mirror/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-01-18 04:53:01 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions
mariadb/sql/table.cc

4081 lines
118 KiB
C++
Raw Normal View History

Import changeset
2000-07-31 21:29:14 +02:00
/* Copyright (C) 2000 MySQL AB & MySQL Finland AB & TCX DataKonsult AB
Update copyright Fixed memory leak on shutdown (Affects the embedded version & MyODBC)
2001-12-06 13:10:51 +01:00
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
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
Update copyright Fixed memory leak on shutdown (Affects the embedded version & MyODBC)
2001-12-06 13:10:51 +01:00
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.
Update copyright Fixed memory leak on shutdown (Affects the embedded version & MyODBC)
2001-12-06 13:10:51 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
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 */
/* Some general useful functions */
#include "mysql_priv.h"
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
#include "sql_trigger.h"
Import changeset
2000-07-31 21:29:14 +02:00
#include <m_ctype.h>
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
#include "md5.h"
Import changeset
2000-07-31 21:29:14 +02:00
/* Functions defined in this 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
void open_table_error(TABLE_SHARE *share, int error, int db_errno,
myf errortype, int errarg);
fixed reading of free'd memory when opening a table with non-existing plugin
2006-04-06 15:49:00 +02:00
static int open_binary_frm(THD *thd, TABLE_SHARE *share,
uchar *head, File file);
Import changeset
2000-07-31 21:29:14 +02:00
static void fix_type_pointers(const char ***array, TYPELIB *point_to_type,
uint types, char **names);
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 uint find_field(Field **fields, uint start, uint length);
Import changeset
2000-07-31 21:29:14 +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
/* Get column name from column hash */
static byte *get_field_name(Field **buff, uint *length,
Import changeset
2000-07-31 21:29:14 +02:00
my_bool not_used __attribute__((unused)))
{
"Fix" for BUG #2050 "10 to 1 performance drop with server 4.1.1". Actually it is not a bug but right behavior observed as pefomance degradation after we have forced Item_field::fix_fields() to re-execute each time when we are executing prep stmt. This patch implements small optimization which heals this bad behavior. We are caching field position in TABLE::field array in Item's member and are using this position for speeding up field lookups in fix_fields() in case of its re-execution.
2004-03-28 01:11:54 +01:00
*length= (uint) strlen((*buff)->field_name);
return (byte*) (*buff)->field_name;
Import changeset
2000-07-31 21:29:14 +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
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
/*
Returns pointer to '.frm' extension of the file name.
SYNOPSIS
fn_rext()
name file name
DESCRIPTION
Checks file name part starting with the rightmost '.' character,
and returns it if it is equal to '.frm'.
TODO
It is a good idea to get rid of this function modifying the code
to garantee that the functions presently calling fn_rext() always
get arguments in the same format: either with '.frm' or without '.frm'.
RETURN VALUES
Pointer to the '.frm' extension. If there is no extension,
or extension is not '.frm', pointer at the end of file name.
*/
char *fn_rext(char *name)
{
char *res= strrchr(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
if (res && !strcmp(res, reg_ext))
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
return res;
return name + strlen(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
/*
Allocate a setup TABLE_SHARE structure
SYNOPSIS
alloc_table_share()
TABLE_LIST Take database and table name from there
key Table cache key (db \0 table_name \0...)
key_length Length of key
RETURN
0 Error (out of memory)
# Share
*/
Fix byte * vs. char * mismatches that break Win32 compilation.
