (COND_global_read_lock) and some were general problems with the global read lock. It is now necessary to signal COND_global_read_lock whenever COND_refresh is signalled. We need to wait for the release of a global read lock if one is set before every operation that requires a write lock. But we must not wait if we have locked tables by LOCK TABLES. After setting a global read lock a thread waits until all write locks are released.
2006-06-26 19:14:35 +02:00
broadcast_refresh();
After merge fixes Fixed compiler warnings Fix core dump when sending SIGHUP to mysqld
2004-12-06 16:15:54 +01:00
DBUG_VOID_RETURN;
Import changeset
2000-07-31 21:29:14 +02:00
}
rename of net_pkg.cc to protocol.cc Class for sending data from server to client (Protocol) This handles both the old ( <= 4.0 ) protocol and then new binary protocol that is used for prepared statements.
2002-12-11 08:17:51 +01:00
static int send_check_errmsg(THD *thd, TABLE_LIST* table,
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
const char* operator_name, const char* errmsg)
sql_table.cc key length limit check disabled for FULLTEXT ft_parser.c Nasty buffer overflow fixed
2000-10-04 17:26:31 +02:00
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
{
rename of net_pkg.cc to protocol.cc Class for sending data from server to client (Protocol) This handles both the old ( <= 4.0 ) protocol and then new binary protocol that is used for prepared statements.
2002-12-11 08:17:51 +01:00
Protocol *protocol= thd->protocol;
protocol->prepare_for_resend();
Preparing to client<->server autoconvert
2003-03-17 10:14:04 +01:00
protocol->store(table->alias, system_charset_info);
protocol->store((char*) operator_name, system_charset_info);
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("error"), system_charset_info);
Preparing to client<->server autoconvert
2003-03-17 10:14:04 +01:00
protocol->store(errmsg, system_charset_info);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
thd->clear_error();
rename of net_pkg.cc to protocol.cc Class for sending data from server to client (Protocol) This handles both the old ( <= 4.0 ) protocol and then new binary protocol that is used for prepared statements.
2002-12-11 08:17:51 +01:00
if (protocol->write())
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
return -1;
return 1;
}
Big code cleanup/review before 4.0.2 release. (All commit emails since 4.0.1 checked) This had to be done now, before the 4.1 tree changes to much, to make it easy to propagate bug fixes to the 4.1 tree.
2002-06-11 10:20:31 +02:00
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
static int prepare_for_restore(THD* thd, TABLE_LIST* table,
Big code cleanup/review before 4.0.2 release. (All commit emails since 4.0.1 checked) This had to be done now, before the 4.1 tree changes to much, to make it easy to propagate bug fixes to the 4.1 tree.
2002-06-11 10:20:31 +02:00
HA_CHECK_OPT *check_opt)
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
{
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
DBUG_ENTER("prepare_for_restore");
sql_table.cc key length limit check disabled for FULLTEXT ft_parser.c Nasty buffer overflow fixed
2000-10-04 17:26:31 +02:00
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
if (table->table) // do not overwrite existing tables on restore
{
DBUG_RETURN(send_check_errmsg(thd, table, "restore",
"table exists, will not overwrite on restore"
));
}
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
else
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
{
THD::lex now points to THD::main_lex like in 5.0 All tests pass (client_test included)
2003-12-19 18:52:13 +01:00
char* backup_dir= thd->lex->backup_dir;
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
char src_path[FN_REFLEN], dst_path[FN_REFLEN], uname[FN_REFLEN];
Fix for bug #12739 "Deadlock in multithreaded environment during creating/ droping trigger on InnoDB table". Deadlock occured in cases when we were trying to create two triggers for the same InnoDB table concurrently and both threads were able to reach close_cached_table() simultaneously. Bugfix implements new approach to table locking and table cache invalidation during creation/dropping of trigger. No testcase is supplied since bug was repeatable only under high concurrency.
2005-10-17 20:37:24 +02:00
char* table_name= table->table_name;
char* db= table->db;
sql_table.cc key length limit check disabled for FULLTEXT ft_parser.c Nasty buffer overflow fixed
2000-10-04 17:26:31 +02:00
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
VOID(tablename_to_filename(table->table_name, uname, sizeof(uname)));
if (fn_format_relative_to_data_home(src_path, uname, backup_dir, reg_ext))
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
DBUG_RETURN(-1); // protect buffer overflow
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
build_table_filename(dst_path, sizeof(dst_path),
db, table_name, reg_ext, 0);
sql_table.cc key length limit check disabled for FULLTEXT ft_parser.c Nasty buffer overflow fixed
2000-10-04 17:26:31 +02:00
Changed DB_TYPE_INNOBASE to DB_TYPE_INNODB Fix that DROP DATABASE works with all table types Use BULK_INSERT when inserting more than one row Better TRUNCATE TABLE
2001-09-02 12:47:00 +02:00
if (lock_and_wait_for_table_name(thd,table))
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
DBUG_RETURN(-1);
sql_table.cc key length limit check disabled for FULLTEXT ft_parser.c Nasty buffer overflow fixed
2000-10-04 17:26:31 +02:00
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
if (my_copy(src_path, dst_path, MYF(MY_WME)))
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
{
Lots of code fixes to the replication code (especially the binary logging and index log file handling) Fixed bugs in my last changeset that made MySQL hard to compile. Added mutex around some data that could cause table cache corruptions when using OPTIMIZE TABLE / REPAIR TABLE or automatic repair of MyISAM tables. Added mutex around some data in the slave start/stop code that could cause THD linked list corruptions Extended my_chsize() to allow one to specify a filler character. Extend vio_blocking to return the old state (This made some usage of this function much simpler) Added testing for some functions that they caller have got the required mutexes before calling the function. Use setrlimit() to ensure that we can write core file if one specifies --core-file. Added --slave-compressed-protocol Made 2 the minimum length for ft_min_word_len Added variables foreign_key_checks & unique_checks. Less logging from replication code (if not started with --log-warnings) Changed that SHOW INNODB STATUS requre the SUPER privilege More DBUG statements and a lot of new code comments
2002-08-08 02:12:02 +02:00
pthread_mutex_lock(&LOCK_open);
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
unlock_table_name(thd, table);
Lots of code fixes to the replication code (especially the binary logging and index log file handling) Fixed bugs in my last changeset that made MySQL hard to compile. Added mutex around some data that could cause table cache corruptions when using OPTIMIZE TABLE / REPAIR TABLE or automatic repair of MyISAM tables. Added mutex around some data in the slave start/stop code that could cause THD linked list corruptions Extended my_chsize() to allow one to specify a filler character. Extend vio_blocking to return the old state (This made some usage of this function much simpler) Added testing for some functions that they caller have got the required mutexes before calling the function. Use setrlimit() to ensure that we can write core file if one specifies --core-file. Added --slave-compressed-protocol Made 2 the minimum length for ft_min_word_len Added variables foreign_key_checks & unique_checks. Less logging from replication code (if not started with --log-warnings) Changed that SHOW INNODB STATUS requre the SUPER privilege More DBUG statements and a lot of new code comments
2002-08-08 02:12:02 +02:00
pthread_mutex_unlock(&LOCK_open);
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
DBUG_RETURN(send_check_errmsg(thd, table, "restore",
"Failed copying .frm file"));
}
Changed DB_TYPE_INNOBASE to DB_TYPE_INNODB Fix that DROP DATABASE works with all table types Use BULK_INSERT when inserting more than one row Better TRUNCATE TABLE
2001-09-02 12:47:00 +02:00
if (mysql_truncate(thd, table, 1))
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
{
Lots of code fixes to the replication code (especially the binary logging and index log file handling) Fixed bugs in my last changeset that made MySQL hard to compile. Added mutex around some data that could cause table cache corruptions when using OPTIMIZE TABLE / REPAIR TABLE or automatic repair of MyISAM tables. Added mutex around some data in the slave start/stop code that could cause THD linked list corruptions Extended my_chsize() to allow one to specify a filler character. Extend vio_blocking to return the old state (This made some usage of this function much simpler) Added testing for some functions that they caller have got the required mutexes before calling the function. Use setrlimit() to ensure that we can write core file if one specifies --core-file. Added --slave-compressed-protocol Made 2 the minimum length for ft_min_word_len Added variables foreign_key_checks & unique_checks. Less logging from replication code (if not started with --log-warnings) Changed that SHOW INNODB STATUS requre the SUPER privilege More DBUG statements and a lot of new code comments
2002-08-08 02:12:02 +02:00
pthread_mutex_lock(&LOCK_open);
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
unlock_table_name(thd, table);
Lots of code fixes to the replication code (especially the binary logging and index log file handling) Fixed bugs in my last changeset that made MySQL hard to compile. Added mutex around some data that could cause table cache corruptions when using OPTIMIZE TABLE / REPAIR TABLE or automatic repair of MyISAM tables. Added mutex around some data in the slave start/stop code that could cause THD linked list corruptions Extended my_chsize() to allow one to specify a filler character. Extend vio_blocking to return the old state (This made some usage of this function much simpler) Added testing for some functions that they caller have got the required mutexes before calling the function. Use setrlimit() to ensure that we can write core file if one specifies --core-file. Added --slave-compressed-protocol Made 2 the minimum length for ft_min_word_len Added variables foreign_key_checks & unique_checks. Less logging from replication code (if not started with --log-warnings) Changed that SHOW INNODB STATUS requre the SUPER privilege More DBUG statements and a lot of new code comments
2002-08-08 02:12:02 +02:00
pthread_mutex_unlock(&LOCK_open);
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
DBUG_RETURN(send_check_errmsg(thd, table, "restore",
"Failed generating table from .frm file"));
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
}
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
}
generalizing mysql_admin_table()
2002-01-18 21:19:41 +01:00
Big code cleanup/review before 4.0.2 release. (All commit emails since 4.0.1 checked) This had to be done now, before the 4.1 tree changes to much, to make it easy to propagate bug fixes to the 4.1 tree.
2002-06-11 10:20:31 +02:00
/*
Now we should be able to open the partially restored table
to finish the restore in the handler later on
*/
Fix for bug #12739 "Deadlock in multithreaded environment during creating/ droping trigger on InnoDB table". Deadlock occured in cases when we were trying to create two triggers for the same InnoDB table concurrently and both threads were able to reach close_cached_table() simultaneously. Bugfix implements new approach to table locking and table cache invalidation during creation/dropping of trigger. No testcase is supplied since bug was repeatable only under high concurrency.
2005-10-17 20:37:24 +02:00
pthread_mutex_lock(&LOCK_open);
if (reopen_name_locked_table(thd, table))
Lots of code fixes to the replication code (especially the binary logging and index log file handling) Fixed bugs in my last changeset that made MySQL hard to compile. Added mutex around some data that could cause table cache corruptions when using OPTIMIZE TABLE / REPAIR TABLE or automatic repair of MyISAM tables. Added mutex around some data in the slave start/stop code that could cause THD linked list corruptions Extended my_chsize() to allow one to specify a filler character. Extend vio_blocking to return the old state (This made some usage of this function much simpler) Added testing for some functions that they caller have got the required mutexes before calling the function. Use setrlimit() to ensure that we can write core file if one specifies --core-file. Added --slave-compressed-protocol Made 2 the minimum length for ft_min_word_len Added variables foreign_key_checks & unique_checks. Less logging from replication code (if not started with --log-warnings) Changed that SHOW INNODB STATUS requre the SUPER privilege More DBUG statements and a lot of new code comments
2002-08-08 02:12:02 +02:00
{
Big code cleanup/review before 4.0.2 release. (All commit emails since 4.0.1 checked) This had to be done now, before the 4.1 tree changes to much, to make it easy to propagate bug fixes to the 4.1 tree.
2002-06-11 10:20:31 +02:00
unlock_table_name(thd, table);
Lots of code fixes to the replication code (especially the binary logging and index log file handling) Fixed bugs in my last changeset that made MySQL hard to compile. Added mutex around some data that could cause table cache corruptions when using OPTIMIZE TABLE / REPAIR TABLE or automatic repair of MyISAM tables. Added mutex around some data in the slave start/stop code that could cause THD linked list corruptions Extended my_chsize() to allow one to specify a filler character. Extend vio_blocking to return the old state (This made some usage of this function much simpler) Added testing for some functions that they caller have got the required mutexes before calling the function. Use setrlimit() to ensure that we can write core file if one specifies --core-file. Added --slave-compressed-protocol Made 2 the minimum length for ft_min_word_len Added variables foreign_key_checks & unique_checks. Less logging from replication code (if not started with --log-warnings) Changed that SHOW INNODB STATUS requre the SUPER privilege More DBUG statements and a lot of new code comments
2002-08-08 02:12:02 +02:00
pthread_mutex_unlock(&LOCK_open);
Fix for bug #12739 "Deadlock in multithreaded environment during creating/ droping trigger on InnoDB table". Deadlock occured in cases when we were trying to create two triggers for the same InnoDB table concurrently and both threads were able to reach close_cached_table() simultaneously. Bugfix implements new approach to table locking and table cache invalidation during creation/dropping of trigger. No testcase is supplied since bug was repeatable only under high concurrency.
2005-10-17 20:37:24 +02:00
DBUG_RETURN(send_check_errmsg(thd, table, "restore",
"Failed to open partially restored table"));
Lots of code fixes to the replication code (especially the binary logging and index log file handling) Fixed bugs in my last changeset that made MySQL hard to compile. Added mutex around some data that could cause table cache corruptions when using OPTIMIZE TABLE / REPAIR TABLE or automatic repair of MyISAM tables. Added mutex around some data in the slave start/stop code that could cause THD linked list corruptions Extended my_chsize() to allow one to specify a filler character. Extend vio_blocking to return the old state (This made some usage of this function much simpler) Added testing for some functions that they caller have got the required mutexes before calling the function. Use setrlimit() to ensure that we can write core file if one specifies --core-file. Added --slave-compressed-protocol Made 2 the minimum length for ft_min_word_len Added variables foreign_key_checks & unique_checks. Less logging from replication code (if not started with --log-warnings) Changed that SHOW INNODB STATUS requre the SUPER privilege More DBUG statements and a lot of new code comments
2002-08-08 02:12:02 +02:00
}
Fix for bug #12739 "Deadlock in multithreaded environment during creating/ droping trigger on InnoDB table". Deadlock occured in cases when we were trying to create two triggers for the same InnoDB table concurrently and both threads were able to reach close_cached_table() simultaneously. Bugfix implements new approach to table locking and table cache invalidation during creation/dropping of trigger. No testcase is supplied since bug was repeatable only under high concurrency.
2005-10-17 20:37:24 +02:00
pthread_mutex_unlock(&LOCK_open);
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
DBUG_RETURN(0);
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
}
Fixed for Ia64 + delayed key creation + a lot of small bug fixes
2000-08-15 19:09:37 +02:00
Big code cleanup/review before 4.0.2 release. (All commit emails since 4.0.1 checked) This had to be done now, before the 4.1 tree changes to much, to make it easy to propagate bug fixes to the 4.1 tree.
2002-06-11 10:20:31 +02:00
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
static int prepare_for_repair(THD *thd, TABLE_LIST *table_list,
Big code cleanup/review before 4.0.2 release. (All commit emails since 4.0.1 checked) This had to be done now, before the 4.1 tree changes to much, to make it easy to propagate bug fixes to the 4.1 tree.
2002-06-11 10:20:31 +02:00
HA_CHECK_OPT *check_opt)
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
{
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
int error= 0;
TABLE tmp_table, *table;
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
TABLE_SHARE *share;
char from[FN_REFLEN],tmp[FN_REFLEN+32];
const char **ext;
MY_STAT stat_info;
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
DBUG_ENTER("prepare_for_repair");
if (!(check_opt->sql_flags & TT_USEFRM))
DBUG_RETURN(0);
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
if (!(table= table_list->table)) /* if open_ltable failed */
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
{
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
char key[MAX_DBKEY_LENGTH];
uint key_length;
key_length= create_table_def_key(thd, key, table_list, 0);
pthread_mutex_lock(&LOCK_open);
if (!(share= (get_table_share(thd, table_list, key, key_length, 0,
&error))))
{
pthread_mutex_unlock(&LOCK_open);
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
DBUG_RETURN(0); // Can't open frm file
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
}
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
if (open_table_from_share(thd, share, "", 0, 0, 0, &tmp_table, FALSE))
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
{
release_table_share(share, RELEASE_NORMAL);
pthread_mutex_unlock(&LOCK_open);
DBUG_RETURN(0); // Out of memory
}
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
table= &tmp_table;
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
pthread_mutex_unlock(&LOCK_open);
avoid sigsegv when open_ltable() fails in REPAIR ... USE_FRM
2003-05-02 22:12:15 +02:00
}
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
/*
REPAIR TABLE ... USE_FRM for temporary tables makes little sense.
*/
if (table->s->tmp_table)
{
error= send_check_errmsg(thd, table_list, "repair",
"Cannot repair temporary table from .frm file");
goto end;
}
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
/*
User gave us USE_FRM which means that the header in the index file is
trashed.
In this case we will try to fix the table the following way:
- Rename the data file to a temporary name
- Truncate the table
- Replace the new data file with the old one
- Run a normal repair using the new index file and the old data file
*/
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
/*
Check if this is a table type that stores index and data separately,
like ISAM or MyISAM
*/
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
ext= table->file->bas_ext();
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
if (!ext[0] || !ext[1])
goto end; // No data file
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
// Name of data file
strxmov(from, table->s->normalized_path.str, ext[1], NullS);
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
if (!my_stat(from, &stat_info, MYF(0)))
goto end; // Can't use USE_FRM flag
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
my_snprintf(tmp, sizeof(tmp), "%s-%lx_%lx",
from, current_pid, thd->thread_id);
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
/* If we could open the table, close it */
if (table_list->table)
{
pthread_mutex_lock(&LOCK_open);
close_cached_table(thd, table);
pthread_mutex_unlock(&LOCK_open);
}
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
if (lock_and_wait_for_table_name(thd,table_list))
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
{
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
error= -1;
goto end;
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
}
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
if (my_rename(from, tmp, MYF(MY_WME)))
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
{
protected close_cached_table() call in REPAIR ... USE_FRM with a mutex test for REPAIR ... USE_FRM added
2002-09-01 20:17:33 +02:00
pthread_mutex_lock(&LOCK_open);
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
unlock_table_name(thd, table_list);
protected close_cached_table() call in REPAIR ... USE_FRM with a mutex test for REPAIR ... USE_FRM added
2002-09-01 20:17:33 +02:00
pthread_mutex_unlock(&LOCK_open);
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
error= send_check_errmsg(thd, table_list, "repair",
"Failed renaming data file");
goto end;
}
if (mysql_truncate(thd, table_list, 1))
{
pthread_mutex_lock(&LOCK_open);
unlock_table_name(thd, table_list);
pthread_mutex_unlock(&LOCK_open);
error= send_check_errmsg(thd, table_list, "repair",
"Failed generating table from .frm file");
goto end;
}
if (my_rename(tmp, from, MYF(MY_WME)))
{
pthread_mutex_lock(&LOCK_open);
unlock_table_name(thd, table_list);
pthread_mutex_unlock(&LOCK_open);
error= send_check_errmsg(thd, table_list, "repair",
"Failed restoring .MYD file");
goto end;
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
}
Big code cleanup/review before 4.0.2 release. (All commit emails since 4.0.1 checked) This had to be done now, before the 4.1 tree changes to much, to make it easy to propagate bug fixes to the 4.1 tree.
2002-06-11 10:20:31 +02:00
/*
Now we should be able to open the partially repaired table
to finish the repair in the handler later on.
*/
Fix for bug #12739 "Deadlock in multithreaded environment during creating/ droping trigger on InnoDB table". Deadlock occured in cases when we were trying to create two triggers for the same InnoDB table concurrently and both threads were able to reach close_cached_table() simultaneously. Bugfix implements new approach to table locking and table cache invalidation during creation/dropping of trigger. No testcase is supplied since bug was repeatable only under high concurrency.
2005-10-17 20:37:24 +02:00
pthread_mutex_lock(&LOCK_open);
if (reopen_name_locked_table(thd, table_list))
merge with 3.23.52
2002-08-08 14:24:47 +02:00
{
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
unlock_table_name(thd, table_list);
merge with 3.23.52
2002-08-08 14:24:47 +02:00
pthread_mutex_unlock(&LOCK_open);
Fix for bug #12739 "Deadlock in multithreaded environment during creating/ droping trigger on InnoDB table". Deadlock occured in cases when we were trying to create two triggers for the same InnoDB table concurrently and both threads were able to reach close_cached_table() simultaneously. Bugfix implements new approach to table locking and table cache invalidation during creation/dropping of trigger. No testcase is supplied since bug was repeatable only under high concurrency.
2005-10-17 20:37:24 +02:00
error= send_check_errmsg(thd, table_list, "repair",
"Failed to open partially repaired table");
goto end;
merge with 3.23.52
2002-08-08 14:24:47 +02:00
}
Fix for bug #12739 "Deadlock in multithreaded environment during creating/ droping trigger on InnoDB table". Deadlock occured in cases when we were trying to create two triggers for the same InnoDB table concurrently and both threads were able to reach close_cached_table() simultaneously. Bugfix implements new approach to table locking and table cache invalidation during creation/dropping of trigger. No testcase is supplied since bug was repeatable only under high concurrency.
2005-10-17 20:37:24 +02:00
pthread_mutex_unlock(&LOCK_open);
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
end:
if (table == &tmp_table)
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
{
pthread_mutex_lock(&LOCK_open);
closefrm(table, 1); // Free allocated memory
pthread_mutex_unlock(&LOCK_open);
}
Fix for MacOSX and symlinks Fix for USE_FRM and crashed index file
2003-05-14 00:27:26 +02:00
DBUG_RETURN(error);
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
}
Changed DB_TYPE_INNOBASE to DB_TYPE_INNODB Fix that DROP DATABASE works with all table types Use BULK_INSERT when inserting more than one row Better TRUNCATE TABLE
2001-09-02 12:47:00 +02:00
Big code cleanup/review before 4.0.2 release. (All commit emails since 4.0.1 checked) This had to be done now, before the 4.1 tree changes to much, to make it easy to propagate bug fixes to the 4.1 tree.
2002-06-11 10:20:31 +02:00
fixed environment creation and cleaning up for processing view one by one during checking (BUG#11337)
2005-07-05 12:37:02 +02:00
Fix for bug#5551 (Version 4). The idea of the fix is that the administrative statements OPTIMIZE TABLE, REPAIR TABLE and ANALYZE TABLE should not generate binlog errors if there is no errors on the master.
2004-11-03 18:23:58 +01:00
/*
RETURN VALUES
merge
2004-11-12 14:36:31 +01:00
FALSE Message sent to net (admin operation went ok)
TRUE Message should be sent by caller
(admin operation or network communication failed)
Fix for bug#5551 (Version 4). The idea of the fix is that the administrative statements OPTIMIZE TABLE, REPAIR TABLE and ANALYZE TABLE should not generate binlog errors if there is no errors on the master.
2004-11-03 18:23:58 +01:00
*/
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
static bool mysql_admin_table(THD* thd, TABLE_LIST* tables,
HA_CHECK_OPT* check_opt,
const char *operator_name,
thr_lock_type lock_type,
bool open_for_modify,
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
bool no_warnings_for_error,
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
uint extra_open_options,
int (*prepare_func)(THD *, TABLE_LIST *,
HA_CHECK_OPT *),
postreview changes
2004-11-21 15:35:42 +01:00
int (handler::*operator_func)(THD *,
HA_CHECK_OPT *),
int (view_operator_func)(THD *, TABLE_LIST*))
Import changeset
2000-07-31 21:29:14 +02:00
{
Fixed compiler warnings detected by option -Wshadow and -Wunused: - Removed not used variables and functions - Added #ifdef around code that is not used - Renamed variables and functions to avoid conflicts - Removed some not used arguments Fixed some class/struct warnings in ndb Added define IS_LONGDATA() to simplify code in libmysql.c I did run gcov on the changes and added 'purecov' comments on almost all lines that was not just variable name changes
2006-12-14 23:51:37 +01:00
TABLE_LIST *table;
fixed environment creation and cleaning up for processing view one by one during checking (BUG#11337)
2005-07-05 12:37:02 +02:00
SELECT_LEX *select= &thd->lex->select_lex;
Import changeset
2000-07-31 21:29:14 +02:00
List<Item> field_list;
rename of net_pkg.cc to protocol.cc Class for sending data from server to client (Protocol) This handles both the old ( <= 4.0 ) protocol and then new binary protocol that is used for prepared statements.
2002-12-11 08:17:51 +01:00
Item *item;
Protocol *protocol= thd->protocol;
fixed environment creation and cleaning up for processing view one by one during checking (BUG#11337)
2005-07-05 12:37:02 +02:00
LEX *lex= thd->lex;
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
int result_code, disable_logs= 0;
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
DBUG_ENTER("mysql_admin_table");
Import changeset
2000-07-31 21:29:14 +02:00
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
if (end_active_trans(thd))
DBUG_RETURN(1);
Import changeset
2000-07-31 21:29:14 +02:00
field_list.push_back(item = new Item_empty_string("Table", NAME_LEN*2));
item->maybe_null = 1;
field_list.push_back(item = new Item_empty_string("Op", 10));
item->maybe_null = 1;
field_list.push_back(item = new Item_empty_string("Msg_type", 10));
item->maybe_null = 1;
field_list.push_back(item = new Item_empty_string("Msg_text", 255));
item->maybe_null = 1;
Port of cursors to be pushed into 5.0 tree: - client side part is simple and may be considered stable - server side part now just joggles with THD state to save execution state and has no additional locking wisdom. Lot's of it are to be rewritten.
2004-08-03 12:32:21 +02:00
if (protocol->send_fields(&field_list,
Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
Import changeset
2000-07-31 21:29:14 +02:00
Bug#14397 - OPTIMIZE TABLE with an open HANDLER causes a crash Version for 5.0. It fixes three problems: 1. The cause of the bug was that we did not check the table version for the HANDLER ... READ commands. We did not notice when a table was replaced by a new one. This can happen during ALTER TABLE, REPAIR TABLE, and OPTIMIZE TABLE (there might be more cases). I call the fix for this problem "the primary bug fix". 2. mysql_ha_flush() was not always called with a locked LOCK_open. Though the function comment clearly said it must. I changed the code so that the locking is done when required. I call the fix for this problem "the secondary fix". 3. In 5.0 (not in 4.1 or 4.0) DROP TABLE had a possible deadlock flaw in concur with FLUSH TABLES WITH READ LOCK. I call the fix for this problem "the 5.0 addendum fix".
2005-11-15 21:57:02 +01:00
mysql_ha_flush(thd, tables, MYSQL_HA_CLOSE_FINAL, FALSE);
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
for (table= tables; table; table= table->next_local)
Import changeset
2000-07-31 21:29:14 +02:00
{
char table_name[NAME_LEN*2+2];
fixes/cleanups according to Coverity report
2005-01-24 15:48:25 +01:00
char* db = table->db;
Fix of LEFT JOIN optimizer bug, problem with key cache on Windows and a lot of portability issues.
2000-10-14 02:16:35 +02:00
bool fatal_error=0;
Import changeset
2000-07-31 21:29:14 +02:00
merged
2005-01-24 19:41:42 +01:00
strxmov(table_name, db, ".", table->table_name, NullS);
Fixes for OSF1 Made test more reliable on slow machines
2001-01-22 14:33:34 +01:00
thd->open_options|= extra_open_options;
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
table->lock_type= lock_type;
/* open only one table from local list of command */
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
{
Fixed compiler warnings detected by option -Wshadow and -Wunused: - Removed not used variables and functions - Added #ifdef around code that is not used - Renamed variables and functions to avoid conflicts - Removed some not used arguments Fixed some class/struct warnings in ndb Added define IS_LONGDATA() to simplify code in libmysql.c I did run gcov on the changes and added 'purecov' comments on almost all lines that was not just variable name changes
2006-12-14 23:51:37 +01:00
TABLE_LIST *save_next_global, *save_next_local;
save_next_global= table->next_global;
table->next_global= 0;
save_next_local= table->next_local;
table->next_local= 0;
select->table_list.first= (byte*)table;
/*
Time zone tables and SP tables can be add to lex->query_tables list,
so it have to be prepared.
TODO: Investigate if we can put extra tables into argument instead of
using lex->query_tables
*/
lex->query_tables= table;
lex->query_tables_last= &table->next_global;
lex->query_tables_own_last= 0;
thd->no_warnings_for_error= no_warnings_for_error;
if (view_operator_func == NULL)
table->required_type=FRMTYPE_TABLE;
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1 Merge of 'remove compiler warnings when using -Wshadow'
2007-01-27 02:46:45 +01:00
/*
If we want to perform an admin operation on the log table
(E.g. rename) and lock_type >= TL_READ_NO_INSERT disable
log tables
*/
rename of net_pkg.cc to protocol.cc Class for sending data from server to client (Protocol) This handles both the old ( <= 4.0 ) protocol and then new binary protocol that is used for prepared statements.
2002-12-11 08:17:51 +01:00
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1 Merge of 'remove compiler warnings when using -Wshadow'
2007-01-27 02:46:45 +01:00
if (check_if_log_table(table->db_length, table->db,
table->table_name_length,
table->table_name, 1) &&
lock_type >= TL_READ_NO_INSERT)
{
disable_logs= 1;
logger.lock();
logger.tmp_close_log_tables(thd);
}
Fixed compiler warnings detected by option -Wshadow and -Wunused: - Removed not used variables and functions - Added #ifdef around code that is not used - Renamed variables and functions to avoid conflicts - Removed some not used arguments Fixed some class/struct warnings in ndb Added define IS_LONGDATA() to simplify code in libmysql.c I did run gcov on the changes and added 'purecov' comments on almost all lines that was not just variable name changes
2006-12-14 23:51:37 +01:00
open_and_lock_tables(thd, table);
thd->no_warnings_for_error= 0;
table->next_global= save_next_global;
table->next_local= save_next_local;
thd->open_options&= ~extra_open_options;
}
generalizing mysql_admin_table()
2002-01-18 21:19:41 +01:00
if (prepare_func)
RIGHT JOIN, part of automatic repair of MyISAM tables, backup on repair, reading from FIFO, fixes for FULLTEXT, @@IDENTITY
2000-09-25 23:33:25 +02:00
{
REPAIR ... USE_FRM
2002-03-22 13:40:29 +01:00
switch ((*prepare_func)(thd, table, check_opt)) {
close table before next iteration of table proccesing in mysql_admin_table (to allow open next table)
2004-10-28 20:14:00 +02:00
case 1: // error, message written to net
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
ha_autocommit_or_rollback(thd, 1);
close table before next iteration of table proccesing in mysql_admin_table (to allow open next table)
2004-10-28 20:14:00 +02:00
close_thread_tables(thd);
continue;
case -1: // error, message could be written to net
goto err;
default: // should be 0 otherwise
;
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
}
RIGHT JOIN, part of automatic repair of MyISAM tables, backup on repair, reading from FIFO, fixes for FULLTEXT, @@IDENTITY
2000-09-25 23:33:25 +02:00
}
sql_table.cc key length limit check disabled for FULLTEXT ft_parser.c Nasty buffer overflow fixed
2000-10-04 17:26:31 +02:00
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
/*
postreview changes
2004-11-21 15:35:42 +01:00
CHECK TABLE command is only command where VIEW allowed here and this
command use only temporary teble method for VIEWs resolving => there
can't be VIEW tree substitition of join view => if opening table
succeed then table->table will have real TABLE pointer as value (in
case of join view substitution table->table can be 0, but here it is
impossible)
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
*/
Import changeset
2000-07-31 21:29:14 +02:00
if (!table->table)
{
postreview changes
2004-11-21 15:35:42 +01:00
char buf[ERRMSGSIZE+ERRMSGSIZE+2];
Import changeset
2000-07-31 21:29:14 +02:00
const char *err_msg;
rename of net_pkg.cc to protocol.cc Class for sending data from server to client (Protocol) This handles both the old ( <= 4.0 ) protocol and then new binary protocol that is used for prepared statements.
2002-12-11 08:17:51 +01:00
protocol->prepare_for_resend();
Preparing to client<->server autoconvert
2003-03-17 10:14:04 +01:00
protocol->store(table_name, system_charset_info);
protocol->store(operator_name, system_charset_info);
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("error"), system_charset_info);
Import changeset
2000-07-31 21:29:14 +02:00
if (!(err_msg=thd->net.last_error))
err_msg=ER(ER_CHECK_NO_SUCH_TABLE);
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
/* if it was a view will check md5 sum */
if (table->view &&
view_checksum(thd, table) == HA_ADMIN_WRONG_CHECKSUM)
{
postreview changes
2004-11-21 15:35:42 +01:00
strxmov(buf, err_msg, "; ", ER(ER_VIEW_CHECKSUM), NullS);
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
err_msg= (const char *)buf;
}
Preparing to client<->server autoconvert
2003-03-17 10:14:04 +01:00
protocol->store(err_msg, system_charset_info);
fixed environment creation and cleaning up for processing view one by one during checking (BUG#11337)
2005-07-05 12:37:02 +02:00
lex->cleanup_after_one_table_open();
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
thd->clear_error();
fixed environment creation and cleaning up for processing view one by one during checking (BUG#11337)
2005-07-05 12:37:02 +02:00
/*
View opening can be interrupted in the middle of process so some
tables can be left opening
*/
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
ha_autocommit_or_rollback(thd, 1);
fixed environment creation and cleaning up for processing view one by one during checking (BUG#11337)
2005-07-05 12:37:02 +02:00
close_thread_tables(thd);
Bug#17204 "second CALL to procedure crashes Server" Bug#18282 "INFORMATION_SCHEMA.TABLES provides inconsistent info about invalid views" This bug caused crashes or resulted in wrong data being returned when one tried to obtain information from I_S tables about views using stored functions. It was caused by the fact that we were using LEX representing statement which were doing select from I_S tables as active LEX when contents of I_S table were built. So state of this LEX both affected and was affected by open_tables() calls which happened during this process. This resulted in wrong behavior and in violations of some of invariants which caused crashes. This fix tries to solve this problem by properly saving/resetting and restoring part of LEX which affects and is affected by the process of opening tables and views in get_all_tables() routine. To simplify things we separated this part of LEX in a new class and made LEX its descendant.
2006-05-30 07:45:23 +02:00
lex->reset_query_tables_list(FALSE);
rename of net_pkg.cc to protocol.cc Class for sending data from server to client (Protocol) This handles both the old ( <= 4.0 ) protocol and then new binary protocol that is used for prepared statements.
2002-12-11 08:17:51 +01:00
if (protocol->write())
Import changeset
2000-07-31 21:29:14 +02:00
goto err;
continue;
}
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
if (table->view)
{
result_code= (*view_operator_func)(thd, table);
goto send_result;
}
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
if ((table->table->db_stat & HA_READ_ONLY) && open_for_modify)
Import changeset
2000-07-31 21:29:14 +02:00
{
Portability fixes
2000-10-10 23:48:03 +02:00
char buff[FN_REFLEN + MYSQL_ERRMSG_SIZE];
Fix that we can read tables with the 'older' decimal format used in 5.0.3 & 5.0.4 We will however give a warning when opening such a table that users should use ALTER TABLE ... FORCE to fix the table. In future release we will fix that REPAIR TABLE will be able to handle this case
2005-05-25 17:33:36 +02:00
uint length;
rename of net_pkg.cc to protocol.cc Class for sending data from server to client (Protocol) This handles both the old ( <= 4.0 ) protocol and then new binary protocol that is used for prepared statements.
2002-12-11 08:17:51 +01:00
protocol->prepare_for_resend();
Preparing to client<->server autoconvert
2003-03-17 10:14:04 +01:00
protocol->store(table_name, system_charset_info);
protocol->store(operator_name, system_charset_info);
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("error"), system_charset_info);
Fix that we can read tables with the 'older' decimal format used in 5.0.3 & 5.0.4 We will however give a warning when opening such a table that users should use ALTER TABLE ... FORCE to fix the table. In future release we will fix that REPAIR TABLE will be able to handle this case
2005-05-25 17:33:36 +02:00
length= my_snprintf(buff, sizeof(buff), ER(ER_OPEN_AS_READONLY),
table_name);
protocol->store(buff, length, system_charset_info);
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
ha_autocommit_or_rollback(thd, 0);
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
close_thread_tables(thd);
Merge a193-229-222-105.elisa-laajakaista.fi:/home/my/bk/mysql-5.0 into a193-229-222-105.elisa-laajakaista.fi:/home/my/bk/mysql-5.1-new
2006-05-30 15:07:49 +02:00
lex->reset_query_tables_list(FALSE);
Update of query cache code
2001-12-06 00:05:30 +01:00
table->table=0; // For query cache
rename of net_pkg.cc to protocol.cc Class for sending data from server to client (Protocol) This handles both the old ( <= 4.0 ) protocol and then new binary protocol that is used for prepared statements.
2002-12-11 08:17:51 +01:00
if (protocol->write())
Import changeset
2000-07-31 21:29:14 +02:00
goto err;
continue;
}
Fixes for automatic recover of MyISAM tables
2000-10-17 15:19:24 +02:00
/* Close all instances of the table to allow repair to rename files */
WL1019: complete patch. Reapplied patch to the clean tree to get rid of multiple typos in CS comments and unify the patch.
2006-01-19 03:56:06 +01:00
if (lock_type == TL_WRITE && table->table->s->version &&
!table->table->s->log_table)
Fixes for automatic recover of MyISAM tables
2000-10-17 15:19:24 +02:00
{
pthread_mutex_lock(&LOCK_open);
Fix race condition in ANALYZE TABLE. Fixed bug where one got an empty set instead of a DEADLOCK error when using BDB tables.
2001-11-26 01:16:38 +01:00
const char *old_message=thd->enter_cond(&COND_refresh, &LOCK_open,
"Waiting to get writelock");
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
mysql_lock_abort(thd,table->table, TRUE);
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
remove_table_from_cache(thd, table->table->s->db.str,
table->table->s->table_name.str,
Review of new code: Change argument order to be database, tablename for some functions
2005-07-27 12:05:30 +02:00
RTFC_WAIT_OTHER_THREAD_FLAG |
RTFC_CHECK_KILLED_FLAG);
Fix race condition in ANALYZE TABLE. Fixed bug where one got an empty set instead of a DEADLOCK error when using BDB tables.
2001-11-26 01:16:38 +01:00
thd->exit_cond(old_message);
Fixes for automatic recover of MyISAM tables
2000-10-17 15:19:24 +02:00
if (thd->killed)
goto err;
Flush entries from the query cache for tables used in administrative statements that may alter the table, such as REPAIR TABLE. (Bug #8480)
2005-03-08 03:15:19 +01:00
/* Flush entries in the query cache involving this table. */
query_cache_invalidate3(thd, table->table, 0);
open_for_modify= 0;
Fixes for automatic recover of MyISAM tables
2000-10-17 15:19:24 +02:00
}
WL#2645 (CHECK TABLE FOR UPGRADE) necessary implementation in the server mysql_upgrade script added
2006-02-17 07:52:32 +01:00
if (table->table->s->crashed && operator_func == &handler::ha_check)
Fix that we can read tables with the 'older' decimal format used in 5.0.3 & 5.0.4 We will however give a warning when opening such a table that users should use ALTER TABLE ... FORCE to fix the table. In future release we will fix that REPAIR TABLE will be able to handle this case
2005-05-25 17:33:36 +02:00
{
protocol->prepare_for_resend();
protocol->store(table_name, system_charset_info);
protocol->store(operator_name, system_charset_info);
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("warning"), system_charset_info);
protocol->store(STRING_WITH_LEN("Table is marked as crashed"),
system_charset_info);
Fix that we can read tables with the 'older' decimal format used in 5.0.3 & 5.0.4 We will however give a warning when opening such a table that users should use ALTER TABLE ... FORCE to fix the table. In future release we will fix that REPAIR TABLE will be able to handle this case
2005-05-25 17:33:36 +02:00
if (protocol->write())
goto err;
}
WL#2645 (CHECK TABLE FOR UPGRADE) necessary implementation in the server mysql_upgrade script added
2006-02-17 07:52:32 +01:00
if (operator_func == &handler::ha_repair)
{
if ((table->table->file->check_old_types() == HA_ADMIN_NEEDS_ALTER) ||
(table->table->file->ha_check_for_upgrade(check_opt) ==
HA_ADMIN_NEEDS_ALTER))
{
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
ha_autocommit_or_rollback(thd, 1);
WL#2645 (CHECK TABLE FOR UPGRADE) necessary implementation in the server mysql_upgrade script added
2006-02-17 07:52:32 +01:00
close_thread_tables(thd);
tmp_disable_binlog(thd); // binlogging is done by caller if wanted
result_code= mysql_recreate_table(thd, table, 0);
reenable_binlog(thd);
goto send_result;
}
}
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
result_code = (table->table->file->*operator_func)(thd, check_opt);
send_result:
fixed environment creation and cleaning up for processing view one by one during checking (BUG#11337)
2005-07-05 12:37:02 +02:00
lex->cleanup_after_one_table_open();
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
thd->clear_error(); // these errors shouldn't get client
rename of net_pkg.cc to protocol.cc Class for sending data from server to client (Protocol) This handles both the old ( <= 4.0 ) protocol and then new binary protocol that is used for prepared statements.
2002-12-11 08:17:51 +01:00
protocol->prepare_for_resend();
Preparing to client<->server autoconvert
2003-03-17 10:14:04 +01:00
protocol->store(table_name, system_charset_info);
protocol->store(operator_name, system_charset_info);
Import changeset
2000-07-31 21:29:14 +02:00
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
send_result_message:
DBUG_PRINT("info", ("result_code: %d", result_code));
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
switch (result_code) {
case HA_ADMIN_NOT_IMPLEMENTED:
corrected error message for OPTIMIZE/ANALYZE/etc, also for InnoDB that supports CHECK table but not REPAIR
2002-11-08 22:22:23 +01:00
{
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
char buf[ERRMSGSIZE+20];
uint length=my_snprintf(buf, ERRMSGSIZE,
rename of net_pkg.cc to protocol.cc Class for sending data from server to client (Protocol) This handles both the old ( <= 4.0 ) protocol and then new binary protocol that is used for prepared statements.
2002-12-11 08:17:51 +01:00
ER(ER_CHECK_NOT_IMPLEMENTED), operator_name);
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("note"), system_charset_info);
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
protocol->store(buf, length, system_charset_info);
corrected error message for OPTIMIZE/ANALYZE/etc, also for InnoDB that supports CHECK table but not REPAIR
2002-11-08 22:22:23 +01:00
}
Import changeset
2000-07-31 21:29:14 +02:00
break;
#view.test#: new file sql_table.cc, handler.h: Fixed bug #14540. Added error mnemonic code HA_ADMIN_NOT_BASE_TABLE to report that an operation cannot be applied for views. view.test, view.result: Added a test case for bug #14540. errmsg.txt: Fixed bug #14540. Added error ER_CHECK_NOT_BASE_TABLE.
2005-11-02 22:44:58 +01:00
case HA_ADMIN_NOT_BASE_TABLE:
{
Post review fixes.
2005-11-03 07:13:10 +01:00
char buf[ERRMSGSIZE+20];
uint length= my_snprintf(buf, ERRMSGSIZE,
Post review fixes.
2005-11-06 07:41:36 +01:00
ER(ER_BAD_TABLE_ERROR), table_name);
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("note"), system_charset_info);
Post review fixes.
2005-11-03 07:13:10 +01:00
protocol->store(buf, length, system_charset_info);
#view.test#: new file sql_table.cc, handler.h: Fixed bug #14540. Added error mnemonic code HA_ADMIN_NOT_BASE_TABLE to report that an operation cannot be applied for views. view.test, view.result: Added a test case for bug #14540. errmsg.txt: Fixed bug #14540. Added error ER_CHECK_NOT_BASE_TABLE.
2005-11-02 22:44:58 +01:00
}
break;
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
case HA_ADMIN_OK:
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("status"), system_charset_info);
protocol->store(STRING_WITH_LEN("OK"), system_charset_info);
Import changeset
2000-07-31 21:29:14 +02:00
break;
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
case HA_ADMIN_FAILED:
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("status"), system_charset_info);
protocol->store(STRING_WITH_LEN("Operation failed"),
system_charset_info);
Import changeset
2000-07-31 21:29:14 +02:00
break;
fixed bug #2342 "Running ANALYZE TABLE on bdb table inside a transaction hangs server thread"
2004-03-26 23:29:31 +01:00
case HA_ADMIN_REJECT:
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("status"), system_charset_info);
protocol->store(STRING_WITH_LEN("Operation need committed state"),
system_charset_info);
false/true -> FALSE/TRUE Fixes after last merge
2004-03-30 01:32:41 +02:00
open_for_modify= FALSE;
fixed bug #2342 "Running ANALYZE TABLE on bdb table inside a transaction hangs server thread" 1. added new status HA_ADMIN_REJECT and processing of it in mysql_admin_table 2. got ha_berkley::analyze to return HA_ADMIN_REJECT if there are any transactions with the table..
2004-03-26 21:35:45 +01:00
break;
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
case HA_ADMIN_ALREADY_DONE:
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("status"), system_charset_info);
protocol->store(STRING_WITH_LEN("Table is already up to date"),
system_charset_info);
Additions for CHECK table + update of benchmarks
2000-08-17 00:05:02 +02:00
break;
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
case HA_ADMIN_CORRUPT:
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("error"), system_charset_info);
protocol->store(STRING_WITH_LEN("Corrupt"), system_charset_info);
Fix of LEFT JOIN optimizer bug, problem with key cache on Windows and a lot of portability issues.
2000-10-14 02:16:35 +02:00
fatal_error=1;
Import changeset
2000-07-31 21:29:14 +02:00
break;
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
case HA_ADMIN_INVALID:
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("error"), system_charset_info);
protocol->store(STRING_WITH_LEN("Invalid argument"),
system_charset_info);
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
break;
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
case HA_ADMIN_TRY_ALTER:
{
/*
This is currently used only by InnoDB. ha_innobase::optimize() answers
"try with alter", so here we close the table, do an ALTER TABLE,
reopen the table and do ha_innobase::analyze() on it.
*/
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
ha_autocommit_or_rollback(thd, 0);
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
close_thread_tables(thd);
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
TABLE_LIST *save_next_local= table->next_local,
*save_next_global= table->next_global;
table->next_local= table->next_global= 0;
Fix for BUG#9149 "OPTIMIZE TABLE statement on InnoDB table is logged twice in the binary log" if we fall back to mysql_alter_table() (for InnoDB), don't do binlogging in mysql_alter_table(), as mysql_admin_table() is not supposed to do any binlogging (it is done by the caller).
2005-04-11 16:46:03 +02:00
tmp_disable_binlog(thd); // binlogging is done by caller if wanted
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
result_code= mysql_recreate_table(thd, table, 0);
Fix for BUG#9149 "OPTIMIZE TABLE statement on InnoDB table is logged twice in the binary log" if we fall back to mysql_alter_table() (for InnoDB), don't do binlogging in mysql_alter_table(), as mysql_admin_table() is not supposed to do any binlogging (it is done by the caller).
2005-04-11 16:46:03 +02:00
reenable_binlog(thd);
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
ha_autocommit_or_rollback(thd, 0);
EXISTS(SELECT * ...) close table before opening in optimize
2004-06-10 20:33:15 +02:00
close_thread_tables(thd);
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
if (!result_code) // recreation went ok
{
if ((table->table= open_ltable(thd, table, lock_type)) &&
((result_code= table->table->file->analyze(thd, check_opt)) > 0))
result_code= 0; // analyze went ok
}
Fix error reporting for 'OPTIMIZE TABLE' on InnoDB tables. (Bug #8135)
2005-05-24 20:11:40 +02:00
if (result_code) // either mysql_recreate_table or analyze failed
{
const char *err_msg;
if ((err_msg= thd->net.last_error))
{
if (!thd->vio_ok())
{
sql_print_error(err_msg);
}
else
{
/* Hijack the row already in-progress. */
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("error"), system_charset_info);
Fix error reporting for 'OPTIMIZE TABLE' on InnoDB tables. (Bug #8135)
2005-05-24 20:11:40 +02:00
protocol->store(err_msg, system_charset_info);
(void)protocol->write();
/* Start off another row for HA_ADMIN_FAILED */
protocol->prepare_for_resend();
protocol->store(table_name, system_charset_info);
protocol->store(operator_name, system_charset_info);
}
}
}
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
result_code= result_code ? HA_ADMIN_FAILED : HA_ADMIN_OK;
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
table->next_local= save_next_local;
table->next_global= save_next_global;
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
goto send_result_message;
}
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
case HA_ADMIN_WRONG_CHECKSUM:
{
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("note"), system_charset_info);
postreview changes
2004-11-21 15:35:42 +01:00
protocol->store(ER(ER_VIEW_CHECKSUM), strlen(ER(ER_VIEW_CHECKSUM)),
system_charset_info);
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
break;
}
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
WL#2645 (CHECK TABLE FOR UPGRADE) necessary implementation in the server mysql_upgrade script added
2006-02-17 07:52:32 +01:00
case HA_ADMIN_NEEDS_UPGRADE:
case HA_ADMIN_NEEDS_ALTER:
{
char buf[ERRMSGSIZE];
uint length;
protocol->store(STRING_WITH_LEN("error"), system_charset_info);
length=my_snprintf(buf, ERRMSGSIZE, ER(ER_TABLE_NEEDS_UPGRADE), table->table_name);
protocol->store(buf, length, system_charset_info);
fatal_error=1;
break;
}
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
default: // Probably HA_ADMIN_INTERNAL_ERROR
Improve default error message for optimize/repair table
2005-08-24 10:41:44 +02:00
{
char buf[ERRMSGSIZE+20];
uint length=my_snprintf(buf, ERRMSGSIZE,
"Unknown - internal error %d during operation",
result_code);
Inefficient usage of String::append() fixed. Bad examples of usage of a string with its length fixed. The incorrect length in the trigger file configuration descriptor fixed (BUG#14090). A hook for unknown keys added to the parser to support old .TRG files.
2005-11-20 19:47:07 +01:00
protocol->store(STRING_WITH_LEN("error"), system_charset_info);
Improve default error message for optimize/repair table
2005-08-24 10:41:44 +02:00
protocol->store(buf, length, system_charset_info);
fatal_error=1;
break;
}
Import changeset
2000-07-31 21:29:14 +02:00
}
#view.test#: new file sql_table.cc, handler.h: Fixed bug #14540. Added error mnemonic code HA_ADMIN_NOT_BASE_TABLE to report that an operation cannot be applied for views. view.test, view.result: Added a test case for bug #14540. errmsg.txt: Fixed bug #14540. Added error ER_CHECK_NOT_BASE_TABLE.
2005-11-02 22:44:58 +01:00
if (table->table)
invalidation moved before tables unlocking
2002-04-28 23:33:52 +02:00
{
WL1019: complete patch. Reapplied patch to the clean tree to get rid of multiple typos in CS comments and unify the patch.
2006-01-19 03:56:06 +01:00
/* in the below check we do not refresh the log tables */
#view.test#: new file sql_table.cc, handler.h: Fixed bug #14540. Added error mnemonic code HA_ADMIN_NOT_BASE_TABLE to report that an operation cannot be applied for views. view.test, view.result: Added a test case for bug #14540. errmsg.txt: Fixed bug #14540. Added error ER_CHECK_NOT_BASE_TABLE.
2005-11-02 22:44:58 +01:00
if (fatal_error)
table->table->s->version=0; // Force close of table
WL1019: complete patch. Reapplied patch to the clean tree to get rid of multiple typos in CS comments and unify the patch.
2006-01-19 03:56:06 +01:00
else if (open_for_modify && !table->table->s->log_table)
#view.test#: new file sql_table.cc, handler.h: Fixed bug #14540. Added error mnemonic code HA_ADMIN_NOT_BASE_TABLE to report that an operation cannot be applied for views. view.test, view.result: Added a test case for bug #14540. errmsg.txt: Fixed bug #14540. Added error ER_CHECK_NOT_BASE_TABLE.
2005-11-02 22:44:58 +01:00
{
merging fix
2006-05-03 14:43:32 +02:00
if (table->table->s->tmp_table)
merging
2006-05-03 13:33:42 +02:00
table->table->file->info(HA_STATUS_CONST);
else
{
pthread_mutex_lock(&LOCK_open);
Merge mysql.com:/home/my/mysql-5.0 into mysql.com:/home/my/mysql-5.1
2006-05-04 14:58:30 +02:00
remove_table_from_cache(thd, table->table->s->db.str,
table->table->s->table_name.str, RTFC_NO_FLAG);
merging
2006-05-03 13:33:42 +02:00
pthread_mutex_unlock(&LOCK_open);
}
/* May be something modified consequently we have to invalidate cache */
#view.test#: new file sql_table.cc, handler.h: Fixed bug #14540. Added error mnemonic code HA_ADMIN_NOT_BASE_TABLE to report that an operation cannot be applied for views. view.test, view.result: Added a test case for bug #14540. errmsg.txt: Fixed bug #14540. Added error ER_CHECK_NOT_BASE_TABLE.
2005-11-02 22:44:58 +01:00
query_cache_invalidate3(thd, table->table, 0);
}
invalidation moved before tables unlocking
2002-04-28 23:33:52 +02:00
}
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
ha_autocommit_or_rollback(thd, 0);
Import changeset
2000-07-31 21:29:14 +02:00
close_thread_tables(thd);
Update of query cache code
2001-12-06 00:05:30 +01:00
table->table=0; // For query cache
rename of net_pkg.cc to protocol.cc Class for sending data from server to client (Protocol) This handles both the old ( <= 4.0 ) protocol and then new binary protocol that is used for prepared statements.
2002-12-11 08:17:51 +01:00
if (protocol->write())
Import changeset
2000-07-31 21:29:14 +02:00
goto err;
}
Fixes and code cleanups after merge with 4.0.3 Warning handling and initial prepared statement handling (last not complete yet) Changed a lot of functions that returned 0/1 to my_bool type. GRANT handling now uses read/write locks instead of mutex Change basic net functions to use THD instead of NET (needed for 4.1 protocol) Use my_sprintf instead of sprintf() + strlen() Added alloc_query() to be able to chare query initialization code with prepared statements. Cleanup handling of SHOW COUNT(*) WARNINGS and SELECT LAST_INSERT_ID() Note that the following test fails (will be fixed ASAP): sub_select, union, rpl_rotate_logs and rpl_mystery22
2002-10-02 12:33:08 +02:00
send_eof(thd);
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
if (disable_logs)
{
if (logger.reopen_log_tables())
my_error(ER_CANT_ACTIVATE_LOG, MYF(0));
logger.unlock();
}
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(FALSE);
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
Import changeset
2000-07-31 21:29:14 +02:00
err:
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
ha_autocommit_or_rollback(thd, 1);
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
close_thread_tables(thd); // Shouldn't be needed
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
/* enable logging back if needed */
if (disable_logs)
{
if (logger.reopen_log_tables())
my_error(ER_CANT_ACTIVATE_LOG, MYF(0));
logger.unlock();
}
Update of query cache code
2001-12-06 00:05:30 +01:00
if (table)
table->table=0;
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
Import changeset
2000-07-31 21:29:14 +02:00
}
Update of query cache code
2001-12-06 00:05:30 +01:00
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool mysql_backup_table(THD* thd, TABLE_LIST* table_list)
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
{
DBUG_ENTER("mysql_backup_table");
DBUG_RETURN(mysql_admin_table(thd, table_list, 0,
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
"backup", TL_READ, 0, 0, 0, 0,
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
&handler::backup, 0));
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
}
Added new tests to benchmark suite
2001-03-11 20:20:15 +01:00
Lots of code fixes to the replication code (especially the binary logging and index log file handling) Fixed bugs in my last changeset that made MySQL hard to compile. Added mutex around some data that could cause table cache corruptions when using OPTIMIZE TABLE / REPAIR TABLE or automatic repair of MyISAM tables. Added mutex around some data in the slave start/stop code that could cause THD linked list corruptions Extended my_chsize() to allow one to specify a filler character. Extend vio_blocking to return the old state (This made some usage of this function much simpler) Added testing for some functions that they caller have got the required mutexes before calling the function. Use setrlimit() to ensure that we can write core file if one specifies --core-file. Added --slave-compressed-protocol Made 2 the minimum length for ft_min_word_len Added variables foreign_key_checks & unique_checks. Less logging from replication code (if not started with --log-warnings) Changed that SHOW INNODB STATUS requre the SUPER privilege More DBUG statements and a lot of new code comments
2002-08-08 02:12:02 +02:00
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool mysql_restore_table(THD* thd, TABLE_LIST* table_list)
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
{
DBUG_ENTER("mysql_restore_table");
DBUG_RETURN(mysql_admin_table(thd, table_list, 0,
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
"restore", TL_WRITE, 1, 1, 0,
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
&prepare_for_restore,
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
&handler::restore, 0));
BACKUP TABLE TO 'directory' RESTORE TABLE FROM 'directory' log on slave when it connects to the master
2000-09-15 00:34:50 +02:00
}
Import changeset
2000-07-31 21:29:14 +02:00
Lots of code fixes to the replication code (especially the binary logging and index log file handling) Fixed bugs in my last changeset that made MySQL hard to compile. Added mutex around some data that could cause table cache corruptions when using OPTIMIZE TABLE / REPAIR TABLE or automatic repair of MyISAM tables. Added mutex around some data in the slave start/stop code that could cause THD linked list corruptions Extended my_chsize() to allow one to specify a filler character. Extend vio_blocking to return the old state (This made some usage of this function much simpler) Added testing for some functions that they caller have got the required mutexes before calling the function. Use setrlimit() to ensure that we can write core file if one specifies --core-file. Added --slave-compressed-protocol Made 2 the minimum length for ft_min_word_len Added variables foreign_key_checks & unique_checks. Less logging from replication code (if not started with --log-warnings) Changed that SHOW INNODB STATUS requre the SUPER privilege More DBUG statements and a lot of new code comments
2002-08-08 02:12:02 +02:00
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool mysql_repair_table(THD* thd, TABLE_LIST* tables, HA_CHECK_OPT* check_opt)
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
{
DBUG_ENTER("mysql_repair_table");
DBUG_RETURN(mysql_admin_table(thd, tables, check_opt,
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
"repair", TL_WRITE, 1,
test(check_opt->sql_flags & TT_USEFRM),
HA_OPEN_FOR_REPAIR,
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
&prepare_for_repair,
WL#2645 (CHECK TABLE FOR UPGRADE) necessary implementation in the server mysql_upgrade script added
2006-02-17 07:52:32 +01:00
&handler::ha_repair, 0));
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
}
Lots of code fixes to the replication code (especially the binary logging and index log file handling) Fixed bugs in my last changeset that made MySQL hard to compile. Added mutex around some data that could cause table cache corruptions when using OPTIMIZE TABLE / REPAIR TABLE or automatic repair of MyISAM tables. Added mutex around some data in the slave start/stop code that could cause THD linked list corruptions Extended my_chsize() to allow one to specify a filler character. Extend vio_blocking to return the old state (This made some usage of this function much simpler) Added testing for some functions that they caller have got the required mutexes before calling the function. Use setrlimit() to ensure that we can write core file if one specifies --core-file. Added --slave-compressed-protocol Made 2 the minimum length for ft_min_word_len Added variables foreign_key_checks & unique_checks. Less logging from replication code (if not started with --log-warnings) Changed that SHOW INNODB STATUS requre the SUPER privilege More DBUG statements and a lot of new code comments
2002-08-08 02:12:02 +02:00
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool mysql_optimize_table(THD* thd, TABLE_LIST* tables, HA_CHECK_OPT* check_opt)
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
{
DBUG_ENTER("mysql_optimize_table");
DBUG_RETURN(mysql_admin_table(thd, tables, check_opt,
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
"optimize", TL_WRITE, 1,0,0,0,
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
&handler::optimize, 0));
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
}
mi_keycache.c: new file Many files: Added CACHE INDEX command
2003-07-16 21:30:49 +02:00
/*
Assigned specified indexes for a table into key cache
SYNOPSIS
mysql_assign_to_keycache()
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
thd Thread object
tables Table list (one table only)
mi_keycache.c: new file Many files: Added CACHE INDEX command
2003-07-16 21:30:49 +02:00
RETURN VALUES
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
FALSE ok
TRUE error
mi_keycache.c: new file Many files: Added CACHE INDEX command
2003-07-16 21:30:49 +02:00
*/
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool mysql_assign_to_keycache(THD* thd, TABLE_LIST* tables,
CLIENT_MULTI_QUERIES -> CLIENT_MULTI_STATEMENTS New multi-key-cache handling. This was needed becasue the old one didn't work reliable with MERGE tables. ALTER TABLE table_name ... CHARACTER SET ... now changes all char/varchar/text columns to the given character set (One must use ALTER TABLE ... DEFAULT CHARACTER SET ... to change the default character set) Fixed that have_compress is detected properly (fixes problems with func_compress.test on platforms without zlib) New syntax for CACHE INDEX ('keys' is optional if no index name is given and one mentions the key cache name only ones) Removed compiler warnings Added mysql_set_server_option() to allow clients like PHP to easaily set/reset the multi-statement flag.
2003-11-18 12:47:27 +01:00
LEX_STRING *key_cache_name)
mi_keycache.c: new file Many files: Added CACHE INDEX command
2003-07-16 21:30:49 +02:00
{
CLIENT_MULTI_QUERIES -> CLIENT_MULTI_STATEMENTS New multi-key-cache handling. This was needed becasue the old one didn't work reliable with MERGE tables. ALTER TABLE table_name ... CHARACTER SET ... now changes all char/varchar/text columns to the given character set (One must use ALTER TABLE ... DEFAULT CHARACTER SET ... to change the default character set) Fixed that have_compress is detected properly (fixes problems with func_compress.test on platforms without zlib) New syntax for CACHE INDEX ('keys' is optional if no index name is given and one mentions the key cache name only ones) Removed compiler warnings Added mysql_set_server_option() to allow clients like PHP to easaily set/reset the multi-statement flag.
2003-11-18 12:47:27 +01:00
HA_CHECK_OPT check_opt;
Merge key cache structures to one Fixed compiler warnings (IRIX C compiler and VC++)
2003-11-20 21:06:25 +01:00
KEY_CACHE *key_cache;
mi_keycache.c: new file Many files: Added CACHE INDEX command
2003-07-16 21:30:49 +02:00
DBUG_ENTER("mysql_assign_to_keycache");
CLIENT_MULTI_QUERIES -> CLIENT_MULTI_STATEMENTS New multi-key-cache handling. This was needed becasue the old one didn't work reliable with MERGE tables. ALTER TABLE table_name ... CHARACTER SET ... now changes all char/varchar/text columns to the given character set (One must use ALTER TABLE ... DEFAULT CHARACTER SET ... to change the default character set) Fixed that have_compress is detected properly (fixes problems with func_compress.test on platforms without zlib) New syntax for CACHE INDEX ('keys' is optional if no index name is given and one mentions the key cache name only ones) Removed compiler warnings Added mysql_set_server_option() to allow clients like PHP to easaily set/reset the multi-statement flag.
2003-11-18 12:47:27 +01:00
check_opt.init();
pthread_mutex_lock(&LOCK_global_system_variables);
if (!(key_cache= get_key_cache(key_cache_name)))
{
pthread_mutex_unlock(&LOCK_global_system_variables);
my_error(ER_UNKNOWN_KEY_CACHE, MYF(0), key_cache_name->str);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
CLIENT_MULTI_QUERIES -> CLIENT_MULTI_STATEMENTS New multi-key-cache handling. This was needed becasue the old one didn't work reliable with MERGE tables. ALTER TABLE table_name ... CHARACTER SET ... now changes all char/varchar/text columns to the given character set (One must use ALTER TABLE ... DEFAULT CHARACTER SET ... to change the default character set) Fixed that have_compress is detected properly (fixes problems with func_compress.test on platforms without zlib) New syntax for CACHE INDEX ('keys' is optional if no index name is given and one mentions the key cache name only ones) Removed compiler warnings Added mysql_set_server_option() to allow clients like PHP to easaily set/reset the multi-statement flag.
2003-11-18 12:47:27 +01:00
}
pthread_mutex_unlock(&LOCK_global_system_variables);
check_opt.key_cache= key_cache;
DBUG_RETURN(mysql_admin_table(thd, tables, &check_opt,
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
"assign_to_keycache", TL_READ_NO_INSERT, 0, 0,
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
0, 0, &handler::assign_to_keycache, 0));
mi_keycache.c: new file Many files: Added CACHE INDEX command
2003-07-16 21:30:49 +02:00
}
Many files: Added key cache assignment mi_locking.c: Added key cache assignment: correction my_sys.h: Added key cache variable structure
2003-08-02 11:43:18 +02:00
/*
Reassign all tables assigned to a key cache to another key cache
SYNOPSIS
reassign_keycache_tables()
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
thd Thread object
src_cache Reference to the key cache to clean up
dest_cache New key cache
Many files: Added key cache assignment mi_locking.c: Added key cache assignment: correction my_sys.h: Added key cache variable structure
2003-08-02 11:43:18 +02:00
CLIENT_MULTI_QUERIES -> CLIENT_MULTI_STATEMENTS New multi-key-cache handling. This was needed becasue the old one didn't work reliable with MERGE tables. ALTER TABLE table_name ... CHARACTER SET ... now changes all char/varchar/text columns to the given character set (One must use ALTER TABLE ... DEFAULT CHARACTER SET ... to change the default character set) Fixed that have_compress is detected properly (fixes problems with func_compress.test on platforms without zlib) New syntax for CACHE INDEX ('keys' is optional if no index name is given and one mentions the key cache name only ones) Removed compiler warnings Added mysql_set_server_option() to allow clients like PHP to easaily set/reset the multi-statement flag.
2003-11-18 12:47:27 +01:00
NOTES
This is called when one sets a key cache size to zero, in which
case we have to move the tables associated to this key cache to
the "default" one.
One has to ensure that one never calls this function while
some other thread is changing the key cache. This is assured by
the caller setting src_cache->in_init before calling this function.
We don't delete the old key cache as there may still be pointers pointing
to it for a while after this function returns.
RETURN VALUES
Many files: Added key cache assignment mi_locking.c: Added key cache assignment: correction my_sys.h: Added key cache variable structure
2003-08-02 11:43:18 +02:00
0 ok
*/
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
int reassign_keycache_tables(THD *thd, KEY_CACHE *src_cache,
KEY_CACHE *dst_cache)
Many files: Added key cache assignment mi_locking.c: Added key cache assignment: correction my_sys.h: Added key cache variable structure
2003-08-02 11:43:18 +02:00
{
DBUG_ENTER("reassign_keycache_tables");
CLIENT_MULTI_QUERIES -> CLIENT_MULTI_STATEMENTS New multi-key-cache handling. This was needed becasue the old one didn't work reliable with MERGE tables. ALTER TABLE table_name ... CHARACTER SET ... now changes all char/varchar/text columns to the given character set (One must use ALTER TABLE ... DEFAULT CHARACTER SET ... to change the default character set) Fixed that have_compress is detected properly (fixes problems with func_compress.test on platforms without zlib) New syntax for CACHE INDEX ('keys' is optional if no index name is given and one mentions the key cache name only ones) Removed compiler warnings Added mysql_set_server_option() to allow clients like PHP to easaily set/reset the multi-statement flag.
2003-11-18 12:47:27 +01:00
DBUG_ASSERT(src_cache != dst_cache);
DBUG_ASSERT(src_cache->in_init);
Merge key cache structures to one Fixed compiler warnings (IRIX C compiler and VC++)
2003-11-20 21:06:25 +01:00
src_cache->param_buff_size= 0; // Free key cache
CLIENT_MULTI_QUERIES -> CLIENT_MULTI_STATEMENTS New multi-key-cache handling. This was needed becasue the old one didn't work reliable with MERGE tables. ALTER TABLE table_name ... CHARACTER SET ... now changes all char/varchar/text columns to the given character set (One must use ALTER TABLE ... DEFAULT CHARACTER SET ... to change the default character set) Fixed that have_compress is detected properly (fixes problems with func_compress.test on platforms without zlib) New syntax for CACHE INDEX ('keys' is optional if no index name is given and one mentions the key cache name only ones) Removed compiler warnings Added mysql_set_server_option() to allow clients like PHP to easaily set/reset the multi-statement flag.
2003-11-18 12:47:27 +01:00
ha_resize_key_cache(src_cache);
ha_change_key_cache(src_cache, dst_cache);
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
DBUG_RETURN(0);
Many files: Added key cache assignment mi_locking.c: Added key cache assignment: correction my_sys.h: Added key cache variable structure
2003-08-02 11:43:18 +02:00
}
Many files: New feature: preload indexes into key cache. mi_preload.c: new file Many files: Added preload statement.
2003-06-12 13:29:02 +02:00
/*
Preload specified indexes for a table into key cache
SYNOPSIS
mysql_preload_keys()
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
thd Thread object
tables Table list (one table only)
Many files: New feature: preload indexes into key cache. mi_preload.c: new file Many files: Added preload statement.
2003-06-12 13:29:02 +02:00
RETURN VALUES
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
FALSE ok
TRUE error
Many files: New feature: preload indexes into key cache. mi_preload.c: new file Many files: Added preload statement.
2003-06-12 13:29:02 +02:00
*/
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool mysql_preload_keys(THD* thd, TABLE_LIST* tables)
Many files: New feature: preload indexes into key cache. mi_preload.c: new file Many files: Added preload statement.
2003-06-12 13:29:02 +02:00
{
DBUG_ENTER("mysql_preload_keys");
DBUG_RETURN(mysql_admin_table(thd, tables, 0,
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
"preload_keys", TL_READ, 0, 0, 0, 0,
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
&handler::preload_keys, 0));
Many files: New feature: preload indexes into key cache. mi_preload.c: new file Many files: Added preload statement.
2003-06-12 13:29:02 +02:00
}
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
/*
Create a table identical to the specified table
SYNOPSIS
mysql_create_like_table()
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
thd Thread object
table Table list (one table only)
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
create_info Create info
table_ident Src table_ident
RETURN VALUES
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
FALSE OK
TRUE error
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
*/
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool mysql_create_like_table(THD* thd, TABLE_LIST* table,
BUG#19309: Problem with calling proecdures twice Need to flag when a copy is needed to not overwrite a create_info object connected to the lex structure
2006-06-21 16:57:30 +02:00
HA_CREATE_INFO *lex_create_info,
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
Table_ident *table_ident)
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
{
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
TABLE *tmp_table;
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
char src_path[FN_REFLEN], dst_path[FN_REFLEN], tmp_path[FN_REFLEN];
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
char src_table_name_buff[FN_REFLEN], src_db_name_buff[FN_REFLEN];
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
uint dst_path_length;
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
char *db= table->db;
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
char *table_name= table->table_name;
Bug#11028 Crash on create table like Report error instead of crashing
2005-06-09 17:06:15 +02:00
char *src_db;
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
char *src_table= table_ident->table.str;
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
int err;
fix test failures in the runtime tree
2006-10-25 10:54:59 +02:00
bool res= TRUE, unlock_dst_table= FALSE;
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
enum legacy_db_type not_used;
BUG#19309: Problem with calling proecdures twice Need to flag when a copy is needed to not overwrite a create_info object connected to the lex structure
2006-06-21 16:57:30 +02:00
HA_CREATE_INFO *create_info;
Changes in get_table_type() and mysql_frm_type(). The main problem was that in mysql_rm_table_part2_with_lock() previously we needed to open same file twice. Now once is enough.
2005-11-03 15:10:11 +01:00
fix test failures in the runtime tree
2006-10-25 10:54:59 +02:00
TABLE_LIST src_tables_list, dst_tables_list;
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
DBUG_ENTER("mysql_create_like_table");
BUG#19309: Problem with calling proecdures twice Need to flag when a copy is needed to not overwrite a create_info object connected to the lex structure
2006-06-21 16:57:30 +02:00
if (!(create_info= copy_create_info(lex_create_info)))
{
DBUG_RETURN(TRUE);
}
A fix and a test case for Bug#19022 "Memory bug when switching db during trigger execution" Bug#17199 "Problem when view calls function from another database." Bug#18444 "Fully qualified stored function names don't work correctly in SELECT statements" Documentation note: this patch introduces a change in behaviour of prepared statements. This patch adds a few new invariants with regard to how THD::db should be used. These invariants should be preserved in future: - one should never refer to THD::db by pointer and always make a deep copy (strmake, strdup) - one should never compare two databases by pointer, but use strncmp or my_strncasecmp - TABLE_LIST object table->db should be always initialized in the parser or by creator of the object. For prepared statements it means that if the current database is changed after a statement is prepared, the database that was current at prepare remains active. This also means that you can not prepare a statement that implicitly refers to the current database if the latter is not set. This is not documented, and therefore needs documentation. This is NOT a change in behavior for almost all SQL statements except: - ALTER TABLE t1 RENAME t2 - OPTIMIZE TABLE t1 - ANALYZE TABLE t1 - TRUNCATE TABLE t1 -- until this patch t1 or t2 could be evaluated at the first execution of prepared statement. CURRENT_DATABASE() still works OK and is evaluated at every execution of prepared statement. Note, that in stored routines this is not an issue as the default database is the database of the stored procedure and "use" statement is prohibited in stored routines. This patch makes obsolete the use of check_db_used (it was never used in the old code too) and all other places that check for table->db and assign it from THD::db if it's NULL, except the parser. How this patch was created: THD::{db,db_length} were replaced with a LEX_STRING, THD::db. All the places that refer to THD::{db,db_length} were manually checked and: - if the place uses thd->db by pointer, it was fixed to make a deep copy - if a place compared two db pointers, it was fixed to compare them by value (via strcmp/my_strcasecmp, whatever was approproate) Then this intermediate patch was used to write a smaller patch that does the same thing but without a rename. TODO in 5.1: - remove check_db_used - deploy THD::set_db in mysql_change_db See also comments to individual files.