2005-11-29 09:07:21 +01:00
TABLE_SHARE *alloc_table_share(TABLE_LIST *table_list, char *key,
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
uint key_length)
{
MEM_ROOT mem_root;
TABLE_SHARE *share;
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
char path[FN_REFLEN];
uint path_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("alloc_table_share");
DBUG_PRINT("enter", ("table: '%s'.'%s'",
table_list->db, table_list->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
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) - 1,
table_list->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_list->table_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
init_sql_alloc(&mem_root, TABLE_ALLOC_BLOCK_SIZE, 0);
if ((share= (TABLE_SHARE*) alloc_root(&mem_root,
sizeof(*share) + key_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
path_length +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
{
bzero((char*) share, sizeof(*share));
share->table_cache_key.str= (char*) (share+1);
share->table_cache_key.length= key_length;
memcpy(share->table_cache_key.str, key, key_length);
/* Use the fact the key is db/0/table_name/0 */
share->db.str= share->table_cache_key.str;
share->db.length= strlen(share->db.str);
share->table_name.str= share->db.str + share->db.length + 1;
share->table_name.length= strlen(share->table_name.str);
share->path.str= share->table_cache_key.str+ key_length;
share->path.length= path_length;
strmov(share->path.str, path);
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
share->normalized_path.str= share->path.str;
share->normalized_path.length= path_length;
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
share->version= refresh_version;
share->flush_version= flush_version;
Bug#17181 (mysqlslap test server crash): Moving assignments to table_map_id for thread-safe handling of table shares.
2006-02-23 13:34:03 +01:00
#ifdef HAVE_ROW_BASED_REPLICATION
/*
This constant is used to mark that no table map version has been
assigned. No arithmetic is done on the value: it will be
overwritten with a value taken from MYSQL_BIN_LOG.
*/
share->table_map_version= ~(ulonglong)0;
/*
Since alloc_table_share() can be called without any locking (for
example, ha_create_table... functions), we do not assign a table
map id here. Instead we assign a value that is not used
elsewhere, and then assign a table map id inside open_table()
under the protection of the LOCK_open mutex.
*/
WL#3023 (Use locks in statement-like manner): Post-merge patches.
2006-03-09 03:56:14 +01:00
share->table_map_id= ~0UL;
Bug#18280 (RBR: Extreneous maps when writing to general_log and slow_log): Filter out replication general_log and slow_log entirely from binary log. Caching result of table share-specific tests.
2006-03-17 18:11:07 +01:00
share->cached_row_logging_check= -1;
Bug#17181 (mysqlslap test server crash): Moving assignments to table_map_id for thread-safe handling of table shares.
2006-02-23 13:34:03 +01:00
#endif
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
memcpy((char*) &share->mem_root, (char*) &mem_root, sizeof(mem_root));
pthread_mutex_init(&share->mutex, MY_MUTEX_INIT_FAST);
pthread_cond_init(&share->cond, NULL);
}
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(share);
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
}
/*
Initialize share for temporary tables
SYNOPSIS
init_tmp_table_share()
share Share to fill
key Table_cache_key, as generated from create_table_def_key.
must start with db name.
key_length Length of key
table_name Table name
path Path to file (possible in lower case) without .frm
NOTES
This is different from alloc_table_share() because temporary tables
don't have to be shared between threads or put into the table def
cache, so we can do some things notable simpler and faster
If table is not put in thd->temporary_tables (happens only when
one uses OPEN TEMPORARY) then one can specify 'db' as key and
use key_length= 0 as neither table_cache_key or key_length will be used).
*/
void init_tmp_table_share(TABLE_SHARE *share, const char *key,
uint key_length, const char *table_name,
const char *path)
{
DBUG_ENTER("init_tmp_table_share");
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", ("table: '%s'.'%s'", key, 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
bzero((char*) share, sizeof(*share));
init_sql_alloc(&share->mem_root, TABLE_ALLOC_BLOCK_SIZE, 0);
share->tmp_table= INTERNAL_TMP_TABLE;
share->db.str= (char*) key;
share->db.length= strlen(key);
share->table_cache_key.str= (char*) key;
share->table_cache_key.length= key_length;
share->table_name.str= (char*) table_name;
share->table_name.length= strlen(table_name);
share->path.str= (char*) path;
share->normalized_path.str= (char*) path;
share->path.length= share->normalized_path.length= strlen(path);
share->frm_version= FRM_VER_TRUE_VARCHAR;
Bug#17181 (mysqlslap test server crash): Moving assignments to table_map_id for thread-safe handling of table shares.