2006-06-26 22:47:52 +02:00
DBUG_ASSERT(table_ident->db.str); /* Must be set in the parser */
src_db= table_ident->db.str;
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
/*
Validate the source table
*/
if (table_ident->table.length > NAME_LEN ||
(table_ident->table.length &&
Bug#11028 Crash on create table like Report error instead of crashing
2005-06-09 17:06:15 +02:00
check_table_name(src_table,table_ident->table.length)))
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
{
Delete duplicate error message names
2003-11-18 16:28:00 +01:00
my_error(ER_WRONG_TABLE_NAME, MYF(0), src_table);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
}
Changed several char* to LEX_STRING*.
2006-10-16 18:57:33 +02:00
if (!src_db || check_db_name(&table_ident->db))
Bug#11028 Crash on create table like Report error instead of crashing
2005-06-09 17:06:15 +02:00
{
my_error(ER_WRONG_DB_NAME, MYF(0), src_db ? src_db : "NULL");
DBUG_RETURN(-1);
}
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
if ((tmp_table= find_temporary_table(thd, src_db, src_table)))
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
strxmov(src_path, tmp_table->s->path.str, reg_ext, NullS);
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
else
{
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
build_table_filename(src_path, sizeof(src_path),
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
src_db, src_table, reg_ext, 0);
Fixed during review of new pulled code
2005-02-02 19:28:01 +01:00
/* Resolve symlinks (for windows) */
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
unpack_filename(src_path, src_path);
Fix 'CREATE TABLE ... LIKE ...' when lower_case_table_names is set on case-sensitive file systems and the source table was specified in something other than lowercase. (Bug #9761)
2005-04-29 00:30:42 +02:00
if (lower_case_table_names)
my_casedn_str(files_charset_info, src_path);
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
if (access(src_path, F_OK))
{
my_error(ER_BAD_TABLE_ERROR, MYF(0), src_table);
added synchronization in mysql_create_like_table ( fixed BUG #2385 CREATE TABLE LIKE lacks locking on source and destination table and added tests for it )
2004-04-01 20:13:25 +02:00
goto err;
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
}
}
Fix for bug #6859 (after Sanja's review) Added check which throws ER_WRONG_OBJECT error in case the .frm doesn't contain a valid table type (e.g. view or trigger)
2005-08-03 07:29:48 +02:00
/*
create like should be not allowed for Views, Triggers, ...
*/
Changes in get_table_type() and mysql_frm_type(). The main problem was that in mysql_rm_table_part2_with_lock() previously we needed to open same file twice. Now once is enough.
2005-11-03 15:10:11 +01:00
if (mysql_frm_type(thd, src_path, &not_used) != FRMTYPE_TABLE)
Fix for bug #6859 (after Sanja's review) Added check which throws ER_WRONG_OBJECT error in case the .frm doesn't contain a valid table type (e.g. view or trigger)
2005-08-03 07:29:48 +02:00
{
Changed/fixed error message from last commit
2005-08-04 02:06:21 +02:00
my_error(ER_WRONG_OBJECT, MYF(0), src_db, src_table, "BASE TABLE");
Fix for bug #6859 (after Sanja's review) Added check which throws ER_WRONG_OBJECT error in case the .frm doesn't contain a valid table type (e.g. view or trigger)
2005-08-03 07:29:48 +02:00
goto err;
}
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
if (lower_case_table_names)
{
if (src_db)
{
strmake(src_db_name_buff, src_db,
min(sizeof(src_db_name_buff) - 1, table_ident->db.length));
my_casedn_str(files_charset_info, src_db_name_buff);
src_db= src_db_name_buff;
}
if (src_table)
{
strmake(src_table_name_buff, src_table,
min(sizeof(src_table_name_buff) - 1, table_ident->table.length));
my_casedn_str(files_charset_info, src_table_name_buff);
src_table= src_table_name_buff;
}
}
bzero((gptr)&src_tables_list, sizeof(src_tables_list));
src_tables_list.db= src_db;
src_tables_list.db_length= table_ident->db.length;
src_tables_list.lock_type= TL_READ;
src_tables_list.table_name= src_table;
src_tables_list.alias= src_table;
if (simple_open_n_lock_tables(thd, &src_tables_list))
DBUG_RETURN(TRUE);
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
/*
Validate the destination table
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
skip the destination table name checking as this is already
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
validated.
*/
if (create_info->options & HA_LEX_CREATE_TMP_TABLE)
{
if (find_temporary_table(thd, db, table_name))
goto table_exists;
Bug#15904 --tmpdir requires trailing slash '/' - Use same code for creating the temptable filename both from "mysql_create_table" and "mysql_create_like_table"
2006-01-23 13:28:42 +01:00
dst_path_length= build_tmptable_filename(thd, dst_path, sizeof(dst_path));
Fix for sporadically failures of innodb.test on Darwin 7.3 which occured because we were not lowering case of file names for temporary tables altough handler assumes so if lower_case_table_names==2. Now we are lowering case for them.
2004-05-25 00:30:09 +02:00
if (lower_case_table_names)
my_casedn_str(files_charset_info, dst_path);
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
create_info->table_options|= HA_CREATE_DELAY_KEY_WRITE;
}
else
{
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
dst_path_length= build_table_filename(dst_path, sizeof(dst_path),
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
db, table_name, reg_ext, 0);
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
if (!access(dst_path, F_OK))
goto table_exists;
}
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
/*
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
Create a new table by copying from source table
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
*/
Fix error message generated when trying to create a table in a non-existent database. (Bug #10407)
2005-07-22 05:08:54 +02:00
if (my_copy(src_path, dst_path, MYF(MY_DONT_OVERWRITE_FILE)))
{
if (my_errno == ENOENT)
my_error(ER_BAD_DB_ERROR,MYF(0),db);
else
my_error(ER_CANT_CREATE_FILE,MYF(0),dst_path,my_errno);
added synchronization in mysql_create_like_table ( fixed BUG #2385 CREATE TABLE LIKE lacks locking on source and destination table and added tests for it )
2004-04-01 20:13:25 +02:00
goto err;
Fix error message generated when trying to create a table in a non-existent database. (Bug #10407)
2005-07-22 05:08:54 +02:00
}
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
/*
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
As mysql_truncate don't work on a new table at this stage of
creation, instead create the table directly (for both normal
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
and temporary tables).
*/
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
/*
For partitioned tables we need to copy the .par file as well since
it is used in open_table_def to even be able to create a new handler.
There is no way to find out here if the original table is a
partitioned table so we copy the file and ignore any errors.
*/
fn_format(tmp_path, dst_path, reg_ext, ".par", MYF(MY_REPLACE_EXT));
strmov(dst_path, tmp_path);
fn_format(tmp_path, src_path, reg_ext, ".par", MYF(MY_REPLACE_EXT));
strmov(src_path, tmp_path);
my_copy(src_path, dst_path, MYF(MY_DONT_OVERWRITE_FILE));
#endif
WL#1324 table name to file name encoding - Encoding itself, implemented as a charset "filename". Originally planned to use '.' as an escape character, but now changed to '@' for two reasons: "ls" does not return file names starting with '.' considering them as a kind of hidden files; some platforms do not allow several dots in a file name. - replacing many calls of my_snprintf() and strnxmov() to the new build_table_filename(). - Adding MY_APPEND_EXT mysys flag, to append an extention rather that replace it. - Replacing all numeric constants in fn_format flag arguments to their mysys definitions, e.g. MY_UNPACK_FILENAME, - Predictability in several function/methods: when a table name can appear with or withot .frm extension. Some functions/methods were changed so accept names strictly with .frm, other - strictly without .frm extensions. Several DBUG_ASSERTs were added to check whether an extension is passed. Many files: table name to file name encoding mysql_priv.h: Prototypes for new table name encoding tools. ctype-utf8.c: Implementing "filename" charset for table name to file name encoding. row0mysql.c: Fixing table name prefix. mf_format.c: Adding MY_APPEND_EXT processing. Many files: Fixing tests. my_sys.h: Adding new flag to append rather than replace an extension. m_ctype.h: Adding "filename" charset definition.
2005-12-31 06:01:26 +01:00
dst_path[dst_path_length - reg_ext_length]= '\0'; // Remove .frm
bug #24301 Next command hangs in mysql after creating new NDB table via LIKE operation.: Locked mutex LOCK_open during operation
2006-11-15 12:05:05 +01:00
pthread_mutex_lock(&LOCK_open);
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
err= ha_create_table(thd, dst_path, db, table_name, create_info, 1);
bug #24301 Next command hangs in mysql after creating new NDB table via LIKE operation.: Locked mutex LOCK_open during operation
2006-11-15 12:05:05 +01:00
pthread_mutex_unlock(&LOCK_open);
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
if (create_info->options & HA_LEX_CREATE_TMP_TABLE)
{
if (err || !open_temporary_table(thd, dst_path, db, table_name, 1))
{
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
(void) rm_temporary_table(create_info->db_type,
dst_path); /* purecov: inspected */
added synchronization in mysql_create_like_table ( fixed BUG #2385 CREATE TABLE LIKE lacks locking on source and destination table and added tests for it )
2004-04-01 20:13:25 +02:00
goto err; /* purecov: inspected */
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
}
}
else if (err)
{
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
(void) quick_rm_table(create_info->db_type, db,
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
table_name, 0); /* purecov: inspected */
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
goto err; /* purecov: inspected */
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
}
Changed wellformedlen to well_formed_len Fixed that blobs >16M can be inserted/updated Fixed bug when doing CREATE TEMPORARY TABLE ... LIKE
2004-02-17 00:35:17 +01:00
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
/*
We have to write the query before we unlock the tables.
*/
WL#2977 and WL#2712 global and session-level variable to set the binlog format (row/statement), and new binlog format called "mixed" (which is statement-based except if only row-based is correct, in this cset it means if UDF or UUID is used; more cases could be added in later 5.1 release): SET GLOBAL|SESSION BINLOG_FORMAT=row|statement|mixed|default; the global default is statement unless cluster is enabled (then it's row) as in 5.1-alpha. It's not possible to use SET on this variable if a session is currently in row-based mode and has open temporary tables (because CREATE TEMPORARY TABLE was not binlogged so temp table is not known on slave), or if NDB is enabled (because NDB does not support such change on-the-fly, though it will later), of if in a stored function (see below). The added tests test the possibility or impossibility to SET, their effects, and the mixed mode, including in prepared statements and in stored procedures and functions. Caveats: a) The mixed mode will not work for stored functions: in mixed mode, a stored function will always be binlogged as one call and in a statement-based way (e.g. INSERT VALUES(myfunc()) or SELECT myfunc()). b) for the same reason, changing the thread's binlog format inside a stored function is refused with an error message. c) the same problems apply to triggers; implementing b) for triggers will be done later (will ask Dmitri). Additionally, as the binlog format is now changeable by each user for his session, I remove the implication which was done at startup, where row-based automatically set log-bin-trust-routine-creators to 1 (not possible anymore as a user can now switch to stmt-based and do nasty things again), and automatically set --innodb-locks-unsafe-for-binlog to 1 (was anyway theoretically incorrect as it disabled phantom protection). Plus fixes for compiler warnings.
2006-02-25 22:21:03 +01:00
if (thd->current_stmt_binlog_row_based)
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
{
/*
Since temporary tables are not replicated under row-based
replication, CREATE TABLE ... LIKE ... needs special
treatement. We have four cases to consider, according to the
following decision table:
==== ========= ========= ==============================
Case Target Source Write to binary log
==== ========= ========= ==============================
1 normal normal Original statement
2 normal temporary Generated statement
3 temporary normal Nothing
4 temporary temporary Nothing
==== ========= ========= ==============================
The variable 'tmp_table' below is used to see if the source
table is a temporary table: if it is set, then the source table
was a temporary table and we can take apropriate actions.
*/
if (!(create_info->options & HA_LEX_CREATE_TMP_TABLE))
{
if (tmp_table) // Case 2
{
char buf[2048];
String query(buf, sizeof(buf), system_charset_info);
query.length(0); // Have to zero it since constructor doesn't
fix test failures in the runtime tree
2006-10-25 10:54:59 +02:00
uint counter;
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
/*
fix test failures in the runtime tree
2006-10-25 10:54:59 +02:00
Here we open the destination table. This is needed for
store_create_info() to work. The table will be closed
by close_thread_tables() at the end of the statement.
*/
if (open_tables(thd, &table, &counter, 0))
goto err;
bzero((gptr)&dst_tables_list, sizeof(dst_tables_list));
dst_tables_list.db= table->db;
dst_tables_list.table_name= table->table_name;
/*
lock destination table name, to make sure that nobody
can drop/alter the table while we execute store_create_info()
*/
if (lock_and_wait_for_table_name(thd, &dst_tables_list))
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
goto err;
fix test failures in the runtime tree
2006-10-25 10:54:59 +02:00
else
unlock_dst_table= TRUE;
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
int result= store_create_info(thd, table, &query, create_info);
DBUG_ASSERT(result == 0); // store_create_info() always return 0
write_bin_log(thd, TRUE, query.ptr(), query.length());
}
else // Case 1
write_bin_log(thd, TRUE, thd->query, thd->query_length);
}
/*
Case 3 and 4 does nothing under RBR
*/
}
Bug#15920 (Temporary tables are not logged in binlog): Fixing faulty tests preventing some CREATE TEMPORARY TABLE statements from being binlogged under statement-based replication.
2005-12-22 11:09:50 +01:00
else
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
write_bin_log(thd, TRUE, thd->query, thd->query_length);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
res= FALSE;
made some optimization of last patch for Bug #2385 "CREATE TABLE LIKE lacks locking on source and destination table" (in sql_table.cc and mysql-test/t/synchronization.test)
2004-04-02 23:44:38 +02:00
goto err;
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
table_exists:
if (create_info->options & HA_LEX_CREATE_IF_NOT_EXISTS)
{
char warn_buff[MYSQL_ERRMSG_SIZE];
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
my_snprintf(warn_buff, sizeof(warn_buff),
ER(ER_TABLE_EXISTS_ERROR), table_name);
After merge fixes of merge with 4.1 that included the new arena code. Fixed (together with Guilhem) bugs in mysqlbinlog regarding --offset Prefix addresses with 0x for easier comparisons of debug logs Fixed problem where MySQL choosed index-read even if there would be a much better range on the same index This fix changed some 'index' queries to 'range' queries in the test suite Don't create 'dummy' WHERE clause for trivial WHERE clauses where we can remove the WHERE clause. This fix removed of a lot of 'Using where' notes in the test suite. Give NOTE instead of WARNING if table/function doesn't exists when using DROP IF EXISTS Give NOTE instead of WARNING for safe field-type conversions
2004-09-09 05:59:26 +02:00
push_warning(thd, MYSQL_ERROR::WARN_LEVEL_NOTE,
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
ER_TABLE_EXISTS_ERROR,warn_buff);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
res= FALSE;
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
}
made some optimization of last patch for Bug #2385 "CREATE TABLE LIKE lacks locking on source and destination table" (in sql_table.cc and mysql-test/t/synchronization.test)
2004-04-02 23:44:38 +02:00
else
my_error(ER_TABLE_EXISTS_ERROR, MYF(0), table_name);
err:
fix test failures in the runtime tree
2006-10-25 10:54:59 +02:00
if (unlock_dst_table)
{
pthread_mutex_lock(&LOCK_open);
unlock_table_name(thd, &dst_tables_list);
pthread_mutex_unlock(&LOCK_open);
}
made some optimization of last patch for Bug #2385 "CREATE TABLE LIKE lacks locking on source and destination table" (in sql_table.cc and mysql-test/t/synchronization.test)
2004-04-02 23:44:38 +02:00
DBUG_RETURN(res);
sql_table.cc: Fix CREATE LIKE TABLE .. temporary case
2002-12-28 10:38:29 +01:00
}
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool mysql_analyze_table(THD* thd, TABLE_LIST* tables, HA_CHECK_OPT* check_opt)
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
{
Fixes for OS2. Fix bug in isamlog Add argument types to function declarations.
2001-08-22 00:45:07 +02:00
thr_lock_type lock_type = TL_READ_NO_INSERT;
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
DBUG_ENTER("mysql_analyze_table");
DBUG_RETURN(mysql_admin_table(thd, tables, check_opt,
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
"analyze", lock_type, 1, 0, 0, 0,
VIEW support for CHECK TABLE command (WL#1984)
2004-10-28 18:37:25 +02:00
&handler::analyze, 0));
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
}
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool mysql_check_table(THD* thd, TABLE_LIST* tables,HA_CHECK_OPT* check_opt)
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
{
Fixes for OS2. Fix bug in isamlog Add argument types to function declarations.
2001-08-22 00:45:07 +02:00
thr_lock_type lock_type = TL_READ_NO_INSERT;
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
DBUG_ENTER("mysql_check_table");
DBUG_RETURN(mysql_admin_table(thd, tables, check_opt,
Fixes for OS2. Fix bug in isamlog Add argument types to function declarations.
2001-08-22 00:45:07 +02:00
"check", lock_type,
Show all generated warnings in SHOW ERROR Previously we only stored the first given error (the error sent to the client)
2005-02-24 22:33:42 +01:00
0, HA_OPEN_FOR_REPAIR, 0, 0,
WL#2645 (CHECK TABLE FOR UPGRADE) necessary implementation in the server mysql_upgrade script added
2006-02-17 07:52:32 +01:00
&handler::ha_check, &view_checksum));
Update to new root alloc, OPTIMIZE TABLE and some other changes
2000-09-12 02:02:33 +02:00
}
After merge fixes Some bigger code changes was necessary becasue of the multi-table-update and the new HANDLER code
2004-10-07 09:50:13 +02:00
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
/* table_list should contain just one table */
After merge fixes Some bigger code changes was necessary becasue of the multi-table-update and the new HANDLER code
2004-10-07 09:50:13 +02:00
static int
mysql_discard_or_import_tablespace(THD *thd,
TABLE_LIST *table_list,
enum tablespace_op_type tablespace_op)
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
{
TABLE *table;
my_bool discard;
int error;
DBUG_ENTER("mysql_discard_or_import_tablespace");
After merge fixes Some bigger code changes was necessary becasue of the multi-table-update and the new HANDLER code
2004-10-07 09:50:13 +02:00
/*
Note that DISCARD/IMPORT TABLESPACE always is the only operation in an
ALTER TABLE
*/
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
thd->proc_info="discard_or_import_tablespace";
After merge fixes Some bigger code changes was necessary becasue of the multi-table-update and the new HANDLER code
2004-10-07 09:50:13 +02:00
discard= test(tablespace_op == DISCARD_TABLESPACE);
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
After merge fixes Some bigger code changes was necessary becasue of the multi-table-update and the new HANDLER code
2004-10-07 09:50:13 +02:00
/*
We set this flag so that ha_innobase::open and ::external_lock() do
not complain when we lock the table
*/
thd->tablespace_op= TRUE;
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
if (!(table=open_ltable(thd,table_list,TL_WRITE)))
{
thd->tablespace_op=FALSE;
DBUG_RETURN(-1);
}
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
error=table->file->discard_or_import_tablespace(discard);
thd->proc_info="end";
if (error)
goto err;
After merge fixes Some bigger code changes was necessary becasue of the multi-table-update and the new HANDLER code
2004-10-07 09:50:13 +02:00
/*
The 0 in the call below means 'not in a transaction', which means
immediate invalidation; that is probably what we wish here
*/
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
query_cache_invalidate3(thd, table_list, 0);
/* The ALTER TABLE is always in its own transaction */
error = ha_commit_stmt(thd);
if (ha_commit(thd))
error=1;
if (error)
goto err;
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
write_bin_log(thd, FALSE, thd->query, thd->query_length);
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
err:
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
ha_autocommit_or_rollback(thd, error);
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
close_thread_tables(thd);
sql_yacc.yy: Add DISCARD etc. in the keyword: rule of sql_yacc.yy sql_table.cc: Add comment that DISCARD/IMPORT TABLESPACE is always the only clause in an ALTER TABLE
2003-10-14 00:52:03 +02:00
thd->tablespace_op=FALSE;
fix for bug 12207 (alter table discard tablespace on MyISAM table causes ERROR 2013). (all-in-one approved patch)
2005-08-29 21:00:43 +02:00
false/true -> FALSE/TRUE Fixes after last merge
2004-03-30 01:32:41 +02:00
if (error == 0)
{
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
send_ok(thd);
false/true -> FALSE/TRUE Fixes after last merge
2004-03-30 01:32:41 +02:00
DBUG_RETURN(0);
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
}
sql_table.cc: Return a sensible error code from DISCARD TABLESPACE, if it fails because the table is referenced by a FOREIGN KEY
2004-12-27 04:03:11 +01:00
fix for bug 12207 (alter table discard tablespace on MyISAM table causes ERROR 2013). (all-in-one approved patch)
2005-08-29 21:00:43 +02:00
table->file->print_error(error, MYF(0));
sql_table.cc: Return a sensible error code from DISCARD TABLESPACE, if it fails because the table is referenced by a FOREIGN KEY
2004-12-27 04:03:11 +01:00
DBUG_RETURN(-1);
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
}
Import changeset
2000-07-31 21:29:14 +02:00
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
/*
SYNOPSIS
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
compare_tables()
table The original table.
create_list The fields for the new table.
key_info_buffer An array of KEY structs for the new indexes.
key_count The number of elements in the array.
create_info Create options for the new table.
alter_info Alter options.
order_num Number of order list elements.
index_drop_buffer OUT An array of offsets into table->key_info.
index_drop_count OUT The number of elements in the array.
index_add_buffer OUT An array of offsets into key_info_buffer.
index_add_count OUT The number of elements in the array.
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
DESCRIPTION
'table' (first argument) contains information of the original
table, which includes all corresponding parts that the new
table has in arguments create_list, key_list and create_info.
By comparing the changes between the original and new table
we can determine how much it has changed after ALTER TABLE
and whether we need to make a copy of the table, or just change
the .frm file.
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
If there are no data changes, but index changes, 'index_drop_buffer'
and/or 'index_add_buffer' are populated with offsets into
table->key_info or key_info_buffer respectively for the indexes
that need to be dropped and/or (re-)created.
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
RETURN VALUES
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
0 No copy needed
ALTER_TABLE_DATA_CHANGED Data changes, copy needed
ALTER_TABLE_INDEX_CHANGED Index changes, copy might be needed
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
*/
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
static uint compare_tables(TABLE *table, List<create_field> *create_list,
KEY *key_info_buffer, uint key_count,
HA_CREATE_INFO *create_info,
ALTER_INFO *alter_info, uint order_num,
uint *index_drop_buffer, uint *index_drop_count,
Merge a193-229-222-105.elisa-laajakaista.fi:/home/my/bk/mysql-5.0 into a193-229-222-105.elisa-laajakaista.fi:/home/my/bk/mysql-5.1-new
2006-05-30 15:07:49 +02:00
uint *index_add_buffer, uint *index_add_count,
bool varchar)
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
{
Field **f_ptr, *field;
uint changes= 0, tmp;
List_iterator_fast<create_field> new_field_it(*create_list);
create_field *new_field;
Fix for Bug#17888 DD: ADD INDEX causes error 756 'Index on disk column is not supported
2006-03-06 10:24:06 +01:00
KEY_PART_INFO *key_part;
KEY_PART_INFO *end;
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
DBUG_ENTER("compare_tables");
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
/*
Some very basic checks. If number of fields changes, or the
handler, we need to run full ALTER TABLE. In the future
new fields can be added and old dropped without copy, but
not yet.
Test also that engine was not given during ALTER TABLE, or
we are force to run regular alter table (copy).
E.g. ALTER TABLE tbl_name ENGINE=MyISAM.
For the following ones we also want to run regular alter table:
ALTER TABLE tbl_name ORDER BY ..
ALTER TABLE tbl_name CONVERT TO CHARACTER SET ..
At the moment we can't handle altering temporary tables without a copy.
We also test if OPTIMIZE TABLE was given and was mapped to alter table.
In that case we always do full copy.
After merge fix.
2006-06-14 13:45:29 +02:00
There was a bug prior to mysql-4.0.25. Number of null fields was
calculated incorrectly. As a result frm and data files gets out of
sync after fast alter table. There is no way to determine by which
mysql version (in 4.0 and 4.1 branches) table was created, thus we
disable fast alter table for all tables created by mysql versions
prior to 5.0 branch.
See BUG#6236.
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
*/
if (table->s->fields != create_list->elements ||
table->s->db_type != create_info->db_type ||
table->s->tmp_table ||
create_info->used_fields & HA_CREATE_USED_ENGINE ||
create_info->used_fields & HA_CREATE_USED_CHARSET ||
create_info->used_fields & HA_CREATE_USED_DEFAULT_CHARSET ||
Bug#18477 - MySQL/InnoDB Ignoring Foreign Keys in ALTER TABLE InnoDB requires a full table rebuild for foreign key changes. It was not possible in compare_tables() to detect such changes. On Heikkis proposal I added a new flag to the syntax parser where foreign key definition changes are done. I test for this flag in compare_tables() now.
2006-03-30 20:55:54 +02:00
(alter_info->flags & (ALTER_RECREATE | ALTER_FOREIGN_KEY)) ||
Merge a193-229-222-105.elisa-laajakaista.fi:/home/my/bk/mysql-5.0 into a193-229-222-105.elisa-laajakaista.fi:/home/my/bk/mysql-5.1-new
2006-05-30 15:07:49 +02:00
order_num ||
After merge fix.
2006-06-14 13:45:29 +02:00
!table->s->mysql_version ||
Merge a193-229-222-105.elisa-laajakaista.fi:/home/my/bk/mysql-5.0 into a193-229-222-105.elisa-laajakaista.fi:/home/my/bk/mysql-5.1-new
2006-05-30 15:07:49 +02:00
(table->s->frm_version < FRM_VER_TRUE_VARCHAR && varchar))
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
DBUG_RETURN(ALTER_TABLE_DATA_CHANGED);
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
/*
Go through fields and check if the original ones are compatible
with new table.
*/
for (f_ptr= table->field, new_field= new_field_it++;
(field= *f_ptr); f_ptr++, new_field= new_field_it++)
{
/* Make sure we have at least the default charset in use. */
if (!new_field->charset)
new_field->charset= create_info->default_table_charset;
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
/* Check that NULL behavior is same for old and new fields */
if ((new_field->flags & NOT_NULL_FLAG) !=
(uint) (field->flags & NOT_NULL_FLAG))
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
DBUG_RETURN(ALTER_TABLE_DATA_CHANGED);
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
/* Don't pack rows in old tables if the user has requested this. */
if (create_info->row_type == ROW_TYPE_DYNAMIC ||
(new_field->flags & BLOB_FLAG) ||
new_field->sql_type == MYSQL_TYPE_VARCHAR &&
create_info->row_type != ROW_TYPE_FIXED)
create_info->table_options|= HA_OPTION_PACK_RECORD;
Added flag to detect renaming of fields (not supported as fast alter table for ndbcluster)
2006-06-14 11:08:36 +02:00
/* Check if field was renamed */
Added flag to detect renaming of fields (not supported as fast alter table for ndbcluster): psot review comment: cleared flag before checking
2006-06-14 12:01:06 +02:00
field->flags&= ~FIELD_IS_RENAMED;
Added flag to detect renaming of fields (not supported as fast alter table for ndbcluster)
2006-06-14 11:08:36 +02:00
if (my_strcasecmp(system_charset_info,
field->field_name,
new_field->field_name))
field->flags|= FIELD_IS_RENAMED;
Added flag to detect renaming of fields (not supported as fast alter table for ndbcluster): psot review comment: cleared flag before checking
2006-06-14 12:01:06 +02:00
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
/* Evaluate changes bitmap and send to check_if_incompatible_data() */
if (!(tmp= field->is_equal(new_field)))
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
DBUG_RETURN(ALTER_TABLE_DATA_CHANGED);
Added possibillity to check what fields will get added indexes (ndb does currently not support indexes on disk stored fields), WL#1892
2006-01-27 17:23:14 +01:00
// Clear indexed marker
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
field->flags&= ~FIELD_IN_ADD_INDEX;
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
changes|= tmp;
}
/*
Go through keys and check if the original ones are compatible
with new table.
*/
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
KEY *table_key;
KEY *table_key_end= table->key_info + table->s->keys;
KEY *new_key;
KEY *new_key_end= key_info_buffer + key_count;
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
DBUG_PRINT("info", ("index count old: %d new: %d",
table->s->keys, key_count));
/*
Step through all keys of the old table and search matching new keys.
*/
*index_drop_count= 0;
*index_add_count= 0;
for (table_key= table->key_info; table_key < table_key_end; table_key++)
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
{
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
KEY_PART_INFO *table_part;
KEY_PART_INFO *table_part_end= table_key->key_part + table_key->key_parts;
KEY_PART_INFO *new_part;
/* Search a new key with the same name. */
for (new_key= key_info_buffer; new_key < new_key_end; new_key++)
{
if (! strcmp(table_key->name, new_key->name))
break;
}
if (new_key >= new_key_end)
{
/* Key not found. Add the offset of the key to the drop buffer. */
index_drop_buffer[(*index_drop_count)++]= table_key - table->key_info;
DBUG_PRINT("info", ("index dropped: '%s'", table_key->name));
continue;
}
/* Check that the key types are compatible between old and new tables. */
if ((table_key->algorithm != new_key->algorithm) ||
((table_key->flags & HA_KEYFLAG_MASK) !=
(new_key->flags & HA_KEYFLAG_MASK)) ||
(table_key->key_parts != new_key->key_parts))
goto index_changed;
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
/*
Check that the key parts remain compatible between the old and
new tables.
*/
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
for (table_part= table_key->key_part, new_part= new_key->key_part;
table_part < table_part_end;
table_part++, new_part++)
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
{
/*
Key definition has changed if we are using a different field or
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
if the used key part length is different. We know that the fields
did not change. Comparing field numbers is sufficient.
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
*/
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
if ((table_part->length != new_part->length) ||
(table_part->fieldnr - 1 != new_part->fieldnr))
goto index_changed;
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
}
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
continue;
index_changed:
/* Key modified. Add the offset of the key to both buffers. */
index_drop_buffer[(*index_drop_count)++]= table_key - table->key_info;
index_add_buffer[(*index_add_count)++]= new_key - key_info_buffer;
Fix for Bug#17888 DD: ADD INDEX causes error 756 'Index on disk column is not supported
2006-03-06 10:24:06 +01:00
key_part= new_key->key_part;
end= key_part + new_key->key_parts;
for(; key_part != end; key_part++)
{
// Mark field to be part of new key
field= table->field[key_part->fieldnr];
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
field->flags|= FIELD_IN_ADD_INDEX;
Fix for Bug#17888 DD: ADD INDEX causes error 756 'Index on disk column is not supported
2006-03-06 10:24:06 +01:00
}
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
DBUG_PRINT("info", ("index changed: '%s'", table_key->name));
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
}
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
/*end of for (; table_key < table_key_end;) */
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
/*
Step through all keys of the new table and find matching old keys.
*/
for (new_key= key_info_buffer; new_key < new_key_end; new_key++)
{
/* Search an old key with the same name. */
for (table_key= table->key_info; table_key < table_key_end; table_key++)
{
if (! strcmp(table_key->name, new_key->name))
break;
}
if (table_key >= table_key_end)
{
/* Key not found. Add the offset of the key to the add buffer. */
index_add_buffer[(*index_add_count)++]= new_key - key_info_buffer;
Fix for Bug#17888 DD: ADD INDEX causes error 756 'Index on disk column is not supported
2006-03-06 10:24:06 +01:00
key_part= new_key->key_part;
end= key_part + new_key->key_parts;
for(; key_part != end; key_part++)
{
// Mark field to be part of new key
field= table->field[key_part->fieldnr];
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
field->flags|= FIELD_IN_ADD_INDEX;
Fix for Bug#17888 DD: ADD INDEX causes error 756 'Index on disk column is not supported
2006-03-06 10:24:06 +01:00
}
Post review fixes
2006-01-31 15:53:35 +01:00
DBUG_PRINT("info", ("index added: '%s'", new_key->name));
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
}
}
Added possibillity to check what fields will get added indexes (ndb does currently not support indexes on disk stored fields), WL#1892
2006-01-27 17:23:14 +01:00
/* Check if changes are compatible with current handler without a copy */
if (table->file->check_if_incompatible_data(create_info, changes))
DBUG_RETURN(ALTER_TABLE_DATA_CHANGED);
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
if (*index_drop_count || *index_add_count)
DBUG_RETURN(ALTER_TABLE_INDEX_CHANGED);
DBUG_RETURN(0); // Tables are compatible
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
}
Fix for bug#24395: ALTER TABLE DISABLE KEYS doesn't work when modifying the table ENABLE|DISABLE KEYS combined with another ALTER TABLE option, different than RENAME TO did nothing. Also, if the table had disabled keys and was ALTER-ed then the end table was with enabled keys. Fixed by checking whether the table had disabled keys and enabling them in the copied table.
2006-11-28 18:27:32 +01:00
/*
Manages enabling/disabling of indexes for ALTER TABLE
SYNOPSIS
alter_table_manage_keys()
table Target table
indexes_were_disabled Whether the indexes of the from table
were disabled
keys_onoff ENABLE | DISABLE | LEAVE_AS_IS
RETURN VALUES
FALSE OK
TRUE Error
*/
static
bool alter_table_manage_keys(TABLE *table, int indexes_were_disabled,
enum enum_enable_or_disable keys_onoff)
{
int error= 0;
DBUG_ENTER("alter_table_manage_keys");
DBUG_PRINT("enter", ("table=%p were_disabled=%d on_off=%d",
table, indexes_were_disabled, keys_onoff));
switch (keys_onoff) {
case ENABLE:
error= table->file->enable_indexes(HA_KEY_SWITCH_NONUNIQ_SAVE);
break;
case LEAVE_AS_IS:
if (!indexes_were_disabled)
break;
/* fall-through: disabled indexes */
case DISABLE:
error= table->file->disable_indexes(HA_KEY_SWITCH_NONUNIQ_SAVE);
}
if (error == HA_ERR_WRONG_COMMAND)
{
push_warning_printf(current_thd, MYSQL_ERROR::WARN_LEVEL_NOTE,
Merge example.com:/work/bug24395-v2/my41 into example.com:/work/bug24395-v2/my50 fix for bug#24395 merged into 5.0
2006-11-30 18:36:15 +01:00
ER_ILLEGAL_HA, ER(ER_ILLEGAL_HA), table->s->table_name);
Fix for bug#24395: ALTER TABLE DISABLE KEYS doesn't work when modifying the table ENABLE|DISABLE KEYS combined with another ALTER TABLE option, different than RENAME TO did nothing. Also, if the table had disabled keys and was ALTER-ed then the end table was with enabled keys. Fixed by checking whether the table had disabled keys and enabling them in the copied table.
2006-11-28 18:27:32 +01:00
error= 0;
} else if (error)
table->file->print_error(error, MYF(0));
DBUG_RETURN(error);
}
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
/*
Alter table
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
SYNOPSIS
mysql_alter_table()
thd Thread handle
new_db If there is a RENAME clause
new_name If there is a RENAME clause
lex_create_info Information from the parsing phase. Since some
clauses are common to CREATE and ALTER TABLE, the
data is stored in lex->create_info. The non-common
is stored in lex->alter_info.
table_list The table to change.
fields lex->create_list - List of fields to be changed,
added or dropped.
keys lex->key_list - List of keys to be changed, added or
dropped.
order_num How many ORDER BY fields has been specified.
order List of fields to ORDER BY.
ignore Whether we have ALTER IGNORE TABLE
alter_info Information from the parsing phase specific to ALTER
TABLE and not shared with CREATE TABLE.
do_send_ok Whether to call send_ok() on success.
DESCRIPTION
This is a veery long function and is everything but the kitchen sink :)
It is used to alter a table and not only by ALTER TABLE but also
CREATE|DROP INDEX are mapped on this function.
When the ALTER TABLE statement just does a RENAME or ENABLE|DISABLE KEYS,
or both, then this function short cuts its operation by renaming
the table and/or enabling/disabling the keys. In this case, the FRM is
not changed, directly by mysql_alter_table. However, if there is a
RENAME + change of a field, or an index, the short cut is not used.
See how `fields` is used to generate the new FRM regarding the structure
of the fields. The same is done for the indices of the table.
Important is the fact, that this function tries to do as little work as
possible, by finding out whether a intermediate table is needed to copy
data into and when finishing the altering to use it as the original table.
For this reason the function compare_tables() is called, which decides
based on all kind of data how similar are the new and the original
tables.
RETURN VALUES
FALSE OK
TRUE Error
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
*/
Worklog#1563 - Support of on-line CREATE/DROP INDEX. This is to enable table handlers to implement online create/drop index. It consists of some parts: - New default handler methods in handler.h - Split of mysql_alter_table. It decides if only one kind of alteration is to be done (e.g. only create indexes or only drop indexes etc.) It then calls the specialized new handler method if the handler implements it. Otherwise it calls real_alter_table. - The parser sets flags for each alter operation detected in a command. These are used by mysql_alter_table for the decision. - mysql_prepare_table is pulled out of mysql_create_table. This is also used by mysql_create_index to prepare the key structure array for the handler. It is also used by mysql_create_index and mysql_drop_index to prepare a call to mysql_create_frm. - mysql_create_frm is pulled out of rea_create_table for use by mysql_create_index and mysql_drop_index after the index is created/dropped. Thanks to Antony who supplied most of the changes.
2004-03-30 19:22:14 +02:00
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool mysql_alter_table(THD *thd,char *new_db, char *new_name,
BUG#19309: Crash if double procedural alter
2006-06-15 20:03:17 +02:00
HA_CREATE_INFO *lex_create_info,
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
TABLE_LIST *table_list,
List<create_field> &fields, List<Key> &keys,
Fix for bug#18437 "Wrong values inserted with a before update trigger on NDB table". SQL-layer was not marking fields which were used in triggers as such. As result these fields were not always properly retrieved/stored by handler layer. So one might got wrong values or lost changes in triggers for NDB, Federated and possibly InnoDB tables. This fix solves the problem by marking fields used in triggers appropriately. Also this patch contains the following cleanup of ha_ndbcluster code: We no longer rely on reading LEX::sql_command value in handler in order to determine if we can enable optimization which allows us to handle REPLACE statement in more efficient way by doing replaces directly in write_row() method without reporting error to SQL-layer. Instead we rely on SQL-layer informing us whether this optimization applicable by calling handler::extra() method with HA_EXTRA_WRITE_CAN_REPLACE flag. As result we no longer apply this optimzation in cases when it should not be used (e.g. if we have on delete triggers on table) and use in some additional cases when it is applicable (e.g. for LOAD DATA REPLACE). Finally this patch includes fix for bug#20728 "REPLACE does not work correctly for NDB table with PK and unique index". This was yet another problem which was caused by improper field mark-up. During row replacement fields which weren't explicity used in REPLACE statement were not marked as fields to be saved (updated) so they have retained values from old row version. The fix is to mark all table fields as set for REPLACE statement. Note that in 5.1 we already solve this problem by notifying handler that it should save values from all fields only in case when real replacement happens.
2006-07-01 23:51:10 +02:00
uint order_num, ORDER *order, bool ignore,
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
ALTER_INFO *alter_info, bool do_send_ok)
Import changeset
2000-07-31 21:29:14 +02:00
{
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
TABLE *table,*new_table=0;
After merge fixes: - Remove compiler warnings - Fix merge errors - Indentation fixes
2006-12-15 05:21:15 +01:00
int error= 0;
Some small portability fixes. Added support for lower_case_table_names=2, which is to be used on case insensitive file systems. This tells MySQL to preserve the used case of filenames and database names to make it esier to move files between cases sensitive can case insensitive file systems (like Windows and Linux)
2003-12-30 12:14:21 +01:00
char tmp_name[80],old_name[32],new_name_buff[FN_REFLEN];
char new_alias_buff[FN_REFLEN], *table_name, *db, *new_alias, *alias;
bug#24667 After ALTER TABLE operation ndb_dd table becomes regular ndb: copy tablespace from old table in copying alter table
2006-12-19 15:53:46 +01:00
char index_file[FN_REFLEN], data_file[FN_REFLEN], tablespace[FN_LEN];
Fixes for fast alter table for ndbcluster
2006-04-11 14:06:32 +02:00
char path[FN_REFLEN];
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
char reg_path[FN_REFLEN+1];
fixed coredump in UDF added Monty's patch for alter table and LAST_INSERT_ID() added a test case for replication of ALTER TABLE on a table with auto_increment
2000-08-07 23:28:56 +02:00
ha_rows copied,deleted;
WL#1266 "Separate auto-set logic from TIMESTAMP type." Final version of patch. Adds support for specifying of DEFAULT NOW() and/or ON UPDATE NOW() clauses for TIMESTAMP field definition. Current implementation allows only one such field per table and uses several unireg types for storing info about this properties of field. It should be replaced with better implementation when new .frm format is introduced.
2004-04-02 08:12:53 +02:00
uint db_create_options, used_fields;
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
handlerton *old_db_type, *new_db_type, *save_old_db_type;
sql_rename.cc, sql_table.cc, type_decimal.result, mix2_myisam.result: merge fixes
2006-10-20 10:12:38 +02:00
legacy_db_type table_type;
BUG#19309: Crash if double procedural alter
2006-06-15 20:03:17 +02:00
HA_CREATE_INFO *create_info;
Bug#14959: ALTER TABLE isn't able to rename a view The mysql_alter_table() was able to rename only a table. The view/table renaming code is moved from the function rename_tables to the new function called do_rename(). The mysql_alter_table() function calls it when it needs to rename a view.
2006-10-13 19:59:52 +02:00
frm_type_enum frm_type;
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
uint need_copy_table= 0;
BUG#17001 - Table and server crash on ALTER TABLE frm and data files for tables created by earlier MySQL versions becomes out of sync after certain ALTER TABLE statements: - One that changes column default value; - One that changes table comment; - One that changes table password. As a result one can expirience either server crash or data corruption/loss. This fix ensures that running ALTER TABLE on tables created by earlier MySQL versions recreates data files.
2006-05-16 15:37:23 +02:00
bool no_table_reopen= FALSE, varchar= FALSE;
Make storage engines "pluggable", handlerton work
2005-11-07 16:25:06 +01:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
uint fast_alter_partition= 0;
Fixed proper merge of partitioning with WL #2301 ALTER TABLE optimisations
2005-07-27 22:56:28 +02:00
bool partition_changed= FALSE;
#endif
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
List<create_field> prepared_create_list;
List<Key> prepared_key_list;
bool need_lock_for_indexes= TRUE;
uint db_options= 0;
uint key_count;
KEY *key_info_buffer;
uint index_drop_count;
uint *index_drop_buffer;
uint index_add_count;
uint *index_add_buffer;
Additional fixes for WL#1892
2006-01-17 12:53:49 +01:00
bool committed= 0;
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_ENTER("mysql_alter_table");
Fixed compiler and valgrind warnings Added missing DBUG_xxx_RETURN statements Fixed some usage of not initialized variables (as found by valgrind) Ensure that we don't remove locked tables used as name locks from open table cache until unlock_table_names() are called. This was fixed by having drop_locked_name() returning any table used as a name lock so that we can free it in unlock_table_names() This will allow Tomas to continue with his work to use namelocks to syncronize things. Note: valgrind still produces a lot of warnings about using not initialized code and shows memory loss errors when running the ndb tests
2006-03-29 13:27:36 +02:00
LINT_INIT(index_add_count);
LINT_INIT(index_drop_count);
LINT_INIT(index_add_buffer);
LINT_INIT(index_drop_buffer);
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
if (table_list && table_list->db && table_list->table_name)
Fix Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on" Log tables rely on concurrent insert machinery to add data. This means that log tables are always opened and locked by special (artificial) logger threads. Because of this, the thread which tries to drop a log table starts to wait for the table to be unlocked. Which will happen only if the log table is disabled. Alike situation happens if one tries to alter a log table. However in addition to the problem above, alter table calls check_if_locking_is_allowed() routine for the engine. The routine does not allow alter for the log tables. So, alter doesn't start waiting forever for logs to be disabled, but returns with an error. Another problem is that not all engines could be used for the log tables. That's because they need concurrent insert. In this patch we: (1) Explicitly disallow to drop/alter a log table if it is currently used by the logger. (2) Update MyISAM to support log tables (3) Allow to drop log tables/alter log tables if log is disabled At the same time we (4) Disallow to alter log tables to unsupported engine (after this patch CSV and MyISAM are alowed) Recommit with review fixes.
2006-08-03 19:28:15 +02:00
{
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
int table_kind= 0;
Fix Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on" Log tables rely on concurrent insert machinery to add data. This means that log tables are always opened and locked by special (artificial) logger threads. Because of this, the thread which tries to drop a log table starts to wait for the table to be unlocked. Which will happen only if the log table is disabled. Alike situation happens if one tries to alter a log table. However in addition to the problem above, alter table calls check_if_locking_is_allowed() routine for the engine. The routine does not allow alter for the log tables. So, alter doesn't start waiting forever for logs to be disabled, but returns with an error. Another problem is that not all engines could be used for the log tables. That's because they need concurrent insert. In this patch we: (1) Explicitly disallow to drop/alter a log table if it is currently used by the logger. (2) Update MyISAM to support log tables (3) Allow to drop log tables/alter log tables if log is disabled At the same time we (4) Disallow to alter log tables to unsupported engine (after this patch CSV and MyISAM are alowed) Recommit with review fixes.
2006-08-03 19:28:15 +02:00
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
table_kind= check_if_log_table(table_list->db_length, table_list->db,
table_list->table_name_length,
table_list->table_name, 0);
Fix Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on" Log tables rely on concurrent insert machinery to add data. This means that log tables are always opened and locked by special (artificial) logger threads. Because of this, the thread which tries to drop a log table starts to wait for the table to be unlocked. Which will happen only if the log table is disabled. Alike situation happens if one tries to alter a log table. However in addition to the problem above, alter table calls check_if_locking_is_allowed() routine for the engine. The routine does not allow alter for the log tables. So, alter doesn't start waiting forever for logs to be disabled, but returns with an error. Another problem is that not all engines could be used for the log tables. That's because they need concurrent insert. In this patch we: (1) Explicitly disallow to drop/alter a log table if it is currently used by the logger. (2) Update MyISAM to support log tables (3) Allow to drop log tables/alter log tables if log is disabled At the same time we (4) Disallow to alter log tables to unsupported engine (after this patch CSV and MyISAM are alowed) Recommit with review fixes.
2006-08-03 19:28:15 +02:00
/* Disable alter of enabled log tables */
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
if (table_kind && logger.is_log_table_enabled(table_kind))
Fix Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on" Log tables rely on concurrent insert machinery to add data. This means that log tables are always opened and locked by special (artificial) logger threads. Because of this, the thread which tries to drop a log table starts to wait for the table to be unlocked. Which will happen only if the log table is disabled. Alike situation happens if one tries to alter a log table. However in addition to the problem above, alter table calls check_if_locking_is_allowed() routine for the engine. The routine does not allow alter for the log tables. So, alter doesn't start waiting forever for logs to be disabled, but returns with an error. Another problem is that not all engines could be used for the log tables. That's because they need concurrent insert. In this patch we: (1) Explicitly disallow to drop/alter a log table if it is currently used by the logger. (2) Update MyISAM to support log tables (3) Allow to drop log tables/alter log tables if log is disabled At the same time we (4) Disallow to alter log tables to unsupported engine (after this patch CSV and MyISAM are alowed) Recommit with review fixes.
2006-08-03 19:28:15 +02:00
{
Remove unused error messages (no release contains them at the moment, so we can safely do that). Update an error mesage to make it translateable.
2006-09-27 15:48:00 +02:00
my_error(ER_BAD_LOG_STATEMENT, MYF(0), "ALTER");
Fix Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on" Log tables rely on concurrent insert machinery to add data. This means that log tables are always opened and locked by special (artificial) logger threads. Because of this, the thread which tries to drop a log table starts to wait for the table to be unlocked. Which will happen only if the log table is disabled. Alike situation happens if one tries to alter a log table. However in addition to the problem above, alter table calls check_if_locking_is_allowed() routine for the engine. The routine does not allow alter for the log tables. So, alter doesn't start waiting forever for logs to be disabled, but returns with an error. Another problem is that not all engines could be used for the log tables. That's because they need concurrent insert. In this patch we: (1) Explicitly disallow to drop/alter a log table if it is currently used by the logger. (2) Update MyISAM to support log tables (3) Allow to drop log tables/alter log tables if log is disabled At the same time we (4) Disallow to alter log tables to unsupported engine (after this patch CSV and MyISAM are alowed) Recommit with review fixes.
2006-08-03 19:28:15 +02:00
DBUG_RETURN(TRUE);
}
/* Disable alter of log tables to unsupported engine */
Fix for Bug #17544 "Cannot do atomic log rotate", Bug #21785 "Server crashes after rename of the log table" and Bug #21966 "Strange warnings on create like/repair of the log tables" According to the patch, from now on, one should use RENAME to perform a log table rotation (this should also be reflected in the manual). Here is a sample: use mysql; CREATE TABLE IF NOT EXISTS general_log2 LIKE general_log; RENAME TABLE general_log TO general_log_backup, general_log2 TO general_log; The rules for Rename of the log tables are following: IF 1. Log tables are enabled AND 2. Rename operates on the log table and nothing is being renamed to the log table. DO 3. Throw an error message. ELSE 4. Perform rename. The very RENAME query will go the the old (backup) table. This is consistent with the behavoiur we have with binlog ROTATE LOGS statement. Other problems, which are solved by the patch are: 1) Now REPAIR of the log table is exclusive operation (as it should be), this also eliminates lock-related warnings. and 2) CREATE LIKE TABLE now usese usual read lock on the source table rather then name lock, which is too restrictive. This way we get rid of another log table-related warning, which occured because of the above fact (as a side-effect, name lock resulted in a warning).
2006-10-13 15:26:46 +02:00
if (table_kind &&
Fix Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on" Log tables rely on concurrent insert machinery to add data. This means that log tables are always opened and locked by special (artificial) logger threads. Because of this, the thread which tries to drop a log table starts to wait for the table to be unlocked. Which will happen only if the log table is disabled. Alike situation happens if one tries to alter a log table. However in addition to the problem above, alter table calls check_if_locking_is_allowed() routine for the engine. The routine does not allow alter for the log tables. So, alter doesn't start waiting forever for logs to be disabled, but returns with an error. Another problem is that not all engines could be used for the log tables. That's because they need concurrent insert. In this patch we: (1) Explicitly disallow to drop/alter a log table if it is currently used by the logger. (2) Update MyISAM to support log tables (3) Allow to drop log tables/alter log tables if log is disabled At the same time we (4) Disallow to alter log tables to unsupported engine (after this patch CSV and MyISAM are alowed) Recommit with review fixes.
2006-08-03 19:28:15 +02:00
(lex_create_info->used_fields & HA_CREATE_USED_ENGINE) &&
Fix windows pushbuild failure: the bug occured because we didn't check for NULL value of the lex_create_info->db_type pointer. The pointer is NULL in the case, when the engine name is unknown to the server. This happens with NDB on Windows.
2006-08-08 10:03:24 +02:00
(!lex_create_info->db_type || /* unknown engine */
Post-review fixes for Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on": 1) Add more generic error messages 2) Add new handlerton flag for engines, which support log tables 3) Remove (log-tables related) mutex lock in myisam to improve performance
2006-09-15 11:07:23 +02:00
!(lex_create_info->db_type->flags & HTON_SUPPORT_LOG_TABLES)))
Fix Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on" Log tables rely on concurrent insert machinery to add data. This means that log tables are always opened and locked by special (artificial) logger threads. Because of this, the thread which tries to drop a log table starts to wait for the table to be unlocked. Which will happen only if the log table is disabled. Alike situation happens if one tries to alter a log table. However in addition to the problem above, alter table calls check_if_locking_is_allowed() routine for the engine. The routine does not allow alter for the log tables. So, alter doesn't start waiting forever for logs to be disabled, but returns with an error. Another problem is that not all engines could be used for the log tables. That's because they need concurrent insert. In this patch we: (1) Explicitly disallow to drop/alter a log table if it is currently used by the logger. (2) Update MyISAM to support log tables (3) Allow to drop log tables/alter log tables if log is disabled At the same time we (4) Disallow to alter log tables to unsupported engine (after this patch CSV and MyISAM are alowed) Recommit with review fixes.
2006-08-03 19:28:15 +02:00
{
Post-review fixes for Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on": 1) Add more generic error messages 2) Add new handlerton flag for engines, which support log tables 3) Remove (log-tables related) mutex lock in myisam to improve performance
2006-09-15 11:07:23 +02:00
my_error(ER_UNSUPORTED_LOG_ENGINE, MYF(0));
Fix Bug #18559 "log tables cannot change engine, and gets deadlocked when dropping w/ log on" Log tables rely on concurrent insert machinery to add data. This means that log tables are always opened and locked by special (artificial) logger threads. Because of this, the thread which tries to drop a log table starts to wait for the table to be unlocked. Which will happen only if the log table is disabled. Alike situation happens if one tries to alter a log table. However in addition to the problem above, alter table calls check_if_locking_is_allowed() routine for the engine. The routine does not allow alter for the log tables. So, alter doesn't start waiting forever for logs to be disabled, but returns with an error. Another problem is that not all engines could be used for the log tables. That's because they need concurrent insert. In this patch we: (1) Explicitly disallow to drop/alter a log table if it is currently used by the logger. (2) Update MyISAM to support log tables (3) Allow to drop log tables/alter log tables if log is disabled At the same time we (4) Disallow to alter log tables to unsupported engine (after this patch CSV and MyISAM are alowed) Recommit with review fixes.
2006-08-03 19:28:15 +02:00
DBUG_RETURN(TRUE);
}
}
Import changeset
2000-07-31 21:29:14 +02:00
thd->proc_info="init";
BUG#19309: Crash if double procedural alter
2006-06-15 20:03:17 +02:00
if (!(create_info= copy_create_info(lex_create_info)))
{
DBUG_RETURN(TRUE);
}
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
table_name=table_list->table_name;
Some small portability fixes. Added support for lower_case_table_names=2, which is to be used on case insensitive file systems. This tells MySQL to preserve the used case of filenames and database names to make it esier to move files between cases sensitive can case insensitive file systems (like Windows and Linux)
2003-12-30 12:14:21 +01:00
alias= (lower_case_table_names == 2) ? table_list->alias : table_name;
Import changeset
2000-07-31 21:29:14 +02:00
db=table_list->db;
Merge with 4.0.18
2004-02-11 00:06:46 +01:00
if (!new_db || !my_strcasecmp(table_alias_charset, new_db, db))
follow-up to bug #2628: attempt to make alter table rename a bit more efficient in case of lower_case_table_names.
2004-02-09 12:29:31 +01:00
new_db= db;
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
build_table_filename(reg_path, sizeof(reg_path), db, table_name, reg_ext, 0);
build_table_filename(path, sizeof(path), db, table_name, "", 0);
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
Fixed that SHOW CREATE TABLE shows all attributes
2001-09-30 04:47:35 +02:00
used_fields=create_info->used_fields;
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
Bug#14397 - OPTIMIZE TABLE with an open HANDLER causes a crash Version for 5.0. It fixes three problems: 1. The cause of the bug was that we did not check the table version for the HANDLER ... READ commands. We did not notice when a table was replaced by a new one. This can happen during ALTER TABLE, REPAIR TABLE, and OPTIMIZE TABLE (there might be more cases). I call the fix for this problem "the primary bug fix". 2. mysql_ha_flush() was not always called with a locked LOCK_open. Though the function comment clearly said it must. I changed the code so that the locking is done when required. I call the fix for this problem "the secondary fix". 3. In 5.0 (not in 4.1 or 4.0) DROP TABLE had a possible deadlock flaw in concur with FLUSH TABLES WITH READ LOCK. I call the fix for this problem "the 5.0 addendum fix".
2005-11-15 21:57:02 +01:00
mysql_ha_flush(thd, table_list, MYSQL_HA_CLOSE_FINAL, FALSE);
Import changeset
2000-07-31 21:29:14 +02:00
sql_yacc.yy: Add DISCARD etc. in the keyword: rule of sql_yacc.yy sql_table.cc: Add comment that DISCARD/IMPORT TABLESPACE is always the only clause in an ALTER TABLE
2003-10-14 00:52:03 +02:00
/* DISCARD/IMPORT TABLESPACE is always alone in an ALTER TABLE */
Introduced parameter object "ALTER_INFO" for mysql_alter_table to make list of parameters in mysql_alter_table shorted to avoid warning in MSVC (windows) building
2004-05-21 16:57:03 +02:00
if (alter_info->tablespace_op != NO_TABLESPACE_OP)
Bug #14262: SP: DROP PROCEDURE|VIEW (maybe more) write to binlog too late \ (race cond) It was possible for one thread to interrupt a Data Definition Language statement and thereby get messages to the binlog out of order. Consider: Connection 1: Drop Foo x Connection 2: Create or replace Foo x Connection 2: Log "Create or replace Foo x" Connection 1: Log "Drop Foo x" Local end would have Foo x, but the replicated slaves would not. The fix for this is to wrap all DDL and logging of a kind in the same mutex. Since we already use mutexes for the various parts of altering the server, this only entails moving the logging events down close to the action, inside the mutex protection.
2006-10-03 19:38:25 +02:00
/* Conditionally writes to binlog. */
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
DBUG_RETURN(mysql_discard_or_import_tablespace(thd,table_list,
Introduced parameter object "ALTER_INFO" for mysql_alter_table to make list of parameters in mysql_alter_table shorted to avoid warning in MSVC (windows) building
2004-05-21 16:57:03 +02:00
alter_info->tablespace_op));
sql_rename.cc, sql_table.cc, type_decimal.result, mix2_myisam.result: merge fixes
2006-10-20 10:12:38 +02:00
strxnmov(new_name_buff, sizeof (new_name_buff) - 1, mysql_data_home, "/", db,
"/", table_name, reg_ext, NullS);
(void) unpack_filename(new_name_buff, new_name_buff);
Bug#14959: ALTER TABLE isn't able to rename a view The mysql_alter_table() was able to rename only a table. The view/table renaming code is moved from the function rename_tables to the new function called do_rename(). The mysql_alter_table() function calls it when it needs to rename a view.
2006-10-13 19:59:52 +02:00
if (lower_case_table_names != 2)
my_casedn_str(files_charset_info, new_name_buff);
frm_type= mysql_frm_type(thd, new_name_buff, &table_type);
/* Rename a view */
if (frm_type == FRMTYPE_VIEW && !(alter_info->flags & ~ALTER_RENAME))
{
/*
Avoid problems with a rename on a table that we have locked or
if the user is trying to to do this in a transcation context
*/
if (thd->locked_tables || thd->active_transaction())
{
my_message(ER_LOCK_OR_ACTIVE_TRANSACTION,
ER(ER_LOCK_OR_ACTIVE_TRANSACTION), MYF(0));
Bug#25183 (Compiler warnings in the sql directory (GCC 4.1)) Fixed compiler warnings. In set_var.cc, the code was not properly returning an error code if close_cached_tables() failed. In sql_tables.cc, the code was not returning properly an error code if lock_table_names() failed. Both cases are bugs, introduced in 5.1 only by recent changes.
2006-12-19 23:02:37 +01:00
DBUG_RETURN(TRUE);
Bug#14959: ALTER TABLE isn't able to rename a view The mysql_alter_table() was able to rename only a table. The view/table renaming code is moved from the function rename_tables to the new function called do_rename(). The mysql_alter_table() function calls it when it needs to rename a view.
2006-10-13 19:59:52 +02:00
}
if (wait_if_global_read_lock(thd,0,1))
Bug#25183 (Compiler warnings in the sql directory (GCC 4.1)) Fixed compiler warnings. In set_var.cc, the code was not properly returning an error code if close_cached_tables() failed. In sql_tables.cc, the code was not returning properly an error code if lock_table_names() failed. Both cases are bugs, introduced in 5.1 only by recent changes.
2006-12-19 23:02:37 +01:00
DBUG_RETURN(TRUE);
Bug#14959: ALTER TABLE isn't able to rename a view The mysql_alter_table() was able to rename only a table. The view/table renaming code is moved from the function rename_tables to the new function called do_rename(). The mysql_alter_table() function calls it when it needs to rename a view.
2006-10-13 19:59:52 +02:00
VOID(pthread_mutex_lock(&LOCK_open));
if (lock_table_names(thd, table_list))
Bug#25183 (Compiler warnings in the sql directory (GCC 4.1)) Fixed compiler warnings. In set_var.cc, the code was not properly returning an error code if close_cached_tables() failed. In sql_tables.cc, the code was not returning properly an error code if lock_table_names() failed. Both cases are bugs, introduced in 5.1 only by recent changes.
2006-12-19 23:02:37 +01:00
{
After merge fixes: - Remove compiler warnings - Fix merge errors - Indentation fixes
2006-12-15 05:21:15 +01:00
error= 1;
Bug#14959: ALTER TABLE isn't able to rename a view The mysql_alter_table() was able to rename only a table. The view/table renaming code is moved from the function rename_tables to the new function called do_rename(). The mysql_alter_table() function calls it when it needs to rename a view.
2006-10-13 19:59:52 +02:00
goto view_err;
Bug#25183 (Compiler warnings in the sql directory (GCC 4.1)) Fixed compiler warnings. In set_var.cc, the code was not properly returning an error code if close_cached_tables() failed. In sql_tables.cc, the code was not returning properly an error code if lock_table_names() failed. Both cases are bugs, introduced in 5.1 only by recent changes.
2006-12-19 23:02:37 +01:00
}
Bug#14959: ALTER TABLE isn't able to rename a view The mysql_alter_table() was able to rename only a table. The view/table renaming code is moved from the function rename_tables to the new function called do_rename(). The mysql_alter_table() function calls it when it needs to rename a view.
2006-10-13 19:59:52 +02:00
if (!do_rename(thd, table_list, new_db, new_name, new_name, 1))
{
if (mysql_bin_log.is_open())
{
thd->clear_error();
Query_log_event qinfo(thd, thd->query, thd->query_length, 0, FALSE);
mysql_bin_log.write(&qinfo);
}
send_ok(thd);
}
unlock_table_names(thd, table_list, (TABLE_LIST*) 0);
view_err:
pthread_mutex_unlock(&LOCK_open);
start_waiting_global_read_lock(thd);
DBUG_RETURN(error);
}
Import changeset
2000-07-31 21:29:14 +02:00
if (!(table=open_ltable(thd,table_list,TL_WRITE_ALLOW_READ)))
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
table->use_all_columns();
Import changeset
2000-07-31 21:29:14 +02:00
/* Check that we are not trying to rename to an existing table */
if (new_name)
{
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
DBUG_PRINT("info", ("new_db.new_name: '%s'.'%s'", new_db, new_name));
Import changeset
2000-07-31 21:29:14 +02:00
strmov(new_name_buff,new_name);
3.23 -> 4.0 merge: after merge fixes
2004-02-06 14:40:44 +01:00
strmov(new_alias= new_alias_buff, new_name);
Portability fixes Added record_rnd_buffer Added --safe-user-create Fix for ALTER TABLE RENAME on windows
2001-08-10 16:37:37 +02:00
if (lower_case_table_names)
Some small portability fixes. Added support for lower_case_table_names=2, which is to be used on case insensitive file systems. This tells MySQL to preserve the used case of filenames and database names to make it esier to move files between cases sensitive can case insensitive file systems (like Windows and Linux)
2003-12-30 12:14:21 +01:00
{
if (lower_case_table_names != 2)
{
Fix for bug which caused grant.test fail on darwin7.3. We were converting db and table names to lower case using latin1 instead of utf-8 in sql_acl.cc if lower_case_table_names was on. Also replaced in other such places system_charset_info with files_charset_info for consistency.
2004-05-22 21:41:58 +02:00
my_casedn_str(files_charset_info, new_name_buff);
Some small portability fixes. Added support for lower_case_table_names=2, which is to be used on case insensitive file systems. This tells MySQL to preserve the used case of filenames and database names to make it esier to move files between cases sensitive can case insensitive file systems (like Windows and Linux)
2003-12-30 12:14:21 +01:00
new_alias= new_name; // Create lower case table name
}
Fix for bug which caused grant.test fail on darwin7.3. We were converting db and table names to lower case using latin1 instead of utf-8 in sql_acl.cc if lower_case_table_names was on. Also replaced in other such places system_charset_info with files_charset_info for consistency.
2004-05-22 21:41:58 +02:00
my_casedn_str(files_charset_info, new_name);
Some small portability fixes. Added support for lower_case_table_names=2, which is to be used on case insensitive file systems. This tells MySQL to preserve the used case of filenames and database names to make it esier to move files between cases sensitive can case insensitive file systems (like Windows and Linux)
2003-12-30 12:14:21 +01:00
}
Fix for bug #2628 "ALTER TABLE destroys table and reports success"
2004-02-06 12:28:57 +01:00
if (new_db == db &&
Merge with 4.0.18
2004-02-11 00:06:46 +01:00
!my_strcasecmp(table_alias_charset, new_name_buff, table_name))
Fix for bug #2628 "ALTER TABLE destroys table and reports success"
2004-02-06 12:28:57 +01:00
{
/*
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
Source and destination table names are equal: make later check
easier.
Fix for bug #2628 "ALTER TABLE destroys table and reports success"
2004-02-06 12:28:57 +01:00
*/
merge 3.23 -> 4.0, 2003/02/06
2004-02-06 13:23:41 +01:00
new_alias= new_name= table_name;
Fix for bug #2628 "ALTER TABLE destroys table and reports success"
2004-02-06 12:28:57 +01:00
}
Import changeset
2000-07-31 21:29:14 +02:00
else
{
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
if (table->s->tmp_table != NO_TMP_TABLE)
Import changeset
2000-07-31 21:29:14 +02:00
{
if (find_temporary_table(thd,new_db,new_name_buff))
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_TABLE_EXISTS_ERROR, MYF(0), new_name_buff);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
Import changeset
2000-07-31 21:29:14 +02:00
}
}
else
{
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
build_table_filename(new_name_buff, sizeof(new_name_buff),
new_db, new_name_buff, reg_ext, 0);
if (!access(new_name_buff, F_OK))
Import changeset
2000-07-31 21:29:14 +02:00
{
/* Table will be closed in do_command() */
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_TABLE_EXISTS_ERROR, MYF(0), new_alias);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
Import changeset
2000-07-31 21:29:14 +02:00
}
}
}
}
else
lower_case_table_names=2 (Keep case for table names) was not honored with ALTER TABLE and CREATE/DROP INDEX. (Bug #3109) Make net_buffer_length visible for mysql clients (Bug #4206)
2004-06-23 15:44:34 +02:00
{
new_alias= (lower_case_table_names == 2) ? alias : table_name;
new_name= table_name;
}
Import changeset
2000-07-31 21:29:14 +02:00
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
old_db_type= table->s->db_type;
handlerton cleanup: duplicate fields removed, st_mysql_storage_engine added to support run-time handlerton initialization (no compiler warnings), handler API is now tied to MySQL version, handlerton->plugin mapping added (slot-based), dummy default_hton removed, plugin-type-specific initialization generalized, built-in plugins are now initialized too, --default-storage-engine no longer needs a list of storage engines in handle_options(). mysql-test-run.pl bugfixes
2006-05-28 14:51:01 +02:00
if (!create_info->db_type)
BUG#20397: Crash at ALTER TABLE t1 engine = x; for partitioned table
2006-06-14 04:46:38 +02:00
{
Fix for broken compile if not compiling with partitioning
2006-06-16 21:04:35 +02:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
BUG#20397: Crash when alter table t1 engine = x; Review fixes
2006-06-15 21:56:47 +02:00
if (table->part_info &&
create_info->used_fields & HA_CREATE_USED_ENGINE)
BUG#20397: Crash at ALTER TABLE t1 engine = x; for partitioned table
2006-06-14 04:46:38 +02:00
{
/*
This case happens when the user specified
ENGINE = x where x is a non-existing storage engine
BUG#20397: Crash when alter table t1 engine = x; Review fixes
2006-06-15 21:56:47 +02:00
We set create_info->db_type to default_engine_type
to ensure we don't change underlying engine type
due to a erroneously given engine name.
BUG#20397: Crash at ALTER TABLE t1 engine = x; for partitioned table
2006-06-14 04:46:38 +02:00
*/
BUG#20397: Crash when alter table t1 engine = x; Review fixes
2006-06-15 21:56:47 +02:00
create_info->db_type= table->part_info->default_engine_type;
BUG#20397: Crash at ALTER TABLE t1 engine = x; for partitioned table
2006-06-14 04:46:38 +02:00
}
BUG#20397: Crash when alter table t1 engine = x; Review fixes
2006-06-15 21:56:47 +02:00
else
Fix for broken compile if not compiling with partitioning
2006-06-16 21:04:35 +02:00
#endif
BUG#20397: Crash when alter table t1 engine = x; Review fixes
2006-06-15 21:56:47 +02:00
create_info->db_type= old_db_type;
BUG#20397: Crash at ALTER TABLE t1 engine = x; for partitioned table
2006-06-14 04:46:38 +02:00
}
WL #2602, #2603, #2604 Added new syntax for partition management
2005-08-19 16:26:05 +02:00
Make storage engines "pluggable", handlerton work
2005-11-07 16:25:06 +01:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
if (prep_alter_part_table(thd, table, alter_info, create_info, old_db_type,
&partition_changed, &fast_alter_partition))
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
{
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
DBUG_RETURN(TRUE);
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
}
#endif
Bug #17210 Create temp table call to ha_ndbcluster::create_handler_files caused core - do not try to create ndb temp tables even if default storage is ndb
2006-02-17 17:12:35 +01:00
if (check_engine(thd, new_name, create_info))
Bug#6877 MySQL should give an error if the requested table type is not available Implement new SQL mode - NO_ENGINE_SUBSTITUTION
2005-06-17 23:14:44 +02:00
DBUG_RETURN(TRUE);
new_db_type= create_info->db_type;
Fixed that SHOW CREATE TABLE shows all attributes
2001-09-30 04:47:35 +02:00
if (create_info->row_type == ROW_TYPE_NOT_USED)
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
create_info->row_type= table->s->row_type;
Import changeset
2000-07-31 21:29:14 +02:00
WL #2747: Fix such that backup and restore works for user defined partitioned tables in NDB
2006-01-17 09:25:12 +01:00
DBUG_PRINT("info", ("old type: %s new type: %s",
ha_resolve_storage_engine_name(old_db_type),
ha_resolve_storage_engine_name(new_db_type)));
Reviewing new pushed code - CHAR() now returns binary string as default - CHAR(X*65536+Y*256+Z) is now equal to CHAR(X,Y,Z) independent of the character set for CHAR() - Test for both ETIMEDOUT and ETIME from pthread_cond_timedwait() (Some old systems returns ETIME and it's safer to test for both values than to try to write a wrapper for each old system) - Fixed new introduced bug in NOT BETWEEN X and X - Ensure we call commit_by_xid or rollback_by_xid for all engines, even if one engine has failed - Use octet2hex() for all conversion of string to hex - Simplify and optimize code
2005-10-11 23:58:22 +02:00
if (ha_check_storage_engine_flag(old_db_type, HTON_ALTER_NOT_SUPPORTED) ||
Revoking patch for Bug#10952 on behalf of Brian.
2006-07-10 20:46:05 +02:00
ha_check_storage_engine_flag(new_db_type, HTON_ALTER_NOT_SUPPORTED))
Move handler specific options into handlerton flag check BUG#13108
2005-10-01 01:26:48 +02:00
{
DBUG_PRINT("info", ("doesn't support alter"));
my_error(ER_ILLEGAL_HA, MYF(0), table_name);
DBUG_RETURN(TRUE);
}
Import changeset
2000-07-31 21:29:14 +02:00
thd->proc_info="setup";
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
if (!(alter_info->flags & ~(ALTER_RENAME | ALTER_KEYS_ONOFF)) &&
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
!table->s->tmp_table) // no need to touch frm
Import changeset
2000-07-31 21:29:14 +02:00
{
Bug #14262: SP: DROP PROCEDURE|VIEW (maybe more) write to binlog too late \ (race cond) It was possible for one thread to interrupt a Data Definition Language statement and thereby get messages to the binlog out of order. Consider: Connection 1: Drop Foo x Connection 2: Create or replace Foo x Connection 2: Log "Create or replace Foo x" Connection 1: Log "Drop Foo x" Local end would have Foo x, but the replicated slaves would not. The fix for this is to wrap all DDL and logging of a kind in the same mutex. Since we already use mutexes for the various parts of altering the server, this only entails moving the logging events down close to the action, inside the mutex protection.
2006-10-03 19:38:25 +02:00
VOID(pthread_mutex_lock(&LOCK_open));
Fix for bug#24219 ALTER TABLE ... RENAME TO ... , DISABLE KEYS leads to crash (this is the 5.0 patch, because 4.1 differs) There was an improper order of doing chained operations. To the documentor: ENABLE|DISABLE KEYS combined with RENAME TO, and no other ALTER TABLE clause, leads to server crash independent of the presence of indices and data in the table.
2006-11-16 14:01:51 +01:00
switch (alter_info->keys_onoff) {
case LEAVE_AS_IS:
break;
case ENABLE:
wait_while_table_is_used(thd, table, HA_EXTRA_FORCE_REOPEN);
error= table->file->enable_indexes(HA_KEY_SWITCH_NONUNIQ_SAVE);
/* COND_refresh will be signaled in close_thread_tables() */
break;
case DISABLE:
wait_while_table_is_used(thd, table, HA_EXTRA_FORCE_REOPEN);
error=table->file->disable_indexes(HA_KEY_SWITCH_NONUNIQ_SAVE);
/* COND_refresh will be signaled in close_thread_tables() */
break;
Bug#25183 (Compiler warnings in the sql directory (GCC 4.1)) Fixed compiler warnings. In set_var.cc, the code was not properly returning an error code if close_cached_tables() failed. In sql_tables.cc, the code was not returning properly an error code if lock_table_names() failed. Both cases are bugs, introduced in 5.1 only by recent changes.
2006-12-19 23:02:37 +01:00
default:
DBUG_ASSERT(FALSE);
error= 0;
break;
Fix for bug#24219 ALTER TABLE ... RENAME TO ... , DISABLE KEYS leads to crash (this is the 5.0 patch, because 4.1 differs) There was an improper order of doing chained operations. To the documentor: ENABLE|DISABLE KEYS combined with RENAME TO, and no other ALTER TABLE clause, leads to server crash independent of the presence of indices and data in the table.
2006-11-16 14:01:51 +01:00
}
if (error == HA_ERR_WRONG_COMMAND)
{
After merge fixes: - Remove compiler warnings - Fix merge errors - Indentation fixes
2006-12-15 05:21:15 +01:00
error= 0;
Fix for bug#24219 ALTER TABLE ... RENAME TO ... , DISABLE KEYS leads to crash (this is the 5.0 patch, because 4.1 differs) There was an improper order of doing chained operations. To the documentor: ENABLE|DISABLE KEYS combined with RENAME TO, and no other ALTER TABLE clause, leads to server crash independent of the presence of indices and data in the table.
2006-11-16 14:01:51 +01:00
push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_NOTE,
ER_ILLEGAL_HA, ER(ER_ILLEGAL_HA),
table->alias);
}
if (!error && (new_name != table_name || new_db != db))
Import changeset
2000-07-31 21:29:14 +02:00
{
ALTER TABLE ... DISABLE KEYS / ALTER TABLE ... ENABLE KEYS
2001-05-11 22:26:12 +02:00
thd->proc_info="rename";
/* Then do a 'simple' rename of the table */
if (!access(new_name_buff,F_OK))
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_TABLE_EXISTS_ERROR, MYF(0), new_name);
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
error= -1;
ALTER TABLE ... DISABLE KEYS / ALTER TABLE ... ENABLE KEYS
2001-05-11 22:26:12 +02:00
}
else
{
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
*fn_ext(new_name)=0;
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
table->s->version= 0; // Force removal of table def
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
close_cached_table(thd, table);
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
if (mysql_rename_table(old_db_type,db,table_name,new_db,new_alias, 0))
ALTER TABLE ... DISABLE KEYS / ALTER TABLE ... ENABLE KEYS
2001-05-11 22:26:12 +02:00
error= -1;
Fix for bug #13525 "Rename table does not keep info of triggers". Let us transfer triggers associated with table when we rename it (but only if we are not changing database to which table belongs, in the latter case we will emit error).
2006-02-24 21:50:36 +01:00
else if (Table_triggers_list::change_table_name(thd, db, table_name,
new_db, new_alias))
{
VOID(mysql_rename_table(old_db_type, new_db, new_alias, db,
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
table_name, 0));
Fix for bug #13525 "Rename table does not keep info of triggers". Let us transfer triggers associated with table when we rename it (but only if we are not changing database to which table belongs, in the latter case we will emit error).
2006-02-24 21:50:36 +01:00
error= -1;
}
ALTER TABLE ... DISABLE KEYS / ALTER TABLE ... ENABLE KEYS
2001-05-11 22:26:12 +02:00
}
Import changeset
2000-07-31 21:29:14 +02:00
}
Many files: ALTER TABLE ... DISCARD/IMPORT TABLESPACE Track crash in buf_LRU_block_remove_hashed_page + 1807 reported in MySQL-3.23.5x
2003-10-13 10:20:19 +02:00
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
if (error == HA_ERR_WRONG_COMMAND)
post-review fixes HA_EXTRA_WRITE_CACHE now part of start_bulk_insert() test cleanups
2004-04-07 16:04:28 +02:00
{
After merge fixes: - Remove compiler warnings - Fix merge errors - Indentation fixes
2006-12-15 05:21:15 +01:00
error= 0;
post-review fixes HA_EXTRA_WRITE_CACHE now part of start_bulk_insert() test cleanups
2004-04-07 16:04:28 +02:00
push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_NOTE,
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
ER_ILLEGAL_HA, ER(ER_ILLEGAL_HA),
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
table->alias);
post-review fixes HA_EXTRA_WRITE_CACHE now part of start_bulk_insert() test cleanups
2004-04-07 16:04:28 +02:00
}
Fix for bug#24219 ALTER TABLE ... RENAME TO ... , DISABLE KEYS leads to crash (this is the 5.0 patch, because 4.1 differs) There was an improper order of doing chained operations. To the documentor: ENABLE|DISABLE KEYS combined with RENAME TO, and no other ALTER TABLE clause, leads to server crash independent of the presence of indices and data in the table.
2006-11-16 14:01:51 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
if (!error)
{
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
write_bin_log(thd, TRUE, thd->query, thd->query_length);
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
if (do_send_ok)
send_ok(thd);
Import changeset
2000-07-31 21:29:14 +02:00
}
bug#4881 - crash in ALTER .. RENAME if rename fails
2004-08-07 18:26:59 +02:00
else if (error > 0)
::reset(), HA_FAST_KEY_READ, disable_indexes(), enable_indexes(), start_bulk_insert(), end_bulk_insert() Field::val_str simplification, comment
2004-04-06 21:35:26 +02:00
{
table->file->print_error(error, MYF(0));
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
error= -1;
::reset(), HA_FAST_KEY_READ, disable_indexes(), enable_indexes(), start_bulk_insert(), end_bulk_insert() Field::val_str simplification, comment
2004-04-06 21:35:26 +02:00
}
Bug #14262: SP: DROP PROCEDURE|VIEW (maybe more) write to binlog too late \ (race cond) It was possible for one thread to interrupt a Data Definition Language statement and thereby get messages to the binlog out of order. Consider: Connection 1: Drop Foo x Connection 2: Create or replace Foo x Connection 2: Log "Create or replace Foo x" Connection 1: Log "Drop Foo x" Local end would have Foo x, but the replicated slaves would not. The fix for this is to wrap all DDL and logging of a kind in the same mutex. Since we already use mutexes for the various parts of altering the server, this only entails moving the logging events down close to the action, inside the mutex protection.
2006-10-03 19:38:25 +02:00
VOID(pthread_mutex_unlock(&LOCK_open));
Fix for bug#24219 ALTER TABLE ... RENAME TO ... , DISABLE KEYS leads to crash (this is the 5.0 patch, because 4.1 differs) There was an improper order of doing chained operations. To the documentor: ENABLE|DISABLE KEYS combined with RENAME TO, and no other ALTER TABLE clause, leads to server crash independent of the presence of indices and data in the table.
2006-11-16 14:01:51 +01:00
table_list->table= NULL; // For query cache
fixed table invalidation in simple renaming
2003-01-24 00:54:39 +01:00
query_cache_invalidate3(thd, table_list, 0);
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(error);
}
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
/* We have to do full alter table */
Fix: create table t1 (a char(10) character set cp1251) SELECT _koi8r'blabla' as a The above query created a field of koi8r charset, not cp1251 Change: CREATE TABLE a (a CHAR(1) CHARACTER SET utf8) Length now means character length, not byte length. The above creates a field that guarantees can store a multibyte value 1 character long. For utf8 the above creates a field that can store 3 bytes.
2003-08-11 15:18:34 +02:00
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
/* Let new create options override the old ones */
Fix: create table t1 (a char(10) character set cp1251) SELECT _koi8r'blabla' as a The above query created a field of koi8r charset, not cp1251 Change: CREATE TABLE a (a CHAR(1) CHARACTER SET utf8) Length now means character length, not byte length. The above creates a field that guarantees can store a multibyte value 1 character long. For utf8 the above creates a field that can store 3 bytes.
2003-08-11 15:18:34 +02:00
if (!(used_fields & HA_CREATE_USED_MIN_ROWS))
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
create_info->min_rows= table->s->min_rows;
Fix: create table t1 (a char(10) character set cp1251) SELECT _koi8r'blabla' as a The above query created a field of koi8r charset, not cp1251 Change: CREATE TABLE a (a CHAR(1) CHARACTER SET utf8) Length now means character length, not byte length. The above creates a field that guarantees can store a multibyte value 1 character long. For utf8 the above creates a field that can store 3 bytes.
2003-08-11 15:18:34 +02:00
if (!(used_fields & HA_CREATE_USED_MAX_ROWS))
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
create_info->max_rows= table->s->max_rows;
Fix: create table t1 (a char(10) character set cp1251) SELECT _koi8r'blabla' as a The above query created a field of koi8r charset, not cp1251 Change: CREATE TABLE a (a CHAR(1) CHARACTER SET utf8) Length now means character length, not byte length. The above creates a field that guarantees can store a multibyte value 1 character long. For utf8 the above creates a field that can store 3 bytes.
2003-08-11 15:18:34 +02:00
if (!(used_fields & HA_CREATE_USED_AVG_ROW_LENGTH))
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
create_info->avg_row_length= table->s->avg_row_length;
CLIENT_MULTI_QUERIES -> CLIENT_MULTI_STATEMENTS New multi-key-cache handling. This was needed becasue the old one didn't work reliable with MERGE tables. ALTER TABLE table_name ... CHARACTER SET ... now changes all char/varchar/text columns to the given character set (One must use ALTER TABLE ... DEFAULT CHARACTER SET ... to change the default character set) Fixed that have_compress is detected properly (fixes problems with func_compress.test on platforms without zlib) New syntax for CACHE INDEX ('keys' is optional if no index name is given and one mentions the key cache name only ones) Removed compiler warnings Added mysql_set_server_option() to allow clients like PHP to easaily set/reset the multi-statement flag.
2003-11-18 12:47:27 +01:00
if (!(used_fields & HA_CREATE_USED_DEFAULT_CHARSET))
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
create_info->default_table_charset= table->s->table_charset;
Added support for key_block_size for key and table level (WL#602) Added support for key_block_size to MyISAM. Simplify interface to 'new Key' to make it easier to add new key options. mysqld option --new is used to define where key options are printed. (In 5.3 we should move all key options to after key part definition to avoid problem with reserved names) Fixed some compiler warnings and a memory leak in ssl
2006-05-03 14:59:17 +02:00
if (!(used_fields & HA_CREATE_USED_KEY_BLOCK_SIZE))
create_info->key_block_size= table->s->key_block_size;
Fix: create table t1 (a char(10) character set cp1251) SELECT _koi8r'blabla' as a The above query created a field of koi8r charset, not cp1251 Change: CREATE TABLE a (a CHAR(1) CHARACTER SET utf8) Length now means character length, not byte length. The above creates a field that guarantees can store a multibyte value 1 character long. For utf8 the above creates a field that can store 3 bytes.
2003-08-11 15:18:34 +02:00
bug#24667 After ALTER TABLE operation ndb_dd table becomes regular ndb: ALTER TABLE must specify STORAGE explicitely to change it, + post review changes
2006-12-19 23:20:43 +01:00
if (!create_info->tablespace && create_info->storage_media != HA_SM_MEMORY)
bug#24667 After ALTER TABLE operation ndb_dd table becomes regular ndb: copy tablespace from old table in copying alter table
2006-12-19 15:53:46 +01:00
{
/*
bug#24667 After ALTER TABLE operation ndb_dd table becomes regular ndb: ALTER TABLE must specify STORAGE explicitely to change it, + post review changes
2006-12-19 23:20:43 +01:00
Regular alter table of disk stored table (no tablespace/storage change)
bug#24667 After ALTER TABLE operation ndb_dd table becomes regular ndb: copy tablespace from old table in copying alter table
2006-12-19 15:53:46 +01:00
Copy tablespace name
*/
bug#24667 After ALTER TABLE operation ndb_dd table becomes regular ndb: ALTER TABLE must specify STORAGE explicitely to change it, + post review changes
2006-12-19 23:20:43 +01:00
if ((table->file->get_tablespace_name(thd, tablespace, FN_LEN)))
create_info->tablespace= tablespace;
bug#24667 After ALTER TABLE operation ndb_dd table becomes regular ndb: copy tablespace from old table in copying alter table
2006-12-19 15:53:46 +01:00
}
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
restore_record(table, s->default_values); // Empty record for DEFAULT
Introduced parameter object "ALTER_INFO" for mysql_alter_table to make list of parameters in mysql_alter_table shorted to avoid warning in MSVC (windows) building
2004-05-21 16:57:03 +02:00
List_iterator<Alter_drop> drop_it(alter_info->drop_list);
Import changeset
2000-07-31 21:29:14 +02:00
List_iterator<create_field> def_it(fields);
Introduced parameter object "ALTER_INFO" for mysql_alter_table to make list of parameters in mysql_alter_table shorted to avoid warning in MSVC (windows) building
2004-05-21 16:57:03 +02:00
List_iterator<Alter_column> alter_it(alter_info->alter_list);
Import changeset
2000-07-31 21:29:14 +02:00
List<create_field> create_list; // Add new fields here
List<Key> key_list; // Add new keys here
Fix: create table t1 (a char(10) character set cp1251) SELECT _koi8r'blabla' as a The above query created a field of koi8r charset, not cp1251 Change: CREATE TABLE a (a CHAR(1) CHARACTER SET utf8) Length now means character length, not byte length. The above creates a field that guarantees can store a multibyte value 1 character long. For utf8 the above creates a field that can store 3 bytes.
2003-08-11 15:18:34 +02:00
create_field *def;
Import changeset
2000-07-31 21:29:14 +02:00
/*
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
First collect all fields from table which isn't in drop_list
Import changeset
2000-07-31 21:29:14 +02:00
*/
Field **f_ptr,*field;
for (f_ptr=table->field ; (field= *f_ptr) ; f_ptr++)
{
BUG#17001 - Table and server crash on ALTER TABLE frm and data files for tables created by earlier MySQL versions becomes out of sync after certain ALTER TABLE statements: - One that changes column default value; - One that changes table comment; - One that changes table password. As a result one can expirience either server crash or data corruption/loss. This fix ensures that running ALTER TABLE on tables created by earlier MySQL versions recreates data files.
2006-05-16 15:37:23 +02:00
if (field->type() == MYSQL_TYPE_STRING)
varchar= TRUE;
Fix incorrect spellings of "dropped" in source and tests. (Bug #12828)
2005-08-30 21:24:37 +02:00
/* Check if field should be dropped */
Import changeset
2000-07-31 21:29:14 +02:00
Alter_drop *drop;
drop_it.rewind();
while ((drop=drop_it++))
{
if (drop->type == Alter_drop::COLUMN &&
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
!my_strcasecmp(system_charset_info,field->field_name, drop->name))
Changed compare in MyISAM to use my_pread() Fixed that @VAR shows all decimals Fixed problem with FLUSH TABLES and LOCK TABLE CREATE ... SELECT now creates keys later Reset AUTO_INCREMENT order if droping AUTO_INCREMENT key
2001-06-11 14:01:28 +02:00
{
/* Reset auto_increment value if it was dropped */
if (MTYP_TYPENR(field->unireg_check) == Field::NEXT_NUMBER &&
Fixed that SHOW CREATE TABLE shows all attributes
2001-09-30 04:47:35 +02:00
!(used_fields & HA_CREATE_USED_AUTO))
Changed compare in MyISAM to use my_pread() Fixed that @VAR shows all decimals Fixed problem with FLUSH TABLES and LOCK TABLE CREATE ... SELECT now creates keys later Reset AUTO_INCREMENT order if droping AUTO_INCREMENT key
2001-06-11 14:01:28 +02:00
{
create_info->auto_increment_value=0;
create_info->used_fields|=HA_CREATE_USED_AUTO;
}
Import changeset
2000-07-31 21:29:14 +02:00
break;
Changed compare in MyISAM to use my_pread() Fixed that @VAR shows all decimals Fixed problem with FLUSH TABLES and LOCK TABLE CREATE ... SELECT now creates keys later Reset AUTO_INCREMENT order if droping AUTO_INCREMENT key
2001-06-11 14:01:28 +02:00
}
Import changeset
2000-07-31 21:29:14 +02:00
}
if (drop)
{
drop_it.remove();
continue;
}
/* Check if field is changed */
def_it.rewind();
while ((def=def_it++))
{
better fix BUG#2361: ALTER TABLE ... DROP PRIMARY KEY drops a non-primary key
2004-01-28 18:35:52 +01:00
if (def->change &&
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
!my_strcasecmp(system_charset_info,field->field_name, def->change))
Import changeset
2000-07-31 21:29:14 +02:00
break;
}
if (def)
{ // Field is changed
def->field=field;
Fixed error message when opening a not-MyISAM file. Extended MODIFY and CHANGE in ALTER TABLE to accept the AFTER keyword. Extended MyISAM to handle records > 16M.
2001-10-29 09:49:35 +01:00
if (!def->after)
{
create_list.push_back(def);
def_it.remove();
}
Import changeset
2000-07-31 21:29:14 +02:00
}
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
else
{
/*
This field was not dropped and not changed, add it to the list
for the new table.
*/
Fix of LEFT JOIN optimizer bug, problem with key cache on Windows and a lot of portability issues.
2000-10-14 02:16:35 +02:00
create_list.push_back(def=new create_field(field,field));
Import changeset
2000-07-31 21:29:14 +02:00
alter_it.rewind(); // Change default if ALTER
Alter_column *alter;
while ((alter=alter_it++))
{
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
if (!my_strcasecmp(system_charset_info,field->field_name, alter->name))
Import changeset
2000-07-31 21:29:14 +02:00
break;
}
if (alter)
{
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
if (def->sql_type == MYSQL_TYPE_BLOB)
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_BLOB_CANT_HAVE_DEFAULT, MYF(0), def->change);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
}
Fix for bug #14693 (ALTER COLUMN SET DEFAULT has no effect)
2005-11-24 17:24:45 +01:00
if ((def->def=alter->def)) // Use new default
def->flags&= ~NO_DEFAULT_VALUE_FLAG;
else
def->flags|= NO_DEFAULT_VALUE_FLAG;
Import changeset
2000-07-31 21:29:14 +02:00
alter_it.remove();
}
}
}
def_it.rewind();
List_iterator<create_field> find_it(create_list);
while ((def=def_it++)) // Add new columns
{
Fixed error message when opening a not-MyISAM file. Extended MODIFY and CHANGE in ALTER TABLE to accept the AFTER keyword. Extended MyISAM to handle records > 16M.
2001-10-29 09:49:35 +01:00
if (def->change && ! def->field)
Import changeset
2000-07-31 21:29:14 +02:00
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_BAD_FIELD_ERROR, MYF(0), def->change, table_name);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
Import changeset
2000-07-31 21:29:14 +02:00
}
if (!def->after)
create_list.push_back(def);
else if (def->after == first_keyword)
create_list.push_front(def);
else
{
create_field *find;
find_it.rewind();
while ((find=find_it++)) // Add new columns
{
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
if (!my_strcasecmp(system_charset_info,def->after, find->field_name))
Import changeset
2000-07-31 21:29:14 +02:00
break;
}
if (!find)
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_BAD_FIELD_ERROR, MYF(0), def->after, table_name);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
Import changeset
2000-07-31 21:29:14 +02:00
}
find_it.after(def); // Put element after this
}
}
Introduced parameter object "ALTER_INFO" for mysql_alter_table to make list of parameters in mysql_alter_table shorted to avoid warning in MSVC (windows) building
2004-05-21 16:57:03 +02:00
if (alter_info->alter_list.elements)
Import changeset
2000-07-31 21:29:14 +02:00
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_BAD_FIELD_ERROR, MYF(0),
alter_info->alter_list.head()->name, table_name);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
Import changeset
2000-07-31 21:29:14 +02:00
}
if (!create_list.elements)
{
post-review fixes
2004-11-12 13:34:00 +01:00
my_message(ER_CANT_REMOVE_ALL_FIELDS, ER(ER_CANT_REMOVE_ALL_FIELDS),
MYF(0));
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
Import changeset
2000-07-31 21:29:14 +02:00
}
/*
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
Collect all keys which isn't in drop list. Add only those
for which some fields exists.
Import changeset
2000-07-31 21:29:14 +02:00
*/
List_iterator<Key> key_it(keys);
List_iterator<create_field> field_it(create_list);
List<key_part_spec> key_parts;
KEY *key_info=table->key_info;
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
for (uint i=0 ; i < table->s->keys ; i++,key_info++)
Import changeset
2000-07-31 21:29:14 +02:00
{
a fix (bug #2361: ALTER TABLE ... DROP PRIMARY KEY drops a non-primary key).
2004-01-13 13:18:37 +01:00
char *key_name= key_info->name;
Import changeset
2000-07-31 21:29:14 +02:00
Alter_drop *drop;
drop_it.rewind();
while ((drop=drop_it++))
{
if (drop->type == Alter_drop::KEY &&
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
!my_strcasecmp(system_charset_info,key_name, drop->name))
Import changeset
2000-07-31 21:29:14 +02:00
break;
}
if (drop)
{
drop_it.remove();
continue;
}
KEY_PART_INFO *key_part= key_info->key_part;
key_parts.empty();
for (uint j=0 ; j < key_info->key_parts ; j++,key_part++)
{
if (!key_part->field)
continue; // Wrong field (from UNIREG)
const char *key_part_name=key_part->field->field_name;
create_field *cfield;
field_it.rewind();
while ((cfield=field_it++))
{
if (cfield->change)
{
Changes for new binary .frm format Fixes after last merge from 4.0. (Code not yet complete, need anoter merge from 4.0)
2002-06-04 07:23:57 +02:00
if (!my_strcasecmp(system_charset_info, key_part_name,
cfield->change))
Import changeset
2000-07-31 21:29:14 +02:00
break;
}
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
else if (!my_strcasecmp(system_charset_info,
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
key_part_name, cfield->field_name))
Changes for new binary .frm format Fixes after last merge from 4.0. (Code not yet complete, need anoter merge from 4.0)
2002-06-04 07:23:57 +02:00
break;
Import changeset
2000-07-31 21:29:14 +02:00
}
if (!cfield)
continue; // Field is removed
uint key_part_length=key_part->length;
if (cfield->field) // Not new field
Fix partial keys when converting VARCHAR to TEXT. (Bug #10543)
2005-05-26 03:11:47 +02:00
{
/*
If the field can't have only a part used in a key according to its
new type, or should not be used partially according to its
previous type, or the field length is less than the key part
length, unset the key part length.
Additional tweak to fix for bug #10543, to prevent a change in behavior when extending fields that were fully part of a multi-part key.
2005-06-02 19:00:36 +02:00
We also unset the key part length if it is the same as the
old field's length, so the whole new field will be used.
Fix partial keys when converting VARCHAR to TEXT. (Bug #10543)
2005-05-26 03:11:47 +02:00
BLOBs may have cfield->length == 0, which is why we test it before
checking whether cfield->length < key_part_length (in chars).
*/
if (!Field::type_can_have_key_part(cfield->field->type()) ||
!Field::type_can_have_key_part(cfield->sql_type) ||
sql_table.cc: Bug#11657 Creation of secondary index fails: If TINYBLOB key part length is 255, it is equal to field length. For BLOB, unlike for CHAR/VARCHAR, we should keep a non-zero key part length, otherwise "BLOB/TEXT column used in key specification without a key length" error is produced afterwards. type_blob.result, type_blob.test: fixing tests accordinly
2005-07-01 12:07:06 +02:00
(cfield->field->field_length == key_part_length &&
!f_is_blob(key_part->key_type)) ||
Fix partial keys when converting VARCHAR to TEXT. (Bug #10543)
2005-05-26 03:11:47 +02:00
(cfield->length && (cfield->length < key_part_length /
key_part->field->charset()->mbmaxlen)))
key_part_length= 0; // Use whole field
Import changeset
2000-07-31 21:29:14 +02:00
}
In CREATE syntax: KEY a (field_name(10)) 10 now means "number of characters", not number of bytes.
2003-09-15 11:45:42 +02:00
key_part_length /= key_part->field->charset()->mbmaxlen;
Import changeset
2000-07-31 21:29:14 +02:00
key_parts.push_back(new key_part_spec(cfield->field_name,
key_part_length));
}
if (key_parts.elements)
Added support for key_block_size for key and table level (WL#602) Added support for key_block_size to MyISAM. Simplify interface to 'new Key' to make it easier to add new key options. mysqld option --new is used to define where key options are printed. (In 5.3 we should move all key options to after key part definition to avoid problem with reserved names) Fixed some compiler warnings and a memory leak in ssl
2006-05-03 14:59:17 +02:00
{
KEY_CREATE_INFO key_create_info;
bzero((char*) &key_create_info, sizeof(key_create_info));
key_create_info.algorithm= key_info->algorithm;
if (key_info->flags & HA_USES_BLOCK_SIZE)
key_create_info.block_size= key_info->block_size;
if (key_info->flags & HA_USES_PARSER)
key_create_info.parser_name= *key_info->parser_name;
Added GIS extension
2002-02-22 12:24:42 +01:00
key_list.push_back(new Key(key_info->flags & HA_SPATIAL ? Key::SPATIAL :
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
(key_info->flags & HA_NOSAME ?
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
(!my_strcasecmp(system_charset_info,
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
key_name, primary_key_name) ?
Key::PRIMARY : Key::UNIQUE) :
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
(key_info->flags & HA_FULLTEXT ?
Key::FULLTEXT : Key::MULTIPLE)),
key_name,
Added support for key_block_size for key and table level (WL#602) Added support for key_block_size to MyISAM. Simplify interface to 'new Key' to make it easier to add new key options. mysqld option --new is used to define where key options are printed. (In 5.3 we should move all key options to after key part definition to avoid problem with reserved names) Fixed some compiler warnings and a memory leak in ssl
2006-05-03 14:59:17 +02:00
&key_create_info,
Don't automaticly generate a new key for a foreign key constraint if there is already a usable key. Prefer not automatic keys before automatic keys. If there is two conf
2004-05-11 23:29:52 +02:00
test(key_info->flags & HA_GENERATED_KEY),
Added support for key_block_size for key and table level (WL#602) Added support for key_block_size to MyISAM. Simplify interface to 'new Key' to make it easier to add new key options. mysqld option --new is used to define where key options are printed. (In 5.3 we should move all key options to after key part definition to avoid problem with reserved names) Fixed some compiler warnings and a memory leak in ssl
2006-05-03 14:59:17 +02:00
key_parts));
}
Import changeset
2000-07-31 21:29:14 +02:00
}
{
Key *key;
while ((key=key_it++)) // Add new keys
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
{
if (key->type != Key::FOREIGN_KEY)
key_list.push_back(key);
fixed bug #856 'Naming a key "Primary" causes trouble'
2004-03-13 22:31:30 +01:00
if (key->name &&
!my_strcasecmp(system_charset_info,key->name,primary_key_name))
{
my_error(ER_WRONG_NAME_FOR_INDEX, MYF(0), key->name);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
fixed bug #856 'Naming a key "Primary" causes trouble'
2004-03-13 22:31:30 +01:00
}
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
}
Import changeset
2000-07-31 21:29:14 +02:00
}
Introduced parameter object "ALTER_INFO" for mysql_alter_table to make list of parameters in mysql_alter_table shorted to avoid warning in MSVC (windows) building
2004-05-21 16:57:03 +02:00
if (alter_info->drop_list.elements)
Import changeset
2000-07-31 21:29:14 +02:00
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_CANT_DROP_FIELD_OR_KEY, MYF(0),
alter_info->drop_list.head()->name);
Import changeset
2000-07-31 21:29:14 +02:00
goto err;
}
Introduced parameter object "ALTER_INFO" for mysql_alter_table to make list of parameters in mysql_alter_table shorted to avoid warning in MSVC (windows) building
2004-05-21 16:57:03 +02:00
if (alter_info->alter_list.elements)
Import changeset
2000-07-31 21:29:14 +02:00
{
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_CANT_DROP_FIELD_OR_KEY, MYF(0),
alter_info->alter_list.head()->name);
Import changeset
2000-07-31 21:29:14 +02:00
goto err;
}
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
db_create_options= table->s->db_create_options & ~(HA_OPTION_PACK_RECORD);
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
my_snprintf(tmp_name, sizeof(tmp_name), "%s-%lx_%lx", tmp_file_prefix,
current_pid, thd->thread_id);
Portablity fix for system with case insenstive table names. (Without this ALTER TABLE can fail for InnoDB)
2004-05-07 05:34:44 +02:00
/* Safety fix for innodb */
if (lower_case_table_names)
Fix for bug which caused grant.test fail on darwin7.3. We were converting db and table names to lower case using latin1 instead of utf-8 in sql_acl.cc if lower_case_table_names was on. Also replaced in other such places system_charset_info with files_charset_info for consistency.
2004-05-22 21:41:58 +02:00
my_casedn_str(files_charset_info, tmp_name);
InnoDB: Prevent ALTER TABLE ... ENGINE=... if there are foreign key constraints on the table. (Bug #5574)
2005-04-07 11:16:41 +02:00
if (new_db_type != old_db_type && !table->file->can_switch_engines()) {
my_error(ER_ROW_IS_REFERENCED, MYF(0));
goto err;
}
Import changeset
2000-07-31 21:29:14 +02:00
create_info->db_type=new_db_type;
Fix for bug#13934 Silent truncation of table comments Table comment: issue a warning(error in traditional mode) if length of comment > 60 symbols Column comment: issue a warning(error in traditional mode) if length of comment > 255 symbols Table 'comment' is changed from char* to LEX_STRING
2006-06-29 15:39:34 +02:00
if (!create_info->comment.str)
{
create_info->comment.str= table->s->comment.str;
create_info->comment.length= table->s->comment.length;
}
Fixed that SHOW CREATE TABLE shows all attributes
2001-09-30 04:47:35 +02:00
table->file->update_create_info(create_info);
if ((create_info->table_options &
(HA_OPTION_PACK_KEYS | HA_OPTION_NO_PACK_KEYS)) ||
(used_fields & HA_CREATE_USED_PACK_KEYS))
Import changeset
2000-07-31 21:29:14 +02:00
db_create_options&= ~(HA_OPTION_PACK_KEYS | HA_OPTION_NO_PACK_KEYS);
if (create_info->table_options &
(HA_OPTION_CHECKSUM | HA_OPTION_NO_CHECKSUM))
db_create_options&= ~(HA_OPTION_CHECKSUM | HA_OPTION_NO_CHECKSUM);
if (create_info->table_options &
(HA_OPTION_DELAY_KEY_WRITE | HA_OPTION_NO_DELAY_KEY_WRITE))
db_create_options&= ~(HA_OPTION_DELAY_KEY_WRITE |
HA_OPTION_NO_DELAY_KEY_WRITE);
create_info->table_options|= db_create_options;
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
if (table->s->tmp_table)
Import changeset
2000-07-31 21:29:14 +02:00
create_info->options|=HA_LEX_CREATE_TMP_TABLE;
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
set_table_default_charset(thd, create_info, db);
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
{
/*
For some purposes we need prepared table structures and translated
key descriptions with proper default key name assignment.
Unfortunately, mysql_prepare_table() modifies the field and key
lists. mysql_create_table() needs the unmodified lists. Hence, we
need to copy the lists and all their elements. The lists contain
pointers to the elements only.
We cannot copy conditionally because the partition code always
needs prepared lists and compare_tables() needs them and is almost
always called.
*/
/* Copy fields. */
List_iterator<create_field> prep_field_it(create_list);
create_field *prep_field;
while ((prep_field= prep_field_it++))
prepared_create_list.push_back(new create_field(*prep_field));
/* Copy keys and key parts. */
List_iterator<Key> prep_key_it(key_list);
Key *prep_key;
while ((prep_key= prep_key_it++))
{
List<key_part_spec> prep_columns;
List_iterator<key_part_spec> prep_col_it(prep_key->columns);
key_part_spec *prep_col;
while ((prep_col= prep_col_it++))
prep_columns.push_back(new key_part_spec(*prep_col));
prepared_key_list.push_back(new Key(prep_key->type, prep_key->name,
Cleanups after review of WL#602 Fixed warnings from test suite Some fixes in mysql-test-run script to catch more warnings
2006-05-03 18:40:52 +02:00
&prep_key->key_create_info,
Added support for key_block_size for key and table level (WL#602) Added support for key_block_size to MyISAM. Simplify interface to 'new Key' to make it easier to add new key options. mysqld option --new is used to define where key options are printed. (In 5.3 we should move all key options to after key part definition to avoid problem with reserved names) Fixed some compiler warnings and a memory leak in ssl
2006-05-03 14:59:17 +02:00
prep_key->generated, prep_columns));
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
}
/* Create the prepared information. */
if (mysql_prepare_table(thd, create_info, &prepared_create_list,
&prepared_key_list,
(table->s->tmp_table != NO_TMP_TABLE), &db_options,
table->file, &key_info_buffer, &key_count, 0))
goto err;
}
if (thd->variables.old_alter_table
|| (table->s->db_type != create_info->db_type)
Make storage engines "pluggable", handlerton work
2005-11-07 16:25:06 +01:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
|| partition_changed
Fixed proper merge of partitioning with WL #2301 ALTER TABLE optimisations
2005-07-27 22:56:28 +02:00
#endif
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
)
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
need_copy_table= 1;
else
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
{
/* Try to optimize ALTER TABLE. Allocate result buffers. */
if (! (index_drop_buffer=
(uint*) thd->alloc(sizeof(uint) * table->s->keys)) ||
! (index_add_buffer=
(uint*) thd->alloc(sizeof(uint) * prepared_key_list.elements)))
goto err;
/* Check how much the tables differ. */
need_copy_table= compare_tables(table, &prepared_create_list,
key_info_buffer, key_count,
create_info, alter_info, order_num,
index_drop_buffer, &index_drop_count,
Merge a193-229-222-105.elisa-laajakaista.fi:/home/my/bk/mysql-5.0 into a193-229-222-105.elisa-laajakaista.fi:/home/my/bk/mysql-5.1-new
2006-05-30 15:07:49 +02:00
index_add_buffer, &index_add_count,
varchar);
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
}
/*
If there are index changes only, try to do them online. "Index
changes only" means also that the handler for the table does not
change. The table is open and locked. The handler can be accessed.
*/
if (need_copy_table == ALTER_TABLE_INDEX_CHANGED)
{
int pk_changed= 0;
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
ulong alter_flags= 0;
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
ulong needed_online_flags= 0;
ulong needed_fast_flags= 0;
KEY *key;
uint *idx_p;
uint *idx_end_p;
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
if (table->s->db_type->alter_table_flags)
alter_flags= table->s->db_type->alter_table_flags(alter_info->flags);
DBUG_PRINT("info", ("alter_flags: %lu", alter_flags));
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
/* Check dropped indexes. */
for (idx_p= index_drop_buffer, idx_end_p= idx_p + index_drop_count;
idx_p < idx_end_p;
idx_p++)
{
key= table->key_info + *idx_p;
DBUG_PRINT("info", ("index dropped: '%s'", key->name));
if (key->flags & HA_NOSAME)
{
/* Unique key. Check for "PRIMARY". */
if (! my_strcasecmp(system_charset_info,
key->name, primary_key_name))
{
/* Primary key. */
needed_online_flags|= HA_ONLINE_DROP_PK_INDEX;
needed_fast_flags|= HA_ONLINE_DROP_PK_INDEX_NO_WRITES;
pk_changed++;
}
else
{
/* Non-primary unique key. */
needed_online_flags|= HA_ONLINE_DROP_UNIQUE_INDEX;
needed_fast_flags|= HA_ONLINE_DROP_UNIQUE_INDEX_NO_WRITES;
}
}
else
{
/* Non-unique key. */
needed_online_flags|= HA_ONLINE_DROP_INDEX;
needed_fast_flags|= HA_ONLINE_DROP_INDEX_NO_WRITES;
}
}
/* Check added indexes. */
for (idx_p= index_add_buffer, idx_end_p= idx_p + index_add_count;
idx_p < idx_end_p;
idx_p++)
{
key= key_info_buffer + *idx_p;
DBUG_PRINT("info", ("index added: '%s'", key->name));
if (key->flags & HA_NOSAME)
{
/* Unique key. Check for "PRIMARY". */
if (! my_strcasecmp(system_charset_info,
key->name, primary_key_name))
{
/* Primary key. */
needed_online_flags|= HA_ONLINE_ADD_PK_INDEX;
needed_fast_flags|= HA_ONLINE_ADD_PK_INDEX_NO_WRITES;
pk_changed++;
}
else
{
/* Non-primary unique key. */
needed_online_flags|= HA_ONLINE_ADD_UNIQUE_INDEX;
needed_fast_flags|= HA_ONLINE_ADD_UNIQUE_INDEX_NO_WRITES;
}
}
else
{
/* Non-unique key. */
needed_online_flags|= HA_ONLINE_ADD_INDEX;
needed_fast_flags|= HA_ONLINE_ADD_INDEX_NO_WRITES;
}
}
/*
Online or fast add/drop index is possible only if
the primary key is not added and dropped in the same statement.
Otherwise we have to recreate the table.
need_copy_table is no-zero at this place.
*/
if ( pk_changed < 2 )
{
if ((alter_flags & needed_online_flags) == needed_online_flags)
{
/* All required online flags are present. */
need_copy_table= 0;
need_lock_for_indexes= FALSE;
}
else if ((alter_flags & needed_fast_flags) == needed_fast_flags)
{
/* All required fast flags are present. */
need_copy_table= 0;
}
}
DBUG_PRINT("info", ("need_copy_table: %u need_lock: %d",
need_copy_table, need_lock_for_indexes));
}
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
/*
better have a negative test here, instead of positive, like
Merge with 4.1
2005-05-14 17:31:22 +02:00
alter_info->flags & ALTER_ADD_COLUMN|ALTER_ADD_INDEX|...
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
so that ALTER TABLE won't break when somebody will add new flag
*/
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
if (!need_copy_table)
create_info->frm_only= 1;
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
Make storage engines "pluggable", handlerton work
2005-11-07 16:25:06 +01:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
if (fast_alter_partition)
WL #2602, #2603, #2604 Added new syntax for partition management
2005-08-19 16:26:05 +02:00
{
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
DBUG_RETURN(fast_alter_partition_table(thd, table, alter_info,
create_info, table_list,
&create_list, &key_list,
db, table_name,
fast_alter_partition));
WL #2602, #2603, #2604 Added new syntax for partition management
2005-08-19 16:26:05 +02:00
}
Fixed back some bin log writes to avoid merge problems Added missing #ifdef for partitioning
2005-08-20 22:56:54 +02:00
#endif
WL #2602, #2603, #2604 Added new syntax for partition management
2005-08-19 16:26:05 +02:00
Added support for symlinked tables. myisamchk: Don't force a repair if the only problem was that the open count wasn't correct. Added missing error messages.
2001-06-01 03:27:59 +02:00
/*
Handling of symlinked tables:
If no rename:
Create new data file and index file on the same disk as the
old data and index files.
Copy data.
Rename new data file over old data file and new index file over
old index file.
Symlinks are not changed.
If rename:
Create new data file and index file on the same disk as the
old data and index files. Create also symlinks to point at
the new tables.
Copy data.
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
At end, rename intermediate tables, and symlinks to intermediate
table, to final table name.
Added support for symlinked tables. myisamchk: Don't force a repair if the only problem was that the open count wasn't correct. Added missing error messages.
2001-06-01 03:27:59 +02:00
Remove old table and old symlinks
If rename is made to another database:
Create new tables in new database.
Copy data.
Remove old table and symlinks.
*/
if (!strcmp(db, new_db)) // Ignore symlink if db changed
{
if (create_info->index_file_name)
{
/* Fix index_file_name to have 'tmp_name' as basename */
strmov(index_file, tmp_name);
create_info->index_file_name=fn_same(index_file,
create_info->index_file_name,
1);
}
if (create_info->data_file_name)
{
/* Fix data_file_name to have 'tmp_name' as basename */
strmov(data_file, tmp_name);
create_info->data_file_name=fn_same(data_file,
create_info->data_file_name,
1);
}
}
Symlinking of tables should not work. Fixed error number handling bug in mysqltest. Fixed that error number from insert delayed is reported correctly. merged new vio code with old violite code.
2001-06-05 02:38:10 +02:00
else
create_info->data_file_name=create_info->index_file_name=0;
merge with 4.1
2004-10-29 18:26:52 +02:00
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
/*
Create a table with a temporary name.
With create_info->frm_only == 1 this creates a .frm file only.
We don't log the statement, it will be logged later.
*/
tmp_disable_binlog(thd);
error= mysql_create_table(thd, new_db, tmp_name,
BUG#19309: Problem with calling proecdures twice Need to flag when a copy is needed to not overwrite a create_info object connected to the lex structure
2006-06-21 16:57:30 +02:00
create_info,create_list,key_list,1,0,0);
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
reenable_binlog(thd);
if (error)
DBUG_RETURN(error);
/* Open the table if we need to copy the data. */
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
if (need_copy_table)
Applied patches for BDB tables Fixes to InnoDB to compile on Windows Fix for temporary InnoDB tables Fixed bug in REPLACE() Fixed sub char keys for InnoDB
2001-05-09 22:02:36 +02:00
{
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
if (table->s->tmp_table)
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
{
TABLE_LIST tbl;
bzero((void*) &tbl, sizeof(tbl));
tbl.db= new_db;
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
tbl.table_name= tbl.alias= tmp_name;
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
/* Table is in thd->temporary_tables */
Bug#12845 - Stress test: Server crashes on CREATE .. SELECT statement Solution for 5.0. Changed calls to open_table(). Requested to ignore flush at places where the command did already lock tables. This could happen in CREATE ... SELECT and ALTER TABLE. No test case. The bug can only be triggered by true concurrency. The stress test suite provides a test case for this.
2005-09-13 14:36:43 +02:00
new_table= open_table(thd, &tbl, thd->mem_root, (bool*) 0,
MYSQL_LOCK_IGNORE_FLUSH);
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
}
else
{
char path[FN_REFLEN];
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
/* table is a normal table: Create temporary table in same directory */
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
build_table_filename(path, sizeof(path), new_db, tmp_name, "",
FN_IS_TMP);
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
/* Open our intermediate table */
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
new_table=open_temporary_table(thd, path, new_db, tmp_name,0);
}
if (!new_table)
Bug#18129 - Fast (online) add index leaves temporary table frm in case of errors ALTER TABLE temporarily creates a new table with a .frm file and optionally other files. For fast ALTER TABLE only the .frm file is created. If the operation succeeds, The temporary files are renamed to their final target. In case of an error, the temporary file was forgotten to remove. Manually tested. The test requires to look at files, which I think cannot be done portably with the test suite. The test file is attached to the bug report.
2006-04-13 16:19:19 +02:00
goto err1;
Import changeset
2000-07-31 21:29:14 +02:00
}
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
/* Copy the data if necessary. */
Add warnings for single line inserts. To do this I had to convert count_cuted_fields to an enum (to be able to detect setting a NOT NULL field to NULL)
2003-10-11 22:26:39 +02:00
thd->count_cuted_fields= CHECK_FIELD_WARN; // calc cuted fields
Import changeset
2000-07-31 21:29:14 +02:00
thd->cuted_fields=0L;
thd->proc_info="copy to tmp table";
Fixed ALTER TABLE on MERGE tables Fixed bug in DISTINCT
2001-01-28 20:35:50 +01:00
copied=deleted=0;
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
if (new_table && !(new_table->file->ha_table_flags() & HA_NO_COPY_ON_ALTER))
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
{
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
/* We don't want update TIMESTAMP fields during ALTER TABLE. */
After merge fixes
2004-11-03 19:07:17 +01:00
new_table->timestamp_field_type= TIMESTAMP_NO_AUTO_SET;
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
new_table->next_number_field=new_table->found_next_number_field;
Fix for bug#18437 "Wrong values inserted with a before update trigger on NDB table". SQL-layer was not marking fields which were used in triggers as such. As result these fields were not always properly retrieved/stored by handler layer. So one might got wrong values or lost changes in triggers for NDB, Federated and possibly InnoDB tables. This fix solves the problem by marking fields used in triggers appropriately. Also this patch contains the following cleanup of ha_ndbcluster code: We no longer rely on reading LEX::sql_command value in handler in order to determine if we can enable optimization which allows us to handle REPLACE statement in more efficient way by doing replaces directly in write_row() method without reporting error to SQL-layer. Instead we rely on SQL-layer informing us whether this optimization applicable by calling handler::extra() method with HA_EXTRA_WRITE_CAN_REPLACE flag. As result we no longer apply this optimzation in cases when it should not be used (e.g. if we have on delete triggers on table) and use in some additional cases when it is applicable (e.g. for LOAD DATA REPLACE). Finally this patch includes fix for bug#20728 "REPLACE does not work correctly for NDB table with PK and unique index". This was yet another problem which was caused by improper field mark-up. During row replacement fields which weren't explicity used in REPLACE statement were not marked as fields to be saved (updated) so they have retained values from old row version. The fix is to mark all table fields as set for REPLACE statement. Note that in 5.1 we already solve this problem by notifying handler that it should save values from all fields only in case when real replacement happens.
2006-07-01 23:51:10 +02:00
error=copy_data_between_tables(table, new_table, create_list, ignore,
Merge example.com:/work/bug24395-v2/my50 into example.com:/work/bug24395-v2/my51
2006-12-01 11:37:33 +01:00
order_num, order, &copied, &deleted,
Fix for bug#24395: ALTER TABLE DISABLE KEYS doesn't work when modifying the table ENABLE|DISABLE KEYS combined with another ALTER TABLE option, different than RENAME TO did nothing. Also, if the table had disabled keys and was ALTER-ed then the end table was with enabled keys. Fixed by checking whether the table had disabled keys and enabling them in the copied table.
2006-11-28 18:27:32 +01:00
alter_info->keys_onoff);
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
}
Merge example.com:/work/bug24395-v2/my50 into example.com:/work/bug24395-v2/my51
2006-12-01 11:37:33 +01:00
else
Merge example.com:/work/bug24395-v2/my41 into example.com:/work/bug24395-v2/my50 fix for bug#24395 merged into 5.0
2006-11-30 18:36:15 +01:00
{
VOID(pthread_mutex_lock(&LOCK_open));
wait_while_table_is_used(thd, table, HA_EXTRA_FORCE_REOPEN);
Merge example.com:/work/bug24395-v2/my50 into example.com:/work/bug24395-v2/my51
2006-12-01 11:37:33 +01:00
table->file->ha_external_lock(thd, F_WRLCK);
Merge example.com:/work/bug24395-v2/my41 into example.com:/work/bug24395-v2/my50 fix for bug#24395 merged into 5.0
2006-11-30 18:36:15 +01:00
alter_table_manage_keys(table, table->file->indexes_are_disabled(),
alter_info->keys_onoff);
Merge example.com:/work/bug24395-v2/my50 into example.com:/work/bug24395-v2/my51
2006-12-01 11:37:33 +01:00
table->file->ha_external_lock(thd, F_UNLCK);
Merge example.com:/work/bug24395-v2/my41 into example.com:/work/bug24395-v2/my50 fix for bug#24395 merged into 5.0
2006-11-30 18:36:15 +01:00
VOID(pthread_mutex_unlock(&LOCK_open));
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
}
Add warnings for single line inserts. To do this I had to convert count_cuted_fields to an enum (to be able to detect setting a NOT NULL field to NULL)
2003-10-11 22:26:39 +02:00
thd->count_cuted_fields= CHECK_FIELD_IGNORE;
Import changeset
2000-07-31 21:29:14 +02:00
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
/* If we did not need to copy, we might still need to add/drop indexes. */
if (! new_table)
{
uint *key_numbers;
uint *keyno_p;
KEY *key_info;
KEY *key;
uint *idx_p;
uint *idx_end_p;
KEY_PART_INFO *key_part;
KEY_PART_INFO *part_end;
DBUG_PRINT("info", ("No new_table, checking add/drop index"));
Fixes for fast alter table for ndbcluster
2006-04-11 14:06:32 +02:00
table->file->prepare_for_alter();
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
if (index_add_count)
{
#ifdef XXX_TO_BE_DONE_LATER_BY_WL3020_AND_WL1892
if (! need_lock_for_indexes)
{
/* Downgrade the write lock. */
mysql_lock_downgrade_write(thd, table, TL_WRITE_ALLOW_WRITE);
}
/* Create a new .frm file for crash recovery. */
/* TODO: Must set INDEX_TO_BE_ADDED flags in the frm file. */
VOID(pthread_mutex_lock(&LOCK_open));
error= (mysql_create_frm(thd, reg_path, db, table_name,
create_info, prepared_create_list, key_count,
key_info_buffer, table->file) ||
WL 2826: Error handling of ALTER TABLE for partitioning Fixed up create_handler_files call
2006-04-20 03:43:30 +02:00
table->file->create_handler_files(reg_path, NULL, CHF_INDEX_FLAG,
Merge c-870ae253.1238-1-64736c10.cust.bredbandsbolaget.se:/home/pappa/clean-mysql-5.1-new into c-870ae253.1238-1-64736c10.cust.bredbandsbolaget.se:/home/pappa/wl2826
2006-04-20 03:22:35 +02:00
create_info));
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
VOID(pthread_mutex_unlock(&LOCK_open));
if (error)
Bug#18129 - Fast (online) add index leaves temporary table frm in case of errors ALTER TABLE temporarily creates a new table with a .frm file and optionally other files. For fast ALTER TABLE only the .frm file is created. If the operation succeeds, The temporary files are renamed to their final target. In case of an error, the temporary file was forgotten to remove. Manually tested. The test requires to look at files, which I think cannot be done portably with the test suite. The test file is attached to the bug report.
2006-04-13 16:19:19 +02:00
goto err1;
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
#endif
/* The add_index() method takes an array of KEY structs. */
key_info= (KEY*) thd->alloc(sizeof(KEY) * index_add_count);
key= key_info;
for (idx_p= index_add_buffer, idx_end_p= idx_p + index_add_count;
idx_p < idx_end_p;
idx_p++, key++)
{
/* Copy the KEY struct. */
*key= key_info_buffer[*idx_p];
/* Fix the key parts. */
part_end= key->key_part + key->key_parts;
for (key_part= key->key_part; key_part < part_end; key_part++)
key_part->field= table->field[key_part->fieldnr];
}
/* Add the indexes. */
if ((error= table->file->add_index(table, key_info, index_add_count)))
{
/*
Exchange the key_info for the error message. If we exchange
key number by key name in the message later, we need correct info.
*/
KEY *save_key_info= table->key_info;
table->key_info= key_info;
table->file->print_error(error, MYF(0));
table->key_info= save_key_info;
Bug#18129 - Fast (online) add index leaves temporary table frm in case of errors ALTER TABLE temporarily creates a new table with a .frm file and optionally other files. For fast ALTER TABLE only the .frm file is created. If the operation succeeds, The temporary files are renamed to their final target. In case of an error, the temporary file was forgotten to remove. Manually tested. The test requires to look at files, which I think cannot be done portably with the test suite. The test file is attached to the bug report.
2006-04-13 16:19:19 +02:00
goto err1;
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
}
}
/*end of if (index_add_count)*/
if (index_drop_count)
{
#ifdef XXX_TO_BE_DONE_LATER_BY_WL3020_AND_WL1892
/* Create a new .frm file for crash recovery. */
/* TODO: Must set INDEX_IS_ADDED in the frm file. */
/* TODO: Must set INDEX_TO_BE_DROPPED in the frm file. */
VOID(pthread_mutex_lock(&LOCK_open));
error= (mysql_create_frm(thd, reg_path, db, table_name,
create_info, prepared_create_list, key_count,
key_info_buffer, table->file) ||
WL 2826: Error handling of ALTER TABLE for partitioning Fixed up create_handler_files call
2006-04-20 03:43:30 +02:00
table->file->create_handler_files(reg_path, NULL, CHF_INDEX_FLAG,
Merge c-870ae253.1238-1-64736c10.cust.bredbandsbolaget.se:/home/pappa/clean-mysql-5.1-new into c-870ae253.1238-1-64736c10.cust.bredbandsbolaget.se:/home/pappa/wl2826
2006-04-20 03:22:35 +02:00
create_info));
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
VOID(pthread_mutex_unlock(&LOCK_open));
if (error)
Bug#18129 - Fast (online) add index leaves temporary table frm in case of errors ALTER TABLE temporarily creates a new table with a .frm file and optionally other files. For fast ALTER TABLE only the .frm file is created. If the operation succeeds, The temporary files are renamed to their final target. In case of an error, the temporary file was forgotten to remove. Manually tested. The test requires to look at files, which I think cannot be done portably with the test suite. The test file is attached to the bug report.
2006-04-13 16:19:19 +02:00
goto err1;
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
if (! need_lock_for_indexes)
{
LOCK_PARAM_TYPE lpt;
lpt.thd= thd;
lpt.table= table;
lpt.db= db;
lpt.table_name= table_name;
lpt.create_info= create_info;
lpt.create_list= &create_list;
lpt.key_count= key_count;
lpt.key_info_buffer= key_info_buffer;
abort_and_upgrade_lock(lpt);
}
#endif
/* The prepare_drop_index() method takes an array of key numbers. */
key_numbers= (uint*) thd->alloc(sizeof(uint) * index_drop_count);
keyno_p= key_numbers;
/* Get the number of each key. */
for (idx_p= index_drop_buffer, idx_end_p= idx_p + index_drop_count;
idx_p < idx_end_p;
idx_p++, keyno_p++)
*keyno_p= *idx_p;
/*
Tell the handler to prepare for drop indexes.
This re-numbers the indexes to get rid of gaps.
*/
if ((error= table->file->prepare_drop_index(table, key_numbers,
index_drop_count)))
{
table->file->print_error(error, MYF(0));
Bug#18129 - Fast (online) add index leaves temporary table frm in case of errors ALTER TABLE temporarily creates a new table with a .frm file and optionally other files. For fast ALTER TABLE only the .frm file is created. If the operation succeeds, The temporary files are renamed to their final target. In case of an error, the temporary file was forgotten to remove. Manually tested. The test requires to look at files, which I think cannot be done portably with the test suite. The test file is attached to the bug report.
2006-04-13 16:19:19 +02:00
goto err1;
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
}
#ifdef XXX_TO_BE_DONE_LATER_BY_WL3020
if (! need_lock_for_indexes)
{
/* Downgrade the lock again. */
if (table->reginfo.lock_type == TL_WRITE_ALLOW_READ)
{
LOCK_PARAM_TYPE lpt;
lpt.thd= thd;
lpt.table= table;
lpt.db= db;
lpt.table_name= table_name;
lpt.create_info= create_info;
lpt.create_list= &create_list;
lpt.key_count= key_count;
lpt.key_info_buffer= key_info_buffer;
close_open_tables_and_downgrade(lpt);
}
}
#endif
/* Tell the handler to finally drop the indexes. */
if ((error= table->file->final_drop_index(table)))
{
table->file->print_error(error, MYF(0));
Bug#18129 - Fast (online) add index leaves temporary table frm in case of errors ALTER TABLE temporarily creates a new table with a .frm file and optionally other files. For fast ALTER TABLE only the .frm file is created. If the operation succeeds, The temporary files are renamed to their final target. In case of an error, the temporary file was forgotten to remove. Manually tested. The test requires to look at files, which I think cannot be done portably with the test suite. The test file is attached to the bug report.
2006-04-13 16:19:19 +02:00
goto err1;
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
}
}
/*end of if (index_drop_count)*/
Fixes for fast alter table for ndbcluster
2006-04-11 14:06:32 +02:00
/*
The final .frm file is already created as a temporary file
and will be renamed to the original table name later.
*/
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
Fixes for fast alter table for ndbcluster
2006-04-11 14:06:32 +02:00
/* Need to commit before a table is unlocked (NDB requirement). */
DBUG_PRINT("info", ("Committing before unlocking table"));
if (ha_commit_stmt(thd) || ha_commit(thd))
Bug#18129 - Fast (online) add index leaves temporary table frm in case of errors ALTER TABLE temporarily creates a new table with a .frm file and optionally other files. For fast ALTER TABLE only the .frm file is created. If the operation succeeds, The temporary files are renamed to their final target. In case of an error, the temporary file was forgotten to remove. Manually tested. The test requires to look at files, which I think cannot be done portably with the test suite. The test file is attached to the bug report.
2006-04-13 16:19:19 +02:00
goto err1;
Fixes for fast alter table for ndbcluster
2006-04-11 14:06:32 +02:00
committed= 1;
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
}
/*end of if (! new_table) for add/drop index*/
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
if (table->s->tmp_table != NO_TMP_TABLE)
Import changeset
2000-07-31 21:29:14 +02:00
{
/* We changed a temporary table */
if (error)
Bug#18129 - Fast (online) add index leaves temporary table frm in case of errors ALTER TABLE temporarily creates a new table with a .frm file and optionally other files. For fast ALTER TABLE only the .frm file is created. If the operation succeeds, The temporary files are renamed to their final target. In case of an error, the temporary file was forgotten to remove. Manually tested. The test requires to look at files, which I think cannot be done portably with the test suite. The test file is attached to the bug report.
2006-04-13 16:19:19 +02:00
goto err1;
Fix bug in ALTER TABLE on a TEMPORARY InnoDB table.
2001-11-07 22:18:12 +01:00
/* Close lock if this is a transactional table */
if (thd->lock)
{
mysql_unlock_tables(thd, thd->lock);
thd->lock=0;
}
Import changeset
2000-07-31 21:29:14 +02:00
/* Remove link to old table and rename the new one */
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
close_temporary_table(thd, table, 1, 1);
A fix (bug #7261: Alter table loses temp table with lower_case_table_names=2).
2005-01-21 12:14:31 +01:00
/* Should pass the 'new_name' as we store table name in the cache */
if (rename_temporary_table(thd, new_table, new_db, new_name))
Bug#18129 - Fast (online) add index leaves temporary table frm in case of errors ALTER TABLE temporarily creates a new table with a .frm file and optionally other files. For fast ALTER TABLE only the .frm file is created. If the operation succeeds, The temporary files are renamed to their final target. In case of an error, the temporary file was forgotten to remove. Manually tested. The test requires to look at files, which I think cannot be done portably with the test suite. The test file is attached to the bug report.
2006-04-13 16:19:19 +02:00
goto err1;
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
/* We don't replicate alter table statement on temporary tables */
WL#2977 and WL#2712 global and session-level variable to set the binlog format (row/statement), and new binlog format called "mixed" (which is statement-based except if only row-based is correct, in this cset it means if UDF or UUID is used; more cases could be added in later 5.1 release): SET GLOBAL|SESSION BINLOG_FORMAT=row|statement|mixed|default; the global default is statement unless cluster is enabled (then it's row) as in 5.1-alpha. It's not possible to use SET on this variable if a session is currently in row-based mode and has open temporary tables (because CREATE TEMPORARY TABLE was not binlogged so temp table is not known on slave), or if NDB is enabled (because NDB does not support such change on-the-fly, though it will later), of if in a stored function (see below). The added tests test the possibility or impossibility to SET, their effects, and the mixed mode, including in prepared statements and in stored procedures and functions. Caveats: a) The mixed mode will not work for stored functions: in mixed mode, a stored function will always be binlogged as one call and in a statement-based way (e.g. INSERT VALUES(myfunc()) or SELECT myfunc()). b) for the same reason, changing the thread's binlog format inside a stored function is refused with an error message. c) the same problems apply to triggers; implementing b) for triggers will be done later (will ask Dmitri). Additionally, as the binlog format is now changeable by each user for his session, I remove the implication which was done at startup, where row-based automatically set log-bin-trust-routine-creators to 1 (not possible anymore as a user can now switch to stmt-based and do nasty things again), and automatically set --innodb-locks-unsafe-for-binlog to 1 (was anyway theoretically incorrect as it disabled phantom protection). Plus fixes for compiler warnings.
2006-02-25 22:21:03 +01:00
if (!thd->current_stmt_binlog_row_based)
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
write_bin_log(thd, TRUE, thd->query, thd->query_length);
Import changeset
2000-07-31 21:29:14 +02:00
goto end_temporary;
}
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
if (new_table)
{
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
/* Close the intermediate table that will be the new table */
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
intern_close_table(new_table);
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
my_free((gptr) new_table,MYF(0));
}
Import changeset
2000-07-31 21:29:14 +02:00
VOID(pthread_mutex_lock(&LOCK_open));
if (error)
{
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
VOID(quick_rm_table(new_db_type, new_db, tmp_name, FN_IS_TMP));
Import changeset
2000-07-31 21:29:14 +02:00
VOID(pthread_mutex_unlock(&LOCK_open));
goto err;
}
Fixed memory leak in DROP DATABASE when using RAID tables (Bug #2882)
2004-03-10 12:46:11 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
/*
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
Data is copied. Now we rename the old table to a temp name,
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
rename the new one to the old name, remove all entries about the old table
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
from the cache, free all locks, close the old table and remove it.
Import changeset
2000-07-31 21:29:14 +02:00
*/
thd->proc_info="rename result table";
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
my_snprintf(old_name, sizeof(old_name), "%s2-%lx-%lx", tmp_file_prefix,
current_pid, thd->thread_id);
Fix for sporadically failures of innodb.test on Darwin 7.3 which occured because we were not lowering case of file names for temporary tables altough handler assumes so if lower_case_table_names==2. Now we are lowering case for them.
2004-05-25 00:30:09 +02:00
if (lower_case_table_names)
my_casedn_str(files_charset_info, old_name);
Enable LOAD DATA LOCAL INFILE in mysql_test Added syntax for column comments (for compability with 4.1) Fix of ALTER TABLE RENAME
2002-06-03 23:40:27 +02:00
if (new_name != table_name || new_db != db)
Import changeset
2000-07-31 21:29:14 +02:00
{
if (!access(new_name_buff,F_OK))
{
error=1;
now my_printf_error is not better then my_error, but my_error call is shorter used only one implementation of format parser of (printf) fixed multistatement
2004-11-13 18:35:51 +01:00
my_error(ER_TABLE_EXISTS_ERROR, MYF(0), new_name_buff);
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
VOID(quick_rm_table(new_db_type, new_db, tmp_name, FN_IS_TMP));
Import changeset
2000-07-31 21:29:14 +02:00
VOID(pthread_mutex_unlock(&LOCK_open));
goto err;
}
}
Final patch to remove the last of the OS2 support.
2006-04-16 03:17:32 +02:00
#if !defined( __WIN__)
Update to autoconf 2.52 Fix of InnoDB variables in SHOW VARIABLES Fix for ALTER TABLE and InnoDB
2002-02-07 20:34:35 +01:00
if (table->file->has_transactions())
#endif
{
/*
Updated changelog of 3.23.48 with InnoDB changes.
2002-02-11 11:56:48 +01:00
Win32 and InnoDB can't drop a table that is in use, so we must
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
close the original table before doing the rename
Update to autoconf 2.52 Fix of InnoDB variables in SHOW VARIABLES Fix for ALTER TABLE and InnoDB
2002-02-07 20:34:35 +01:00
*/
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
table->s->version= 0; // Force removal of table def
WL#1218 "Triggers". Some very preliminary version of patch. Mostly needed for Monty for him getting notion what needed for triggers from new .FRM format. Things to be done: - Right placement of trigger's invocations - Right handling of errors in triggers (including transaction rollback) - Support for priviliges - Right handling of DROP/RENAME table (hope that it will be handled automatically with merging of .TRG into .FRM file) - Saving/restoring some information critical for trigger creation and replication with their definitions (e.g. sql_mode, creator, ...) - Replication Already has some known bugs so probably not for general review.
2004-09-07 14:29:46 +02:00
close_cached_table(thd, table);
Update to autoconf 2.52 Fix of InnoDB variables in SHOW VARIABLES Fix for ALTER TABLE and InnoDB
2002-02-07 20:34:35 +01:00
table=0; // Marker that table is closed
Follow-up for the fix for bug #18153 "ALTER/OPTIMIZE/REPAIR on transactional tables corrupt triggers". It turned out that we also have relied at certain places that (new_table != table_name) were always true on Windows and for transactional tables. Since our fix for the bug brakes this assumption we have to add new flag to pass this information around. This code needs to be refactored but I dare not to do this in 5.0.
2006-03-24 20:46:33 +01:00
no_table_reopen= TRUE;
Import changeset
2000-07-31 21:29:14 +02:00
}
Final patch to remove the last of the OS2 support.
2006-04-16 03:17:32 +02:00
#if !defined( __WIN__)
Update to autoconf 2.52 Fix of InnoDB variables in SHOW VARIABLES Fix for ALTER TABLE and InnoDB
2002-02-07 20:34:35 +01:00
else
table->file->extra(HA_EXTRA_FORCE_REOPEN); // Don't use this file anymore
Import changeset
2000-07-31 21:29:14 +02:00
#endif
Update to autoconf 2.52 Fix of InnoDB variables in SHOW VARIABLES Fix for ALTER TABLE and InnoDB
2002-02-07 20:34:35 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
error=0;
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
save_old_db_type= old_db_type;
/*
This leads to the storage engine (SE) not being notified for renames in
mysql_rename_table(), because we just juggle with the FRM and nothing
more. If we have an intermediate table, then we notify the SE that
it should become the actual table. Later, we will recycle the old table.
However, in case of ALTER TABLE RENAME there might be no intermediate
table. This is when the old and new tables are compatible, according to
compare_table(). Then, we need one additional call to
mysql_rename_table() with flag NO_FRM_RENAME, which does nothing else but
actual rename in the SE and the FRM is not touched. Note that, if the
table is renamed and the SE is also changed, then an intermediate table
is created and the additional call will not take place.
*/
smarter ALTER TABLE - don't copy the table if only comment or default values are changed
2004-10-21 18:10:58 +02:00
if (!need_copy_table)
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
new_db_type=old_db_type= NULL; // this type cannot happen in regular ALTER
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
if (mysql_rename_table(old_db_type, db, table_name, db, old_name,
FN_TO_IS_TMP))
Import changeset
2000-07-31 21:29:14 +02:00
{
error=1;
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
VOID(quick_rm_table(new_db_type, new_db, tmp_name, FN_IS_TMP));
Import changeset
2000-07-31 21:29:14 +02:00
}
else if (mysql_rename_table(new_db_type,new_db,tmp_name,new_db,
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
new_alias, FN_FROM_IS_TMP) ||
Fix for bug #13525 "Rename table does not keep info of triggers". Let us transfer triggers associated with table when we rename it (but only if we are not changing database to which table belongs, in the latter case we will emit error).
2006-02-24 21:50:36 +01:00
(new_name != table_name || new_db != db) && // we also do rename
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
(need_copy_table ||
mysql_rename_table(save_old_db_type, db, table_name, new_db,
new_alias, NO_FRM_RENAME)) &&
Fix for bug #13525 "Rename table does not keep info of triggers". Let us transfer triggers associated with table when we rename it (but only if we are not changing database to which table belongs, in the latter case we will emit error).
2006-02-24 21:50:36 +01:00
Table_triggers_list::change_table_name(thd, db, table_name,
new_db, new_alias))
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
{
/* Try to get everything back. */
Import changeset
2000-07-31 21:29:14 +02:00
error=1;
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
VOID(quick_rm_table(new_db_type,new_db,new_alias, 0));
VOID(quick_rm_table(new_db_type, new_db, tmp_name, FN_IS_TMP));
VOID(mysql_rename_table(old_db_type, db, old_name, db, alias,
FN_FROM_IS_TMP));
Import changeset
2000-07-31 21:29:14 +02:00
}
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
if (error)
{
Update of query cache code
2001-12-06 00:05:30 +01:00
/*
This shouldn't happen. We solve this the safe way by
closing the locked table.
*/
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
if (table)
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
{
table->s->version= 0; // Force removal of table def
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
close_cached_table(thd,table);
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
}
Import changeset
2000-07-31 21:29:14 +02:00
VOID(pthread_mutex_unlock(&LOCK_open));
goto err;
}
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
if (! need_copy_table)
{
if (! table)
{
Additional fix for bug#22369
2006-12-07 13:57:39 +01:00
if (new_name != table_name || new_db != db)
{
table_list->alias= new_name;
table_list->table_name= new_name;
table_list->table_name_length= strlen(new_name);
table_list->db= new_db;
table_list->db_length= strlen(new_db);
}
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
VOID(pthread_mutex_unlock(&LOCK_open));
if (! (table= open_ltable(thd, table_list, TL_WRITE_ALLOW_READ)))
goto err;
VOID(pthread_mutex_lock(&LOCK_open));
}
/* Tell the handler that a new frm file is in place. */
WL 2826: Error handling of ALTER TABLE for partitioning Fixed merge issue
2006-04-22 10:38:19 +02:00
if (table->file->create_handler_files(path, NULL, CHF_INDEX_FLAG,
Merge c-870ae253.1238-1-64736c10.cust.bredbandsbolaget.se:/home/pappa/clean-mysql-5.1-new into c-870ae253.1238-1-64736c10.cust.bredbandsbolaget.se:/home/pappa/wl2826
2006-04-20 03:22:35 +02:00
create_info))
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
{
VOID(pthread_mutex_unlock(&LOCK_open));
goto err;
}
}
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
Follow-up for the fix for bug #18153 "ALTER/OPTIMIZE/REPAIR on transactional tables corrupt triggers". It turned out that we also have relied at certain places that (new_table != table_name) were always true on Windows and for transactional tables. Since our fix for the bug brakes this assumption we have to add new flag to pass this information around. This code needs to be refactored but I dare not to do this in 5.0.
2006-03-24 20:46:33 +01:00
if (thd->lock || new_name != table_name || no_table_reopen) // True if WIN32
Import changeset
2000-07-31 21:29:14 +02:00
{
Update of query cache code
2001-12-06 00:05:30 +01:00
/*
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
Not table locking or alter table with rename.
Free locks and remove old table
Update of query cache code
2001-12-06 00:05:30 +01:00
*/
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
if (table)
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
{
table->s->version= 0; // Force removal of table def
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
close_cached_table(thd,table);
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
}
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
VOID(quick_rm_table(old_db_type, db, old_name, FN_IS_TMP));
Import changeset
2000-07-31 21:29:14 +02:00
}
else
{
Update of query cache code
2001-12-06 00:05:30 +01:00
/*
Using LOCK TABLES without rename.
This code is never executed on WIN32!
Remove old renamed table, reopen table and get new locks
*/
Import changeset
2000-07-31 21:29:14 +02:00
if (table)
{
VOID(table->file->extra(HA_EXTRA_FORCE_REOPEN)); // Use new file
Review of new code: Change argument order to be database, tablename for some functions
2005-07-27 12:05:30 +02:00
/* Mark in-use copies old */
Bug #10600 After review fixes
2005-07-20 21:19:01 +02:00
remove_table_from_cache(thd,db,table_name,RTFC_NO_FLAG);
Review of new code: Change argument order to be database, tablename for some functions
2005-07-27 12:05:30 +02:00
/* end threads waiting on lock */
WL #2604: Partition Management Optimised version of ADD/DROP/REORGANIZE partitions for non-NDB storage engines. New syntax to handle REBUILD/OPTIMIZE/ANALYZE/CHECK/REPAIR partitions Quite a few bug fixes
2006-01-17 08:40:00 +01:00
mysql_lock_abort(thd,table, TRUE);
Import changeset
2000-07-31 21:29:14 +02:00
}
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
VOID(quick_rm_table(old_db_type, db, old_name, FN_IS_TMP));
Import changeset
2000-07-31 21:29:14 +02:00
if (close_data_tables(thd,db,table_name) ||
reopen_tables(thd,1,0))
{ // This shouldn't happen
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
if (table)
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
{
table->s->version= 0; // Force removal of table def
Fixed bug in ALTER TABLE DISABLE KEYS and INSERT DELAYED. Bug #478
2003-06-01 11:32:53 +02:00
close_cached_table(thd,table); // Remove lock for table
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
}
Import changeset
2000-07-31 21:29:14 +02:00
VOID(pthread_mutex_unlock(&LOCK_open));
goto err;
}
}
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
VOID(pthread_mutex_unlock(&LOCK_open));
Merge mysql.com:/home/mydev/mysql-5.0-tmp_merge into mysql.com:/home/mydev/mysql-5.1-amerge
2006-07-04 10:02:11 +02:00
broadcast_refresh();
WL#1563 - Modify MySQL to support fast CREATE/DROP INDEX NDB cluster is not fully supported. This will be added with WL 1892 (NDB Handler: Add support for CREATE/DROP INDEX). Some preparatory code for this is already present though. A change for the "duplicate key" error message is planned for another changeset.
2006-01-12 10:05:07 +01:00
/*
The ALTER TABLE is always in its own transaction.
Commit must not be called while LOCK_open is locked. It could call
wait_if_global_read_lock(), which could create a deadlock if called
with LOCK_open.
*/
Additional fixes for WL#1892
2006-01-17 12:53:49 +01:00
if (!committed)
{
error = ha_commit_stmt(thd);
if (ha_commit(thd))
error=1;
if (error)
goto err;
}
Import changeset
2000-07-31 21:29:14 +02:00
thd->proc_info="end";
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
Bug #17038, distribution of schema operation to multiple binlogs missing/multiple entries, partial fix - log alter table directly in server instead of in handler - acknowledge alter table _after_ all binlog events have been processed
2006-02-06 11:47:12 +01:00
ha_binlog_log_query(thd, create_info->db_type, LOGCOM_ALTER_TABLE,
thd->query, thd->query_length,
db, table_name);
WL#2977 and WL#2712 global and session-level variable to set the binlog format (row/statement), and new binlog format called "mixed" (which is statement-based except if only row-based is correct, in this cset it means if UDF or UUID is used; more cases could be added in later 5.1 release): SET GLOBAL|SESSION BINLOG_FORMAT=row|statement|mixed|default; the global default is statement unless cluster is enabled (then it's row) as in 5.1-alpha. It's not possible to use SET on this variable if a session is currently in row-based mode and has open temporary tables (because CREATE TEMPORARY TABLE was not binlogged so temp table is not known on slave), or if NDB is enabled (because NDB does not support such change on-the-fly, though it will later), of if in a stored function (see below). The added tests test the possibility or impossibility to SET, their effects, and the mixed mode, including in prepared statements and in stored procedures and functions. Caveats: a) The mixed mode will not work for stored functions: in mixed mode, a stored function will always be binlogged as one call and in a statement-based way (e.g. INSERT VALUES(myfunc()) or SELECT myfunc()). b) for the same reason, changing the thread's binlog format inside a stored function is refused with an error message. c) the same problems apply to triggers; implementing b) for triggers will be done later (will ask Dmitri). Additionally, as the binlog format is now changeable by each user for his session, I remove the implication which was done at startup, where row-based automatically set log-bin-trust-routine-creators to 1 (not possible anymore as a user can now switch to stmt-based and do nasty things again), and automatically set --innodb-locks-unsafe-for-binlog to 1 (was anyway theoretically incorrect as it disabled phantom protection). Plus fixes for compiler warnings.
2006-02-25 22:21:03 +01:00
DBUG_ASSERT(!(mysql_bin_log.is_open() && thd->current_stmt_binlog_row_based &&
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
(create_info->options & HA_LEX_CREATE_TMP_TABLE)));
write_bin_log(thd, TRUE, thd->query, thd->query_length);
Fix for bug#22369: Alter table rename combined with other alterations causes lost tables Using RENAME clause combined with other clauses of ALTER TABLE led to data loss (the data was there but not accessible). This could happen if the changes do not change the table much. Adding and droppping of fields and indices was safe. Renaming a column with MODIFY or CHANGE was unsafe operation, if the actual column didn't change (changing from int to int, which is a noop) Depending on the storage engine (SE) the behavior is different: 1)MyISAM/MEMORY - the ALTER TABLE statement completes without any error but next SELECT against the new table fails. 2)InnoDB (and every other transactional table) - The ALTER TABLE statement fails. There are the the following files in the db dir - `new_table_name.frm` and a temporary table's frm. If the SE is file based, then the data and index files will be present but with the old names. What happens is that for InnoDB the table is not renamed in the internal DDIC. Fixed by adding additional call to mysql_rename_table() method, which should not include FRM file rename, because it has been already done during file names juggling.
2006-12-04 18:22:38 +01:00
Make storage engines "pluggable", handlerton work
2005-11-07 16:25:06 +01:00
if (ha_check_storage_engine_flag(old_db_type,HTON_FLUSH_AFTER_RENAME))
Little but nasty bug with BDB and alter table
2002-08-28 15:41:23 +02:00
{
Some trivial optimzations Check if AND/OR expression can be NULL; Fixed bug in GROUP BY and-or-expression where expression could be NULL Bug fix for SHOW OPEN TABLES when user didn't have privilege to access all open tables. Better fix for ALTER TABLE on BDB tables.
2002-09-03 14:44:25 +02:00
/*
For the alter table to be properly flushed to the logs, we
have to open the new table. If not, we get a problem on server
shutdown.
*/
Fixed problem with BDB tables and ALTER TABLE
2002-10-24 00:29:29 +02:00
char path[FN_REFLEN];
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
build_table_filename(path, sizeof(path), new_db, table_name, "", 0);
Fixed problem with BDB tables and ALTER TABLE
2002-10-24 00:29:29 +02:00
table=open_temporary_table(thd, path, new_db, tmp_name,0);
if (table)
Some trivial optimzations Check if AND/OR expression can be NULL; Fixed bug in GROUP BY and-or-expression where expression could be NULL Bug fix for SHOW OPEN TABLES when user didn't have privilege to access all open tables. Better fix for ALTER TABLE on BDB tables.
2002-09-03 14:44:25 +02:00
{
Fixed problem with BDB tables and ALTER TABLE
2002-10-24 00:29:29 +02:00
intern_close_table(table);
my_free((char*) table, MYF(0));
Some trivial optimzations Check if AND/OR expression can be NULL; Fixed bug in GROUP BY and-or-expression where expression could be NULL Bug fix for SHOW OPEN TABLES when user didn't have privilege to access all open tables. Better fix for ALTER TABLE on BDB tables.
2002-09-03 14:44:25 +02:00
}
Fixed problem with BDB tables and ALTER TABLE
2002-10-24 00:29:29 +02:00
else
Make storage engines "pluggable", handlerton work
2005-11-07 16:25:06 +01:00
sql_print_warning("Could not open table %s.%s after rename\n",
sql_print_error cleanup
2004-09-04 20:17:09 +02:00
new_db,table_name);
Make storage engines "pluggable", handlerton work
2005-11-07 16:25:06 +01:00
ha_flush_logs(old_db_type);
Little but nasty bug with BDB and alter table
2002-08-28 15:41:23 +02:00
}
Update of query cache code
2001-12-06 00:05:30 +01:00
table_list->table=0; // For query cache
added building without query cache
2002-03-22 21:55:08 +01:00
query_cache_invalidate3(thd, table_list, 0);
Import changeset
2000-07-31 21:29:14 +02:00
end_temporary:
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
my_snprintf(tmp_name, sizeof(tmp_name), ER(ER_INSERT_INFO),
(ulong) (copied + deleted), (ulong) deleted,
(ulong) thd->cuted_fields);
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
if (do_send_ok)
send_ok(thd,copied+deleted,0L,tmp_name);
Import changeset
2000-07-31 21:29:14 +02:00
thd->some_tables_deleted=0;
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(FALSE);
Import changeset
2000-07-31 21:29:14 +02:00
Merge zippy.cornsilk.net:/home/cmiller/work/mysql/bug14262/my50-bug14262 into zippy.cornsilk.net:/home/cmiller/work/mysql/bug14262/my51
2006-10-11 22:49:59 +02:00
err1:
Bug#18129 - Fast (online) add index leaves temporary table frm in case of errors ALTER TABLE temporarily creates a new table with a .frm file and optionally other files. For fast ALTER TABLE only the .frm file is created. If the operation succeeds, The temporary files are renamed to their final target. In case of an error, the temporary file was forgotten to remove. Manually tested. The test requires to look at files, which I think cannot be done portably with the test suite. The test file is attached to the bug report.
2006-04-13 16:19:19 +02:00
if (new_table)
{
/* close_temporary_table() frees the new_table pointer. */
close_temporary_table(thd, new_table, 1, 1);
}
else
Bug#18775 - Temporary table from alter table visible to other threads Continued implementation of WL#1324 (table name to filename encoding) The intermediate (not temporary) files of the new table during ALTER TABLE was visible for SHOW TABLES. These intermediate files are copies of the original table with the changes done by ALTER TABLE. After all the data is copied over from the original table, these files are renamed to the original tables file names. So they are not temporary files. They persist after ALTER TABLE, but just with another name. In 5.0 the intermediate files are invisible for SHOW TABLES because all file names beginning with "#sql" were suppressed. This failed since 5.1.6 because even temporary table names were converted when making file names from them. The prefix became converted to "@0023sql". Converting the prefix during SHOW TABLES would suppress the listing of user tables that start with "#sql". The solution of the problem is to continue the implementation of the table name to file name conversion feature. One requirement is to suppress the conversion for temporary table names. This change is straightforward for real temporary tables as there is a function that creates temporary file names. But the generated path names are located in TMPDIR and have no relation to the internal table name. This cannot be used for ALTER TABLE. Its intermediate files need to be in the same directory as the old table files. And it is necessary to be able to deduce the same path from the same table name repeatedly. Consequently the intermediate table files must be handled like normal tables. Their internal names shall start with tmp_file_prefix (#sql) and they shall not be converted like normal table names. I added a flags parameter to all relevant functions that are called from ALTER TABLE. It is used to suppress the conversion for the intermediate table files. The outcome is that the suppression of #sql in SHOW TABLES works again. It does not suppress user tables as these are converted to @0023sql on file level. This patch does also fix ALTER TABLE ... RENAME, which could not rename a table with non-ASCII characters in its name. It does also fix the problem that a user could create a table like `#sql-xxxx-yyyy`, where xxxx is mysqld's pid and yyyy is the thread ID of some other thread, which prevented this thread from running ALTER TABLE. Some of the above problems are mentioned in Bug 1405, which can be closed with this patch. This patch does also contain some minor fixes for other forgotten conversions. Still known problems are reported as bugs 21370, 21373, and 21387.
2006-08-02 17:57:06 +02:00
VOID(quick_rm_table(new_db_type, new_db, tmp_name, FN_IS_TMP));
Bug#18129 - Fast (online) add index leaves temporary table frm in case of errors ALTER TABLE temporarily creates a new table with a .frm file and optionally other files. For fast ALTER TABLE only the .frm file is created. If the operation succeeds, The temporary files are renamed to their final target. In case of an error, the temporary file was forgotten to remove. Manually tested. The test requires to look at files, which I think cannot be done portably with the test suite. The test file is attached to the bug report.
2006-04-13 16:19:19 +02:00
Bug #14262: SP: DROP PROCEDURE|VIEW (maybe more) write to binlog too late \ (race cond) It was possible for one thread to interrupt a Data Definition Language statement and thereby get messages to the binlog out of order. Consider: Connection 1: Drop Foo x Connection 2: Create or replace Foo x Connection 2: Log "Create or replace Foo x" Connection 1: Log "Drop Foo x" Local end would have Foo x, but the replicated slaves would not. The fix for this is to wrap all DDL and logging of a kind in the same mutex. Since we already use mutexes for the various parts of altering the server, this only entails moving the logging events down close to the action, inside the mutex protection.
2006-10-03 19:38:25 +02:00
err:
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
Import changeset
2000-07-31 21:29:14 +02:00
}
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
/* mysql_alter_table */
Import changeset
2000-07-31 21:29:14 +02:00
static int
Added ALTER TABLE ... ORDER BY ...
2000-11-11 19:27:34 +01:00
copy_data_between_tables(TABLE *from,TABLE *to,
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
List<create_field> &create,
Remove DUP_IGNORE from enum_duplicates and instead use a separate ignore flag This allows use to use INSERT IGNORE ... ON DUPLICATE ...
2004-12-31 11:04:35 +01:00
bool ignore,
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
uint order_num, ORDER *order,
Added ALTER TABLE ... ORDER BY ...
2000-11-11 19:27:34 +01:00
ha_rows *copied,
Fix for bug#24395: ALTER TABLE DISABLE KEYS doesn't work when modifying the table ENABLE|DISABLE KEYS combined with another ALTER TABLE option, different than RENAME TO did nothing. Also, if the table had disabled keys and was ALTER-ed then the end table was with enabled keys. Fixed by checking whether the table had disabled keys and enabling them in the copied table.
2006-11-28 18:27:32 +01:00
ha_rows *deleted,
enum enum_enable_or_disable keys_onoff)
Import changeset
2000-07-31 21:29:14 +02:00
{
int error;
Copy_field *copy,*copy_end;
ulong found_count,delete_count;
THD *thd= current_thd;
Added ALTER TABLE ... ORDER BY ...
2000-11-11 19:27:34 +01:00
uint length;
SORT_FIELD *sortorder;
READ_RECORD info;
TABLE_LIST tables;
List<Item> fields;
List<Item> all_fields;
Portability fixes for Ia64 and Sgi compiler DROP DATABASE in mysqladmin asks for confirmation (again) Added examined_rows to slow query log SHOW CREATE for TEMPORARY table
2001-04-25 21:44:27 +02:00
ha_rows examined_rows;
Portablity fixes & simple optimizations
2004-08-24 17:00:45 +02:00
bool auto_increment_field_copied= 0;
Fix for bug #5915 "ALTER TABLE behaves differently when converting column to auto_increment in 4.1". Now we are enforcing NO_AUTO_VALUE_ON_ZERO mode during ALTER TABLE only if we are converting one auto_increment column to another auto_increment column (this also includes most common case when we don't do anything with such column). Also now when we convert some column to TIMESTAMP NOT NULL column with ALTER TABLE we convert NULL values to current timestamp, (as we do this in INSERT). One can still get old behavior by setting system TIMESTAMP variable to 0.
2004-10-07 11:02:39 +02:00
ulong save_sql_mode;
WL#3146 "less locking in auto_increment": this is a cleanup patch for our current auto_increment handling: new names for auto_increment variables in THD, new methods to manipulate them (see sql_class.h), some move into handler::, causing less backup/restore work when executing substatements. This makes the logic hopefully clearer, less work is is needed in mysql_insert(). By cleaning up, using different variables for different purposes (instead of one for 3 things...), we fix those bugs, which someone may want to fix in 5.0 too: BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate statement-based" BUG#20341 "stored function inserting into one auto_increment puts bad data in slave" BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE" (now if a row is updated, LAST_INSERT_ID() will return its id) and re-fixes: BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT" (already fixed differently by Ramil in 4.1) Test of documented behaviour of mysql_insert_id() (there was no test). The behaviour changes introduced are: - LAST_INSERT_ID() now returns "the first autogenerated auto_increment value successfully inserted", instead of "the first autogenerated auto_increment value if any row was successfully inserted", see auto_increment.test. Same for mysql_insert_id(), see mysql_client_test.c. - LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY UPDATE, see auto_increment.test. Same for mysql_insert_id(), see mysql_client_test.c. - LAST_INSERT_ID() does not change if no autogenerated value was successfully inserted (it used to then be 0), see auto_increment.test. - if in INSERT SELECT no autogenerated value was successfully inserted, mysql_insert_id() now returns the id of the last inserted row (it already did this for INSERT VALUES), see mysql_client_test.c. - if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X (it already did this for INSERT VALUES), see mysql_client_test.c. - NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE, the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID influences not only the first row now. Additionally, when unlocking a table we check that the thread is not keeping a next_insert_id (as the table is unlocked that id is potentially out-of-date); forgetting about this next_insert_id is done in a new handler::ha_release_auto_increment(). Finally we prepare for engines capable of reserving finite-length intervals of auto_increment values: we store such intervals in THD. The next step (to be done by the replication team in 5.1) is to read those intervals from THD and actually store them in the statement-based binary log. NDB will be a good engine to test that.
2006-07-09 17:52:19 +02:00
ulonglong prev_insert_id;
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_ENTER("copy_data_between_tables");
bug#7761 - ndb does not autocommit during alter table
2005-01-10 11:17:01 +01:00
/*
Turn off recovery logging since rollback of an alter table is to
delete the new table so there is no need to log the changes to it.
This needs to be done before external_lock
*/
bug#7761 - ndb does not autcommit - postreview fixes
2005-01-10 11:33:08 +01:00
error= ha_enable_transaction(thd, FALSE);
bug#7761 - ndb does not autocommit during alter table
2005-01-10 11:17:01 +01:00
if (error)
DBUG_RETURN(-1);
bug#7761 - ndb does not autcommit - postreview fixes
2005-01-10 11:33:08 +01:00
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
if (!(copy= new Copy_field[to->s->fields]))
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(-1); /* purecov: inspected */
WL#3023 (RBR: Use locks in a statment-like manner): Interface changes pushed early. Separation of public and implementation interface for external_lock() in preparation for implementation.
2006-01-26 09:25:37 +01:00
if (to->file->ha_external_lock(thd, F_WRLCK))
bug#2831 - --extenral-locking does not fully work with --myisam-recover. Changed the semantics of open_count so that it is decremented at every unlock (if it was incremented due to data changes). So it indicates a crash, if it is non-zero after a lock. The table will then be repaired.
2004-09-18 20:33:39 +02:00
DBUG_RETURN(-1);
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
Fix for bug#24395: ALTER TABLE DISABLE KEYS doesn't work when modifying the table ENABLE|DISABLE KEYS combined with another ALTER TABLE option, different than RENAME TO did nothing. Also, if the table had disabled keys and was ALTER-ed then the end table was with enabled keys. Fixed by checking whether the table had disabled keys and enabling them in the copied table.
2006-11-28 18:27:32 +01:00
/* We need external lock before we can disable/enable keys */
alter_table_manage_keys(to, from->file->indexes_are_disabled(), keys_onoff);
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
/* We can abort alter table for any table type */
thd->no_trans_update= 0;
thd->abort_on_warning= !ignore && test(thd->variables.sql_mode &
(MODE_STRICT_TRANS_TABLES |
MODE_STRICT_ALL_TABLES));
Fixed for Ia64 + delayed key creation + a lot of small bug fixes
2000-08-15 19:09:37 +02:00
from->file->info(HA_STATUS_VARIABLE);
Merge bk-internal.mysql.com:/home/bk/mysql-5.1-new into mysql.com:/home/my/mysql-5.1
2006-06-04 20:05:22 +02:00
to->file->ha_start_bulk_insert(from->file->stats.records);
Import changeset
2000-07-31 21:29:14 +02:00
Manual merge of tree containing fix for bug #5915 "ALTER TABLE behaves differently when converting column to auto_increment in 4.1" with current tree.
2004-10-07 11:51:32 +02:00
save_sql_mode= thd->variables.sql_mode;
Import changeset
2000-07-31 21:29:14 +02:00
List_iterator<create_field> it(create);
create_field *def;
copy_end=copy;
for (Field **ptr=to->field ; *ptr ; ptr++)
{
def=it++;
if (def->field)
no_auto_value_on_zero + alter table bug
2004-07-30 14:17:12 +02:00
{
if (*ptr == to->next_number_field)
Fix for bug #5915 "ALTER TABLE behaves differently when converting column to auto_increment in 4.1". Now we are enforcing NO_AUTO_VALUE_ON_ZERO mode during ALTER TABLE only if we are converting one auto_increment column to another auto_increment column (this also includes most common case when we don't do anything with such column). Also now when we convert some column to TIMESTAMP NOT NULL column with ALTER TABLE we convert NULL values to current timestamp, (as we do this in INSERT). One can still get old behavior by setting system TIMESTAMP variable to 0.
2004-10-07 11:02:39 +02:00
{
Portablity fixes & simple optimizations
2004-08-24 17:00:45 +02:00
auto_increment_field_copied= TRUE;
Fix for bug #5915 "ALTER TABLE behaves differently when converting column to auto_increment in 4.1". Now we are enforcing NO_AUTO_VALUE_ON_ZERO mode during ALTER TABLE only if we are converting one auto_increment column to another auto_increment column (this also includes most common case when we don't do anything with such column). Also now when we convert some column to TIMESTAMP NOT NULL column with ALTER TABLE we convert NULL values to current timestamp, (as we do this in INSERT). One can still get old behavior by setting system TIMESTAMP variable to 0.
2004-10-07 11:02:39 +02:00
/*
If we are going to copy contents of one auto_increment column to
another auto_increment column it is sensible to preserve zeroes.
This condition also covers case when we are don't actually alter
auto_increment column.
*/
if (def->field == from->found_next_number_field)
thd->variables.sql_mode|= MODE_NO_AUTO_VALUE_ON_ZERO;
}
Import changeset
2000-07-31 21:29:14 +02:00
(copy_end++)->set(*ptr,def->field,0);
no_auto_value_on_zero + alter table bug
2004-07-30 14:17:12 +02:00
}
Import changeset
2000-07-31 21:29:14 +02:00
}
Fixed temp tables to work with replication. It will not work if you stop the slave while slave thread has temp tables - not hard to fix though, but it is time to go to bed added a new test case for temp table replication and Slave_open_temp_tables status variable. Misc fixes
2000-11-26 07:29:01 +01:00
found_count=delete_count=0;
Fixes for innobase Added test for ALTER TABLE ORDER BY
2001-02-21 13:16:00 +01:00
if (order)
{
Many files: Added the MAX_LENGTH_FOR_SORT_DATA option filesort.cc: Added the MAX_LENGTH_FOR_SORT_DATA option
2003-04-24 13:33:33 +02:00
from->sort.io_cache=(IO_CACHE*) my_malloc(sizeof(IO_CACHE),
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
MYF(MY_FAE | MY_ZEROFILL));
Added ALTER TABLE ... ORDER BY ...
2000-11-11 19:27:34 +01:00
bzero((char*) &tables,sizeof(tables));
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
tables.table= from;
Table definition cache, part 2 The table opening process now works the following way: - Create common TABLE_SHARE object - Read the .frm file and unpack it into the TABLE_SHARE object - Create a TABLE object based on the information in the TABLE_SHARE object and open a handler to the table object Other noteworthy changes: - In TABLE_SHARE the most common strings are now LEX_STRING's - Better error message when table is not found - Variable table_cache is now renamed 'table_open_cache' - New variable 'table_definition_cache' that is the number of table defintions that will be cached - strxnmov() calls are now fixed to avoid overflows - strxnmov() will now always add one end \0 to result - engine objects are now created with a TABLE_SHARE object instead of a TABLE object. - After creating a field object one must call field->init(table) before using it - For a busy system this change will give you: - Less memory usage for table object - Faster opening of tables (if it's has been in use or is in table definition cache) - Allow you to cache many table definitions objects - Faster drop of table
2005-11-23 21:45:02 +01:00
tables.alias= tables.table_name= from->s->table_name.str;
tables.db= from->s->db.str;
Added ALTER TABLE ... ORDER BY ...
2000-11-11 19:27:34 +01:00
error=1;
Post-merge fixes.
2003-08-26 17:41:40 +02:00
if (thd->lex->select_lex.setup_ref_array(thd, order_num) ||
made lex a pointer in THD
2003-05-05 20:54:37 +02:00
setup_order(thd, thd->lex->select_lex.ref_pointer_array,
fixed subselects with temporary tables (SCRUM) fixed memory leacks
2003-01-25 01:25:52 +01:00
&tables, fields, all_fields, order) ||
Bug#21798: memory leak during query execution with subquery in column list using a function When executing dependent subqueries they are re-inited and re-exec() for each row of the outer context. The cause for the bug is that during subquery reinitialization/re-execution, the optimizer reallocates JOIN::join_tab, JOIN::table in make_simple_join() and the local variable in 'sortorder' in create_sort_index(), which is allocated by make_unireg_sortorder(). Care must be taken not to allocate anything into the thread's memory pool while re-initializing query plan structures between subquery re-executions. All such items mush be cached and reused because the thread's memory pool is freed at the end of the whole query. Note that they must be cached and reused even for queries that are not otherwise cacheable because otherwise it will grow the thread's memory pool every time a cacheable query is re-executed. We provide additional members to the JOIN structure to store references to the items that need to be cached.
2006-10-17 15:20:26 +02:00
!(sortorder=make_unireg_sortorder(order, &length, NULL)) ||
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
(from->sort.found_records = filesort(thd, from, sortorder, length,
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
(SQL_SELECT *) 0, HA_POS_ERROR, 1,
Merge with 4.1
2005-05-14 17:31:22 +02:00
&examined_rows)) ==
HA_POS_ERROR)
Added ALTER TABLE ... ORDER BY ...
2000-11-11 19:27:34 +01:00
goto err;
};
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
/* Tell handler that we have values for all columns in the to table */
to->use_all_columns();
Import changeset
2000-07-31 21:29:14 +02:00
init_read_record(&info, thd, from, (SQL_SELECT *) 0, 1,1);
Fix for bug#18437 "Wrong values inserted with a before update trigger on NDB table". SQL-layer was not marking fields which were used in triggers as such. As result these fields were not always properly retrieved/stored by handler layer. So one might got wrong values or lost changes in triggers for NDB, Federated and possibly InnoDB tables. This fix solves the problem by marking fields used in triggers appropriately. Also this patch contains the following cleanup of ha_ndbcluster code: We no longer rely on reading LEX::sql_command value in handler in order to determine if we can enable optimization which allows us to handle REPLACE statement in more efficient way by doing replaces directly in write_row() method without reporting error to SQL-layer. Instead we rely on SQL-layer informing us whether this optimization applicable by calling handler::extra() method with HA_EXTRA_WRITE_CAN_REPLACE flag. As result we no longer apply this optimzation in cases when it should not be used (e.g. if we have on delete triggers on table) and use in some additional cases when it is applicable (e.g. for LOAD DATA REPLACE). Finally this patch includes fix for bug#20728 "REPLACE does not work correctly for NDB table with PK and unique index". This was yet another problem which was caused by improper field mark-up. During row replacement fields which weren't explicity used in REPLACE statement were not marked as fields to be saved (updated) so they have retained values from old row version. The fix is to mark all table fields as set for REPLACE statement. Note that in 5.1 we already solve this problem by notifying handler that it should save values from all fields only in case when real replacement happens.
2006-07-01 23:51:10 +02:00
if (ignore)
New benchmark test Fixed bug in REPLACE with BDB tables Prepare for write lock on read for BDB Inform the handler when we want to use IGNORE / REPLACE New manual pages
2000-12-24 14:19:00 +01:00
to->file->extra(HA_EXTRA_IGNORE_DUP_KEY);
Support of warnings for all DML statements (Insert, Update and Alter) Fix LOAD DATA INFILE warnings to have a better meanigful messages Fix to make the mysql command line to automatically show the warnings count for all basic commands
2003-04-30 09:02:28 +02:00
thd->row_count= 0;
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
restore_record(to, s->default_values); // Create empty record
Import changeset
2000-07-31 21:29:14 +02:00
while (!(error=info.read_record(&info)))
{
if (thd->killed)
{
SCRUM two KILL versions code trimming with headquarter's suggestions
2003-04-08 16:18:33 +02:00
thd->send_kill_message();
Import changeset
2000-07-31 21:29:14 +02:00
error= 1;
break;
}
Support of warnings for all DML statements (Insert, Update and Alter) Fix LOAD DATA INFILE warnings to have a better meanigful messages Fix to make the mysql command line to automatically show the warnings count for all basic commands
2003-04-30 09:02:28 +02:00
thd->row_count++;
no_auto_value_on_zero + alter table bug
2004-07-30 14:17:12 +02:00
if (to->next_number_field)
{
Portablity fixes & simple optimizations
2004-08-24 17:00:45 +02:00
if (auto_increment_field_copied)
no_auto_value_on_zero + alter table bug
2004-07-30 14:17:12 +02:00
to->auto_increment_field_not_null= TRUE;
Portablity fixes & simple optimizations
2004-08-24 17:00:45 +02:00
else
to->next_number_field->reset();
}
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
Import changeset
2000-07-31 21:29:14 +02:00
for (Copy_field *copy_ptr=copy ; copy_ptr != copy_end ; copy_ptr++)
Portablity fixes & simple optimizations
2004-08-24 17:00:45 +02:00
{
Import changeset
2000-07-31 21:29:14 +02:00
copy_ptr->do_copy(copy_ptr);
no_auto_value_on_zero + alter table bug
2004-07-30 14:17:12 +02:00
}
WL#3146 "less locking in auto_increment": this is a cleanup patch for our current auto_increment handling: new names for auto_increment variables in THD, new methods to manipulate them (see sql_class.h), some move into handler::, causing less backup/restore work when executing substatements. This makes the logic hopefully clearer, less work is is needed in mysql_insert(). By cleaning up, using different variables for different purposes (instead of one for 3 things...), we fix those bugs, which someone may want to fix in 5.0 too: BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate statement-based" BUG#20341 "stored function inserting into one auto_increment puts bad data in slave" BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE" (now if a row is updated, LAST_INSERT_ID() will return its id) and re-fixes: BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT" (already fixed differently by Ramil in 4.1) Test of documented behaviour of mysql_insert_id() (there was no test). The behaviour changes introduced are: - LAST_INSERT_ID() now returns "the first autogenerated auto_increment value successfully inserted", instead of "the first autogenerated auto_increment value if any row was successfully inserted", see auto_increment.test. Same for mysql_insert_id(), see mysql_client_test.c. - LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY UPDATE, see auto_increment.test. Same for mysql_insert_id(), see mysql_client_test.c. - LAST_INSERT_ID() does not change if no autogenerated value was successfully inserted (it used to then be 0), see auto_increment.test. - if in INSERT SELECT no autogenerated value was successfully inserted, mysql_insert_id() now returns the id of the last inserted row (it already did this for INSERT VALUES), see mysql_client_test.c. - if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X (it already did this for INSERT VALUES), see mysql_client_test.c. - NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE, the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID influences not only the first row now. Additionally, when unlocking a table we check that the thread is not keeping a next_insert_id (as the table is unlocked that id is potentially out-of-date); forgetting about this next_insert_id is done in a new handler::ha_release_auto_increment(). Finally we prepare for engines capable of reserving finite-length intervals of auto_increment values: we store such intervals in THD. The next step (to be done by the replication team in 5.1) is to read those intervals from THD and actually store them in the statement-based binary log. NDB will be a good engine to test that.
2006-07-09 17:52:19 +02:00
prev_insert_id= to->file->next_insert_id;
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
if ((error=to->file->ha_write_row((byte*) to->record[0])))
Import changeset
2000-07-31 21:29:14 +02:00
{
Fix for bug#18437 "Wrong values inserted with a before update trigger on NDB table". SQL-layer was not marking fields which were used in triggers as such. As result these fields were not always properly retrieved/stored by handler layer. So one might got wrong values or lost changes in triggers for NDB, Federated and possibly InnoDB tables. This fix solves the problem by marking fields used in triggers appropriately. Also this patch contains the following cleanup of ha_ndbcluster code: We no longer rely on reading LEX::sql_command value in handler in order to determine if we can enable optimization which allows us to handle REPLACE statement in more efficient way by doing replaces directly in write_row() method without reporting error to SQL-layer. Instead we rely on SQL-layer informing us whether this optimization applicable by calling handler::extra() method with HA_EXTRA_WRITE_CAN_REPLACE flag. As result we no longer apply this optimzation in cases when it should not be used (e.g. if we have on delete triggers on table) and use in some additional cases when it is applicable (e.g. for LOAD DATA REPLACE). Finally this patch includes fix for bug#20728 "REPLACE does not work correctly for NDB table with PK and unique index". This was yet another problem which was caused by improper field mark-up. During row replacement fields which weren't explicity used in REPLACE statement were not marked as fields to be saved (updated) so they have retained values from old row version. The fix is to mark all table fields as set for REPLACE statement. Note that in 5.1 we already solve this problem by notifying handler that it should save values from all fields only in case when real replacement happens.
2006-07-01 23:51:10 +02:00
if (!ignore ||
BUG#17138: Crashes in stored procedure Last round of review fixes
2006-07-01 06:01:37 +02:00
to->file->is_fatal_error(error, HA_CHECK_DUP))
Import changeset
2000-07-31 21:29:14 +02:00
{
BUG#17138: Crashes in stored procedure Last round of review fixes
2006-07-01 06:01:37 +02:00
if (!to->file->is_fatal_error(error, HA_CHECK_DUP))
bug #14573 (Error on adding auto_increment on to a column with '0' values)
2006-05-12 15:02:42 +02:00
{
uint key_nr= to->file->get_dup_key(error);
if ((int) key_nr >= 0)
{
Give warnings for unused objects Changed error message to be compatible with old error file Added new error message for new DUP_ENTRY syntax
2007-01-22 17:42:52 +01:00
const char *err_msg= ER(ER_DUP_ENTRY_WITH_KEY_NAME);
bug #14573 (Error on adding auto_increment on to a column with '0' values)
2006-05-12 15:02:42 +02:00
if (key_nr == 0 &&
After merge fixes Remove compiler warnings
2006-06-05 05:16:08 +02:00
(to->key_info[0].key_part[0].field->flags &
AUTO_INCREMENT_FLAG))
bug #14573 (Error on adding auto_increment on to a column with '0' values)
2006-05-12 15:02:42 +02:00
err_msg= ER(ER_DUP_ENTRY_AUTOINCREMENT_CASE);
After merge fixes Remove compiler warnings
2006-06-05 05:16:08 +02:00
to->file->print_keydup_error(key_nr, err_msg);
bug #14573 (Error on adding auto_increment on to a column with '0' values)
2006-05-12 15:02:42 +02:00
break;
}
}
Import changeset
2000-07-31 21:29:14 +02:00
to->file->print_error(error,MYF(0));
break;
}
WL#3146 "less locking in auto_increment": this is a cleanup patch for our current auto_increment handling: new names for auto_increment variables in THD, new methods to manipulate them (see sql_class.h), some move into handler::, causing less backup/restore work when executing substatements. This makes the logic hopefully clearer, less work is is needed in mysql_insert(). By cleaning up, using different variables for different purposes (instead of one for 3 things...), we fix those bugs, which someone may want to fix in 5.0 too: BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate statement-based" BUG#20341 "stored function inserting into one auto_increment puts bad data in slave" BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE" (now if a row is updated, LAST_INSERT_ID() will return its id) and re-fixes: BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT" (already fixed differently by Ramil in 4.1) Test of documented behaviour of mysql_insert_id() (there was no test). The behaviour changes introduced are: - LAST_INSERT_ID() now returns "the first autogenerated auto_increment value successfully inserted", instead of "the first autogenerated auto_increment value if any row was successfully inserted", see auto_increment.test. Same for mysql_insert_id(), see mysql_client_test.c. - LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY UPDATE, see auto_increment.test. Same for mysql_insert_id(), see mysql_client_test.c. - LAST_INSERT_ID() does not change if no autogenerated value was successfully inserted (it used to then be 0), see auto_increment.test. - if in INSERT SELECT no autogenerated value was successfully inserted, mysql_insert_id() now returns the id of the last inserted row (it already did this for INSERT VALUES), see mysql_client_test.c. - if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X (it already did this for INSERT VALUES), see mysql_client_test.c. - NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE, the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID influences not only the first row now. Additionally, when unlocking a table we check that the thread is not keeping a next_insert_id (as the table is unlocked that id is potentially out-of-date); forgetting about this next_insert_id is done in a new handler::ha_release_auto_increment(). Finally we prepare for engines capable of reserving finite-length intervals of auto_increment values: we store such intervals in THD. The next step (to be done by the replication team in 5.1) is to read those intervals from THD and actually store them in the statement-based binary log. NDB will be a good engine to test that.
2006-07-09 17:52:19 +02:00
to->file->restore_auto_increment(prev_insert_id);
Import changeset
2000-07-31 21:29:14 +02:00
delete_count++;
}
else
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
found_count++;
Import changeset
2000-07-31 21:29:14 +02:00
}
end_read_record(&info);
Added new mutex for hostname lookup Added free of io_cache when using ALTER TABLE ... ORDER BY
2001-01-23 00:25:07 +01:00
free_io_cache(from);
Updated manual about embedded version. Speed up column-completion in 'mysql' Don't use ISAM if HAVE_ISAM is not defined A lot of fixes for the embedded version. All libraries are now included in libmysqld.a Changed arguments to convert_dirname() to make it more general. Renamed files in the 'merge' directory to all use a common prefix. Don't compile both assembler and C functions on x86
2001-10-08 03:58:07 +02:00
delete [] copy; // This is never 0
post-review fixes HA_EXTRA_WRITE_CACHE now part of start_bulk_insert() test cleanups
2004-04-07 16:04:28 +02:00
First push for WL#3146 "less locking in auto_increment". It is a 0-real-change patch. New prototype for get_auto_increment() (but new arguments not yet used), to be able to reserve a finite interval of auto_increment values from cooperating engines. A hint on how many values to reserve is found in handler::estimation_rows_to_insert, filled by ha_start_bulk_insert(), new wrapper around start_bulk_insert(). NOTE: this patch changes nothing, for all engines. But it makes the API ready for those engines which will want to do reservation. More csets will come to complete WL#3146.
2006-06-02 22:21:32 +02:00
if (to->file->ha_end_bulk_insert() && error <= 0)
Import changeset
2000-07-31 21:29:14 +02:00
{
post-review fixes HA_EXTRA_WRITE_CACHE now part of start_bulk_insert() test cleanups
2004-04-07 16:04:28 +02:00
to->file->print_error(my_errno,MYF(0));
Import changeset
2000-07-31 21:29:14 +02:00
error=1;
}
New benchmark test Fixed bug in REPLACE with BDB tables Prepare for write lock on read for BDB Inform the handler when we want to use IGNORE / REPLACE New manual pages
2000-12-24 14:19:00 +01:00
to->file->extra(HA_EXTRA_NO_IGNORE_DUP_KEY);
All statements are now run in a sub transaction Give warning if we are using non transactional tables and do ROLLBACK Porting of crash-me and benchmarks to FrontBase
2000-11-24 00:51:18 +01:00
renamed ha_recovery_logging to ha_enable_transaction added tests to alter table for "large" alter tables and truncates in ndbcluster added debug printout in restart() in ndbcluster added flag THD::transaction.on to enable/disable transaction
2004-09-03 17:11:09 +02:00
ha_enable_transaction(thd,TRUE);
Setting MODE_NO_AUTO_VALUE_ON_ZERO at copying in copy_data_between_tables
2004-09-26 17:11:28 +02:00
All statements are now run in a sub transaction Give warning if we are using non transactional tables and do ROLLBACK Porting of crash-me and benchmarks to FrontBase
2000-11-24 00:51:18 +01:00
/*
Ensure that the new table is saved properly to disk so that we
can do a rename
*/
if (ha_commit_stmt(thd))
error=1;
if (ha_commit(thd))
error=1;
Reverted patch for new usage of open_count as it caused more problems than it solved Cleaned up patch for checking locks for multi-table updates
2004-10-06 00:24:21 +02:00
Added ALTER TABLE ... ORDER BY ...
2000-11-11 19:27:34 +01:00
err:
Fix for bug #5915 "ALTER TABLE behaves differently when converting column to auto_increment in 4.1". Now we are enforcing NO_AUTO_VALUE_ON_ZERO mode during ALTER TABLE only if we are converting one auto_increment column to another auto_increment column (this also includes most common case when we don't do anything with such column). Also now when we convert some column to TIMESTAMP NOT NULL column with ALTER TABLE we convert NULL values to current timestamp, (as we do this in INSERT). One can still get old behavior by setting system TIMESTAMP variable to 0.
2004-10-07 11:02:39 +02:00
thd->variables.sql_mode= save_sql_mode;
Invalid DEFAULT values for CREATE TABLE now generates errors. (Bug #5902) CAST() now produces warnings when casting a wrong INTEGER or CHAR values. This also applies to implicite string to number casts. (Bug #5912) ALTER TABLE now fails in STRICT mode if it generates warnings. Inserting a zero date in a DATE, DATETIME or TIMESTAMP column during TRADITIONAL mode now produces an error. (Bug #5933)
2005-04-01 14:04:50 +02:00
thd->abort_on_warning= 0;
Added ALTER TABLE ... ORDER BY ...
2000-11-11 19:27:34 +01:00
free_io_cache(from);
Import changeset
2000-07-31 21:29:14 +02:00
*copied= found_count;
*deleted=delete_count;
WL#3146 "less locking in auto_increment": this is a cleanup patch for our current auto_increment handling: new names for auto_increment variables in THD, new methods to manipulate them (see sql_class.h), some move into handler::, causing less backup/restore work when executing substatements. This makes the logic hopefully clearer, less work is is needed in mysql_insert(). By cleaning up, using different variables for different purposes (instead of one for 3 things...), we fix those bugs, which someone may want to fix in 5.0 too: BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate statement-based" BUG#20341 "stored function inserting into one auto_increment puts bad data in slave" BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE" (now if a row is updated, LAST_INSERT_ID() will return its id) and re-fixes: BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT" (already fixed differently by Ramil in 4.1) Test of documented behaviour of mysql_insert_id() (there was no test). The behaviour changes introduced are: - LAST_INSERT_ID() now returns "the first autogenerated auto_increment value successfully inserted", instead of "the first autogenerated auto_increment value if any row was successfully inserted", see auto_increment.test. Same for mysql_insert_id(), see mysql_client_test.c. - LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY UPDATE, see auto_increment.test. Same for mysql_insert_id(), see mysql_client_test.c. - LAST_INSERT_ID() does not change if no autogenerated value was successfully inserted (it used to then be 0), see auto_increment.test. - if in INSERT SELECT no autogenerated value was successfully inserted, mysql_insert_id() now returns the id of the last inserted row (it already did this for INSERT VALUES), see mysql_client_test.c. - if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X (it already did this for INSERT VALUES), see mysql_client_test.c. - NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE, the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID influences not only the first row now. Additionally, when unlocking a table we check that the thread is not keeping a next_insert_id (as the table is unlocked that id is potentially out-of-date); forgetting about this next_insert_id is done in a new handler::ha_release_auto_increment(). Finally we prepare for engines capable of reserving finite-length intervals of auto_increment values: we store such intervals in THD. The next step (to be done by the replication team in 5.1) is to read those intervals from THD and actually store them in the statement-based binary log. NDB will be a good engine to test that.
2006-07-09 17:52:19 +02:00
to->file->ha_release_auto_increment();
WL#3023 (RBR: Use locks in a statment-like manner): Interface changes pushed early. Separation of public and implementation interface for external_lock() in preparation for implementation.
2006-01-26 09:25:37 +01:00
if (to->file->ha_external_lock(thd,F_UNLCK))
bug#2831 - --extenral-locking does not fully work with --myisam-recover. Changed the semantics of open_count so that it is decremented at every unlock (if it was incremented due to data changes). So it indicates a crash, if it is non-zero after a lock. The table will then be repaired.
2004-09-18 20:33:39 +02:00
error=1;
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(error > 0 ? -1 : 0);
}
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
After merge fixes. Note that mix_innodb_myisam_binlog and union fails after this patch (Will be fixed shortly by maintaners of this code)
2003-09-11 18:06:23 +02:00
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
/*
Recreates tables by calling mysql_alter_table().
SYNOPSIS
mysql_recreate_table()
thd Thread handler
tables Tables to recreate
do_send_ok If we should send_ok() or leave it to caller
RETURN
Like mysql_alter_table().
*/
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
bool mysql_recreate_table(THD *thd, TABLE_LIST *table_list,
bool do_send_ok)
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
{
DBUG_ENTER("mysql_recreate_table");
LEX *lex= thd->lex;
HA_CREATE_INFO create_info;
lex->create_list.empty();
lex->key_list.empty();
lex->col_list.empty();
lex->alter_info.reset();
bzero((char*) &create_info,sizeof(create_info));
handlerton cleanup: duplicate fields removed, st_mysql_storage_engine added to support run-time handlerton initialization (no compiler warnings), handler API is now tied to MySQL version, handlerton->plugin mapping added (slot-based), dummy default_hton removed, plugin-type-specific initialization generalized, built-in plugins are now initialized too, --default-storage-engine no longer needs a list of storage engines in handle_options(). mysql-test-run.pl bugfixes
2006-05-28 14:51:01 +02:00
create_info.db_type= 0;
typo fixed
2005-01-02 14:23:34 +01:00
create_info.row_type=ROW_TYPE_NOT_USED;
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
create_info.default_table_charset=default_charset_info;
After merge fixes Added push_back(void *, MEM_ROOT *) to make some list-handling code easier that needs to be allocated in a different mem-root (Before one had to change thd->mem_root ; push_back(); restore mem_root.
2004-11-09 02:58:44 +01:00
/* Force alter table to recreate table */
Faster alter table code for 5.1.
2005-07-22 22:43:59 +02:00
lex->alter_info.flags= (ALTER_CHANGE_COLUMN | ALTER_RECREATE);
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
DBUG_RETURN(mysql_alter_table(thd, NullS, NullS, &create_info,
table_list, lex->create_list,
lex->key_list, 0, (ORDER *) 0,
Fix for bug#18437 "Wrong values inserted with a before update trigger on NDB table". SQL-layer was not marking fields which were used in triggers as such. As result these fields were not always properly retrieved/stored by handler layer. So one might got wrong values or lost changes in triggers for NDB, Federated and possibly InnoDB tables. This fix solves the problem by marking fields used in triggers appropriately. Also this patch contains the following cleanup of ha_ndbcluster code: We no longer rely on reading LEX::sql_command value in handler in order to determine if we can enable optimization which allows us to handle REPLACE statement in more efficient way by doing replaces directly in write_row() method without reporting error to SQL-layer. Instead we rely on SQL-layer informing us whether this optimization applicable by calling handler::extra() method with HA_EXTRA_WRITE_CAN_REPLACE flag. As result we no longer apply this optimzation in cases when it should not be used (e.g. if we have on delete triggers on table) and use in some additional cases when it is applicable (e.g. for LOAD DATA REPLACE). Finally this patch includes fix for bug#20728 "REPLACE does not work correctly for NDB table with PK and unique index". This was yet another problem which was caused by improper field mark-up. During row replacement fields which weren't explicity used in REPLACE statement were not marked as fields to be saved (updated) so they have retained values from old row version. The fix is to mark all table fields as set for REPLACE statement. Note that in 5.1 we already solve this problem by notifying handler that it should save values from all fields only in case when real replacement happens.
2006-07-01 23:51:10 +02:00
0, &lex->alter_info, do_send_ok));
WL#1689 "Map OPTIMIZE TABLE to ALTER TABLE for InnoDB" exactly it's mapped to "ALTER TABLE t; ANALYZE TABLE t;"
2004-06-10 16:41:24 +02:00
}
Don't use row level logging on optimize or repair table. (Fixes core dump in rpl_failed_optimize.test) Ensure we end active transcations if we do an admin command (like optimize, repair etc)
2006-05-05 19:08:40 +02:00
bool mysql_checksum_table(THD *thd, TABLE_LIST *tables,
HA_CHECK_OPT *check_opt)
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
{
TABLE_LIST *table;
List<Item> field_list;
Item *item;
Protocol *protocol= thd->protocol;
fixed debug information
2004-03-23 18:28:18 +01:00
DBUG_ENTER("mysql_checksum_table");
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
field_list.push_back(item = new Item_empty_string("Table", NAME_LEN*2));
item->maybe_null= 1;
field_list.push_back(item=new Item_int("Checksum",(longlong) 1,21));
item->maybe_null= 1;
Port of cursors to be pushed into 5.0 tree: - client side part is simple and may be considered stable - server side part now just joggles with THD state to save execution state and has no additional locking wisdom. Lot's of it are to be rewritten.
2004-08-03 12:32:21 +02:00
if (protocol->send_fields(&field_list,
Protocol::SEND_NUM_ROWS | Protocol::SEND_EOF))
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
for (table= tables; table; table= table->next_local)
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
{
char table_name[NAME_LEN*2+2];
CHECKSUM TABLE table1, table2, ... [ QUICK | EXTENDED ]
2003-09-03 11:34:32 +02:00
TABLE *t;
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
strxmov(table_name, table->db ,".", table->table_name, NullS);
After merge fixes. Note that mix_innodb_myisam_binlog and union fails after this patch (Will be fixed shortly by maintaners of this code)
2003-09-11 18:06:23 +02:00
CHECKSUM TABLE now allows concurrent inserts
2005-09-27 20:11:09 +02:00
t= table->table= open_ltable(thd, table, TL_READ);
After merge fixes. Note that mix_innodb_myisam_binlog and union fails after this patch (Will be fixed shortly by maintaners of this code)
2003-09-11 18:06:23 +02:00
thd->clear_error(); // these errors shouldn't get client
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
protocol->prepare_for_resend();
protocol->store(table_name, system_charset_info);
CHECKSUM TABLE table1, table2, ... [ QUICK | EXTENDED ]
2003-09-03 11:34:32 +02:00
if (!t)
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
{
After merge fixes. Note that mix_innodb_myisam_binlog and union fails after this patch (Will be fixed shortly by maintaners of this code)
2003-09-11 18:06:23 +02:00
/* Table didn't exist */
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
protocol->store_null();
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
thd->clear_error();
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
}
else
{
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
if (t->file->ha_table_flags() & HA_HAS_CHECKSUM &&
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
!(check_opt->flags & T_EXTEND))
protocol->store((ulonglong)t->file->checksum());
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
else if (!(t->file->ha_table_flags() & HA_HAS_CHECKSUM) &&
After merge fixes. Note that mix_innodb_myisam_binlog and union fails after this patch (Will be fixed shortly by maintaners of this code)
2003-09-11 18:06:23 +02:00
(check_opt->flags & T_QUICK))
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
protocol->store_null();
CHECKSUM TABLE table1, table2, ... [ QUICK | EXTENDED ]
2003-09-03 11:34:32 +02:00
else
{
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
/* calculating table's checksum */
ha_checksum crc= 0;
sql_table.cc, table.cc, table.h: fixed CHECKSUM TABLE to be independent from last - undefined in InnoDB - bits in the null bitmask bug#13710
2005-10-04 17:04:20 +02:00
uchar null_mask=256 - (1 << t->s->last_null_bit_pos);
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes Changes that requires code changes in other code of other storage engines. (Note that all changes are very straightforward and one should find all issues by compiling a --debug build and fixing all compiler errors and all asserts in field.cc while running the test suite), - New optional handler function introduced: reset() This is called after every DML statement to make it easy for a handler to statement specific cleanups. (The only case it's not called is if force the file to be closed) - handler::extra(HA_EXTRA_RESET) is removed. Code that was there before should be moved to handler::reset() - table->read_set contains a bitmap over all columns that are needed in the query. read_row() and similar functions only needs to read these columns - table->write_set contains a bitmap over all columns that will be updated in the query. write_row() and update_row() only needs to update these columns. The above bitmaps should now be up to date in all context (including ALTER TABLE, filesort()). The handler is informed of any changes to the bitmap after fix_fields() by calling the virtual function handler::column_bitmaps_signal(). If the handler does caching of these bitmaps (instead of using table->read_set, table->write_set), it should redo the caching in this code. as the signal() may be sent several times, it's probably best to set a variable in the signal and redo the caching on read_row() / write_row() if the variable was set. - Removed the read_set and write_set bitmap objects from the handler class - Removed all column bit handling functions from the handler class. (Now one instead uses the normal bitmap functions in my_bitmap.c instead of handler dedicated bitmap functions) - field->query_id is removed. One should instead instead check table->read_set and table->write_set if a field is used in the query. - handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now instead use table->read_set to check for which columns to retrieve. - If a handler needs to call Field->val() or Field->store() on columns that are not used in the query, one should install a temporary all-columns-used map while doing so. For this, we provide the following functions: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set); field->val(); dbug_tmp_restore_column_map(table->read_set, old_map); and similar for the write map: my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set); field->val(); dbug_tmp_restore_column_map(table->write_set, old_map); If this is not done, you will sooner or later hit a DBUG_ASSERT in the field store() / val() functions. (For not DBUG binaries, the dbug_tmp_restore_column_map() and dbug_tmp_restore_column_map() are inline dummy functions and should be optimized away be the compiler). - If one needs to temporary set the column map for all binaries (and not just to avoid the DBUG_ASSERT() in the Field::store() / Field::val() methods) one should use the functions tmp_use_all_columns() and tmp_restore_column_map() instead of the above dbug_ variants. - All 'status' fields in the handler base class (like records, data_file_length etc) are now stored in a 'stats' struct. This makes it easier to know what status variables are provided by the base handler. This requires some trivial variable names in the extra() function. - New virtual function handler::records(). This is called to optimize COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true. (stats.records is not supposed to be an exact value. It's only has to be 'reasonable enough' for the optimizer to be able to choose a good optimization path). - Non virtual handler::init() function added for caching of virtual constants from engine. - Removed has_transactions() virtual method. Now one should instead return HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support transactions. - The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument that is to be used with 'new handler_name()' to allocate the handler in the right area. The xxxx_create_handler() function is also responsible for any initialization of the object before returning. For example, one should change: static handler *myisam_create_handler(TABLE_SHARE *table) { return new ha_myisam(table); } -> static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root) { return new (mem_root) ha_myisam(table); } - New optional virtual function: use_hidden_primary_key(). This is called in case of an update/delete when (table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined but we don't have a primary key. This allows the handler to take precisions in remembering any hidden primary key to able to update/delete any found row. The default handler marks all columns to be read. - handler::table_flags() now returns a ulonglong (to allow for more flags). - New/changed table_flags() - HA_HAS_RECORDS Set if ::records() is supported - HA_NO_TRANSACTIONS Set if engine doesn't support transactions - HA_PRIMARY_KEY_REQUIRED_FOR_DELETE Set if we should mark all primary key columns for read when reading rows as part of a DELETE statement. If there is no primary key, all columns are marked for read. - HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some cases (based on table->read_set) - HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION. - HA_DUPP_POS Renamed to HA_DUPLICATE_POS - HA_REQUIRES_KEY_COLUMNS_FOR_DELETE Set this if we should mark ALL key columns for read when when reading rows as part of a DELETE statement. In case of an update we will mark all keys for read for which key part changed value. - HA_STATS_RECORDS_IS_EXACT Set this if stats.records is exact. (This saves us some extra records() calls when optimizing COUNT(*)) - Removed table_flags() - HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if handler::records() gives an exact count() and HA_STATS_RECORDS_IS_EXACT if stats.records is exact. - HA_READ_RND_SAME Removed (no one supported this one) - Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk() - Renamed handler::dupp_pos to handler::dup_pos - Removed not used variable handler::sortkey Upper level handler changes: - ha_reset() now does some overall checks and calls ::reset() - ha_table_flags() added. This is a cached version of table_flags(). The cache is updated on engine creation time and updated on open. MySQL level changes (not obvious from the above): - DBUG_ASSERT() added to check that column usage matches what is set in the column usage bit maps. (This found a LOT of bugs in current column marking code). - In 5.1 before, all used columns was marked in read_set and only updated columns was marked in write_set. Now we only mark columns for which we need a value in read_set. - Column bitmaps are created in open_binary_frm() and open_table_from_share(). (Before this was in table.cc) - handler::table_flags() calls are replaced with handler::ha_table_flags() - For calling field->val() you must have the corresponding bit set in table->read_set. For calling field->store() you must have the corresponding bit set in table->write_set. (There are asserts in all store()/val() functions to catch wrong usage) - thd->set_query_id is renamed to thd->mark_used_columns and instead of setting this to an integer value, this has now the values: MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE Changed also all variables named 'set_query_id' to mark_used_columns. - In filesort() we now inform the handler of exactly which columns are needed doing the sort and choosing the rows. - The TABLE_SHARE object has a 'all_set' column bitmap one can use when one needs a column bitmap with all columns set. (This is used for table->use_all_columns() and other places) - The TABLE object has 3 column bitmaps: - def_read_set Default bitmap for columns to be read - def_write_set Default bitmap for columns to be written - tmp_set Can be used as a temporary bitmap when needed. The table object has also two pointer to bitmaps read_set and write_set that the handler should use to find out which columns are used in which way. - count() optimization now calls handler::records() instead of using handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true). - Added extra argument to Item::walk() to indicate if we should also traverse sub queries. - Added TABLE parameter to cp_buffer_from_ref() - Don't close tables created with CREATE ... SELECT but keep them in the table cache. (Faster usage of newly created tables). New interfaces: - table->clear_column_bitmaps() to initialize the bitmaps for tables at start of new statements. - table->column_bitmaps_set() to set up new column bitmaps and signal the handler about this. - table->column_bitmaps_set_no_signal() for some few cases where we need to setup new column bitmaps but don't signal the handler (as the handler has already been signaled about these before). Used for the momement only in opt_range.cc when doing ROR scans. - table->use_all_columns() to install a bitmap where all columns are marked as use in the read and the write set. - table->default_column_bitmaps() to install the normal read and write column bitmaps, but not signaling the handler about this. This is mainly used when creating TABLE instances. - table->mark_columns_needed_for_delete(), table->mark_columns_needed_for_delete() and table->mark_columns_needed_for_insert() to allow us to put additional columns in column usage maps if handler so requires. (The handler indicates what it neads in handler->table_flags()) - table->prepare_for_position() to allow us to tell handler that it needs to read primary key parts to be able to store them in future table->position() calls. (This replaces the table->file->ha_retrieve_all_pk function) - table->mark_auto_increment_column() to tell handler are going to update columns part of any auto_increment key. - table->mark_columns_used_by_index() to mark all columns that is part of an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow it to quickly know that it only needs to read colums that are part of the key. (The handler can also use the column map for detecting this, but simpler/faster handler can just monitor the extra() call). - table->mark_columns_used_by_index_no_reset() to in addition to other columns, also mark all columns that is used by the given key. - table->restore_column_maps_after_mark_index() to restore to default column maps after a call to table->mark_columns_used_by_index(). - New item function register_field_in_read_map(), for marking used columns in table->read_map. Used by filesort() to mark all used columns - Maintain in TABLE->merge_keys set of all keys that are used in query. (Simplices some optimization loops) - Maintain Field->part_of_key_not_clustered which is like Field->part_of_key but the field in the clustered key is not assumed to be part of all index. (used in opt_range.cc for faster loops) - dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map() tmp_use_all_columns() and tmp_restore_column_map() functions to temporally mark all columns as usable. The 'dbug_' version is primarily intended inside a handler when it wants to just call Field:store() & Field::val() functions, but don't need the column maps set for any other usage. (ie:: bitmap_is_set() is never called) - We can't use compare_records() to skip updates for handlers that returns a partial column set and the read_set doesn't cover all columns in the write set. The reason for this is that if we have a column marked only for write we can't in the MySQL level know if the value changed or not. The reason this worked before was that MySQL marked all to be written columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden bug'. - open_table_from_share() does not anymore setup temporary MEM_ROOT object as a thread specific variable for the handler. Instead we send the to-be-used MEMROOT to get_new_handler(). (Simpler, faster code) Bugs fixed: - Column marking was not done correctly in a lot of cases. (ALTER TABLE, when using triggers, auto_increment fields etc) (Could potentially result in wrong values inserted in table handlers relying on that the old column maps or field->set_query_id was correct) Especially when it comes to triggers, there may be cases where the old code would cause lost/wrong values for NDB and/or InnoDB tables. - Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags: OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG. This allowed me to remove some wrong warnings about: "Some non-transactional changed tables couldn't be rolled back" - Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset (thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose some warnings about "Some non-transactional changed tables couldn't be rolled back") - Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table() which could cause delete_table to report random failures. - Fixed core dumps for some tests when running with --debug - Added missing FN_LIBCHAR in mysql_rm_tmp_tables() (This has probably caused us to not properly remove temporary files after crash) - slow_logs was not properly initialized, which could maybe cause extra/lost entries in slow log. - If we get an duplicate row on insert, change column map to read and write all columns while retrying the operation. This is required by the definition of REPLACE and also ensures that fields that are only part of UPDATE are properly handled. This fixed a bug in NDB and REPLACE where REPLACE wrongly copied some column values from the replaced row. - For table handler that doesn't support NULL in keys, we would give an error when creating a primary key with NULL fields, even after the fields has been automaticly converted to NOT NULL. - Creating a primary key on a SPATIAL key, would fail if field was not declared as NOT NULL. Cleanups: - Removed not used condition argument to setup_tables - Removed not needed item function reset_query_id_processor(). - Field->add_index is removed. Now this is instead maintained in (field->flags & FIELD_IN_ADD_INDEX) - Field->fieldnr is removed (use field->field_index instead) - New argument to filesort() to indicate that it should return a set of row pointers (not used columns). This allowed me to remove some references to sql_command in filesort and should also enable us to return column results in some cases where we couldn't before. - Changed column bitmap handling in opt_range.cc to be aligned with TABLE bitmap, which allowed me to use bitmap functions instead of looping over all fields to create some needed bitmaps. (Faster and smaller code) - Broke up found too long lines - Moved some variable declaration at start of function for better code readability. - Removed some not used arguments from functions. (setup_fields(), mysql_prepare_insert_check_table()) - setup_fields() now takes an enum instead of an int for marking columns usage. - For internal temporary tables, use handler::write_row(), handler::delete_row() and handler::update_row() instead of handler::ha_xxxx() for faster execution. - Changed some constants to enum's and define's. - Using separate column read and write sets allows for easier checking of timestamp field was set by statement. - Remove calls to free_io_cache() as this is now done automaticly in ha_reset() - Don't build table->normalized_path as this is now identical to table->path (after bar's fixes to convert filenames) - Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to do comparision with the 'convert-dbug-for-diff' tool. Things left to do in 5.1: - We wrongly log failed CREATE TABLE ... SELECT in some cases when using row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result) Mats has promised to look into this. - Test that my fix for CREATE TABLE ... SELECT is indeed correct. (I added several test cases for this, but in this case it's better that someone else also tests this throughly). Lars has promosed to do this.
2006-06-04 17:52:22 +02:00
t->use_all_columns();
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
handler interface cleanups: more logical table/index_flags return HA_ERR_WRONG_COMMAND instead of abstract methods where appropriate max_keys and other limits renamed to max_supported_keys/etc max_keys/etc are now wrappers to max_supported_keys/etc ha_index_init/ha_rnd_init/ha_index_end/ha_rnd_end are now wrappers to real {index,rnd}_{init,end} to enforce strict pairing
2004-06-23 12:29:05 +02:00
if (t->file->ha_rnd_init(1))
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
protocol->store_null();
else
{
Bug#12296 - CHECKSUM TABLE reports 0 for the table Skipping deleted records instead of breaking the loop during checksum calculation.
2005-08-29 17:08:41 +02:00
for (;;)
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
{
ha_checksum row_crc= 0;
Bug#12296 - CHECKSUM TABLE reports 0 for the table Skipping deleted records instead of breaking the loop during checksum calculation.
2005-08-29 17:08:41 +02:00
int error= t->file->rnd_next(t->record[0]);
if (unlikely(error))
{
if (error == HA_ERR_RECORD_DELETED)
continue;
break;
}
sql_table.cc, table.cc, table.h: fixed CHECKSUM TABLE to be independent from last - undefined in InnoDB - bits in the null bitmask bug#13710
2005-10-04 17:04:20 +02:00
if (t->s->null_bytes)
{
/* fix undefined null bits */
t->record[0][t->s->null_bytes-1] |= null_mask;
sql_table.cc: fixed CHECKSUM TABLE to be independent from the first 'deleted' bit in the null bitmask (undefined in the InnoDB) bug#13710
2005-10-05 19:36:20 +02:00
if (!(t->s->db_create_options & HA_OPTION_PACK_RECORD))
t->record[0][0] |= 1;
sql_table.cc, table.cc, table.h: fixed CHECKSUM TABLE to be independent from last - undefined in InnoDB - bits in the null bitmask bug#13710
2005-10-04 17:04:20 +02:00
row_crc= my_checksum(row_crc, t->record[0], t->s->null_bytes);
}
CHECKSUM TABLE table1, table2, ... [ QUICK | EXTENDED ]
2003-09-03 11:34:32 +02:00
First stage of table definition cache Split TABLE to TABLE and TABLE_SHARE (TABLE_SHARE is still allocated as part of table, will be fixed soon) Created Field::make_field() and made Field_num::make_field() to call this Added 'TABLE_SHARE->db' that points to database name; Changed all usage of table_cache_key as database name to use this instead Changed field->table_name to point to pointer to alias. This allows us to change alias for a table by just updating one pointer. Renamed TABLE_SHARE->real_name to table_name Renamed TABLE->table_name to alias Renamed TABLE_LIST->real_name to table_name
2005-01-06 12:00:13 +01:00
for (uint i= 0; i < t->s->fields; i++ )
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
{
Field *f= t->field[i];
This finishes the work (someone) started to remove FIELD_ types and use the Enum MYSQL types. The second part to this is to actually deprecate the FIELD defines in mysql_com.h
2006-12-02 02:26:52 +01:00
if ((f->type() == MYSQL_TYPE_BLOB) ||
Bug#8841 - CHECKSUM TABLE is broken in MyISAM There are (at least) two implementations of the checksum computation. One is in MyISAM for the quick checksum. It is executed on every row change. The other is in the SQL layer for the extended checksum. It retrieves all rows of a table via the respective storage engine. In former MySQL versions varchars were stored with their maximum length, but now with their real length similar to blobs. This change had been forgotten to take care of in the extended checksum calculation. Hence too much data was checksumed. In MyISAM this change had been taken care of already. Only the real data is included in the checksum. I changed mysql_checksum_table() so that it uses the length information of true varchar fields instead of the field length like in former varchar implementations.
2006-02-01 16:46:44 +01:00
(f->type() == MYSQL_TYPE_VARCHAR))
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
{
String tmp;
f->val_str(&tmp);
row_crc= my_checksum(row_crc, (byte*) tmp.ptr(), tmp.length());
}
else
row_crc= my_checksum(row_crc, (byte*) f->ptr,
Portability fixes for windows After merge fixes
2003-10-15 21:40:36 +02:00
f->pack_length());
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
}
CHECKSUM TABLE table1, table2, ... [ QUICK | EXTENDED ]
2003-09-03 11:34:32 +02:00
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
crc+= row_crc;
}
protocol->store((ulonglong)crc);
handler interface cleanups: more logical table/index_flags return HA_ERR_WRONG_COMMAND instead of abstract methods where appropriate max_keys and other limits renamed to max_supported_keys/etc max_keys/etc are now wrappers to max_supported_keys/etc ha_index_init/ha_rnd_init/ha_index_end/ha_rnd_end are now wrappers to real {index,rnd}_{init,end} to enforce strict pairing
2004-06-23 12:29:05 +02:00
t->file->ha_rnd_end();
Don't enable HA_EXTRA_WRITE_CACHE if too few rows Revert main parts of patch for online index builds. Should be done differently Added support for %lx in my_snprintf()
2004-04-08 16:56:45 +02:00
}
CHECKSUM TABLE table1, table2, ... [ QUICK | EXTENDED ]
2003-09-03 11:34:32 +02:00
}
After merge fixes. Note that mix_innodb_myisam_binlog and union fails after this patch (Will be fixed shortly by maintaners of this code)
2003-09-11 18:06:23 +02:00
thd->clear_error();
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
close_thread_tables(thd);
table->table=0; // For query cache
}
if (protocol->write())
goto err;
}
send_eof(thd);
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(FALSE);
After merge fixes. Note that mix_innodb_myisam_binlog and union fails after this patch (Will be fixed shortly by maintaners of this code)
2003-09-11 18:06:23 +02:00
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
err:
close_thread_tables(thd); // Shouldn't be needed
if (table)
table->table=0;
errors without code removed net_printf/send_error calls replaced by my_error family functions -1/1 (sent/unsent) error reporting removed (WL#2133)
2004-10-20 03:04:37 +02:00
DBUG_RETURN(TRUE);
fixed a crash on COMPRESS() and other zlib-dependent functions when compiled w/o zlib moved them all from different places to item_strfunc.{h,cc} checksum table command Com_xxx status variables updated
2003-08-21 16:15:06 +02:00
}
Bug#6877 MySQL should give an error if the requested table type is not available Implement new SQL mode - NO_ENGINE_SUBSTITUTION
2005-06-17 23:14:44 +02:00
static bool check_engine(THD *thd, const char *table_name,
Bug #17210 Create temp table call to ha_ndbcluster::create_handler_files caused core - do not try to create ndb temp tables even if default storage is ndb
2006-02-17 17:12:35 +01:00
HA_CREATE_INFO *create_info)
Bug#6877 MySQL should give an error if the requested table type is not available Implement new SQL mode - NO_ENGINE_SUBSTITUTION
2005-06-17 23:14:44 +02:00
{
Bug #17210 Create temp table call to ha_ndbcluster::create_handler_files caused core - do not try to create ndb temp tables even if default storage is ndb
2006-02-17 17:12:35 +01:00
handlerton **new_engine= &create_info->db_type;
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
handlerton *req_engine= *new_engine;
sql_table.cc, table.cc, table.h: fixed CHECKSUM TABLE to be independent from last - undefined in InnoDB - bits in the null bitmask bug#13710
2005-10-04 17:04:20 +02:00
bool no_substitution=
Bug#6877 MySQL should give an error if the requested table type is not available Implement new SQL mode - NO_ENGINE_SUBSTITUTION
2005-06-17 23:14:44 +02:00
test(thd->variables.sql_mode & MODE_NO_ENGINE_SUBSTITUTION);
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
if (!(*new_engine= ha_checktype(thd, ha_legacy_type(req_engine),
no_substitution, 1)))
Bug#6877 MySQL should give an error if the requested table type is not available Implement new SQL mode - NO_ENGINE_SUBSTITUTION
2005-06-17 23:14:44 +02:00
return TRUE;
handlerton cleanup: duplicate fields removed, st_mysql_storage_engine added to support run-time handlerton initialization (no compiler warnings), handler API is now tied to MySQL version, handlerton->plugin mapping added (slot-based), dummy default_hton removed, plugin-type-specific initialization generalized, built-in plugins are now initialized too, --default-storage-engine no longer needs a list of storage engines in handle_options(). mysql-test-run.pl bugfixes
2006-05-28 14:51:01 +02:00
if (req_engine && req_engine != *new_engine)
Bug#6877 MySQL should give an error if the requested table type is not available Implement new SQL mode - NO_ENGINE_SUBSTITUTION
2005-06-17 23:14:44 +02:00
{
push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_WARN,
ER_WARN_USING_OTHER_HANDLER,
ER(ER_WARN_USING_OTHER_HANDLER),
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
ha_resolve_storage_engine_name(*new_engine),
Bug#6877 MySQL should give an error if the requested table type is not available Implement new SQL mode - NO_ENGINE_SUBSTITUTION
2005-06-17 23:14:44 +02:00
table_name);
}
Bug #17210 Create temp table call to ha_ndbcluster::create_handler_files caused core - do not try to create ndb temp tables even if default storage is ndb
2006-02-17 17:12:35 +01:00
if (create_info->options & HA_LEX_CREATE_TMP_TABLE &&
ha_check_storage_engine_flag(*new_engine, HTON_TEMPORARY_NOT_SUPPORTED))
{
if (create_info->used_fields & HA_CREATE_USED_ENGINE)
{
handlerton cleanup: duplicate fields removed, st_mysql_storage_engine added to support run-time handlerton initialization (no compiler warnings), handler API is now tied to MySQL version, handlerton->plugin mapping added (slot-based), dummy default_hton removed, plugin-type-specific initialization generalized, built-in plugins are now initialized too, --default-storage-engine no longer needs a list of storage engines in handle_options(). mysql-test-run.pl bugfixes
2006-05-28 14:51:01 +02:00
my_error(ER_ILLEGAL_HA_CREATE_OPTION, MYF(0),
after review fixes
2006-05-31 18:07:32 +02:00
hton2plugin[(*new_engine)->slot]->name.str, "TEMPORARY");
Bug #17210 Create temp table call to ha_ndbcluster::create_handler_files caused core - do not try to create ndb temp tables even if default storage is ndb
2006-02-17 17:12:35 +01:00
*new_engine= 0;
return TRUE;
}
This changes the order of the universe, black is now the new white. In practice this means that handlerton is now created by the server and is passed to the engine. Plugin startups can now also control how plugins are inited (and can optionally pass values). Bit more flexibility to those who want to write plugin interfaces to the database.
2006-09-15 19:28:00 +02:00
*new_engine= myisam_hton;
Bug #17210 Create temp table call to ha_ndbcluster::create_handler_files caused core - do not try to create ndb temp tables even if default storage is ndb
2006-02-17 17:12:35 +01:00
}
Bug#6877 MySQL should give an error if the requested table type is not available Implement new SQL mode - NO_ENGINE_SUBSTITUTION
2005-06-17 23:14:44 +02:00
return FALSE;
}
Reference in a new issue Copy permalink