2006-02-23 13:34:03 +01:00
#ifdef HAVE_ROW_BASED_REPLICATION
/*
Temporary tables are not replicated, but we set up these fields
anyway to be able to catch errors.
*/
share->table_map_version= ~(ulonglong)0;
WL#3023 (Use locks in statement-like manner): Post-merge patches.
2006-03-09 03:56:14 +01:00
share->table_map_id= ~0UL;
Bug#18280 (RBR: Extreneous maps when writing to general_log and slow_log): Filter out replication general_log and slow_log entirely from binary log. Caching result of table share-specific tests.
2006-03-17 18:11:07 +01:00
share->cached_row_logging_check= -1;
Bug#17181 (mysqlslap test server crash): Moving assignments to table_map_id for thread-safe handling of table shares.
2006-02-23 13:34:03 +01:00
#endif
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_VOID_RETURN;
}
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
/*
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
Free table share and memory used by it
SYNOPSIS
free_table_share()
share Table share
NOTES
share->mutex must be locked when we come here if it's not a temp table
*/
void free_table_share(TABLE_SHARE *share)
{
MEM_ROOT mem_root;
DBUG_ENTER("free_table_share");
DBUG_PRINT("enter", ("table: %s.%s", share->db.str, share->table_name.str));
DBUG_ASSERT(share->ref_count == 0);
/*
If someone is waiting for this to be deleted, inform it about this.
Don't do a delete until we know that no one is refering to this anymore.
*/
if (share->tmp_table == NO_TMP_TABLE)
{
/* share->mutex is locked in release_table_share() */
while (share->waiting_on_cond)
{
pthread_cond_broadcast(&share->cond);
pthread_cond_wait(&share->cond, &share->mutex);
}
/* No thread refers to this anymore */
pthread_mutex_unlock(&share->mutex);
pthread_mutex_destroy(&share->mutex);
pthread_cond_destroy(&share->cond);
}
hash_free(&share->name_hash);
/* We must copy mem_root from share because share is allocated through it */
memcpy((char*) &mem_root, (char*) &share->mem_root, sizeof(mem_root));
free_root(&mem_root, MYF(0)); // Free's share
DBUG_VOID_RETURN;
}
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
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
/*
Read table definition from a binary / text based .frm file
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
SYNOPSIS
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
open_table_def()
thd Thread handler
share Fill this with table definition
db_flags Bit mask of the following flags: OPEN_VIEW
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
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
NOTES
This function is called when the table definition is not cached in
table_def_cache
The data is returned in 'share', which is alloced by
alloc_table_share().. The code assumes that share is initialized.
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
RETURN VALUES
0 ok
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
1 Error (see open_table_error)
2 Error (see open_table_error)
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
3 Wrong data in .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
4 Error (see open_table_error)
5 Error (see open_table_error: charset unavailable)
Add special error message for .frm files with an un-understood version, like those from 5.0 with new datatypes. (Bug #10435)
2005-05-07 22:57:22 +02:00
6 Unknown .frm version
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
*/
Import changeset
2000-07-31 21:29:14 +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
int open_table_def(THD *thd, TABLE_SHARE *share, uint db_flags)
{
int error, table_type;
bool error_given;
File file;
uchar head[288], *disk_buff;
char path[FN_REFLEN];
Simpler arena swapping code Now thd->mem_root is a pointer to thd->main_mem_root and THR_MALLOC is a pointer to thd->mem_root. This gives us the following benefits: - Allow us to easily detect if arena has already been swapped before (this fixes a bug in setup_conds() where arena was swaped twice in some cases) - Faster swaps of arenas (as we don't have to copy the whole MEM_ROOT) - We don't anymore have to call my_pthread_setspecific_ptr(THR_MALLOC,...) to change where memory is alloced. Now it's enough to set thd->mem_root
2004-11-08 00:13:54 +01:00
MEM_ROOT **root_ptr, *old_root;
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_ENTER("open_table_def");
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", ("table: '%s'.'%s' path: '%s'", share->db.str,
share->table_name.str, share->normalized_path.str));
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
error= 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
error_given= 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
disk_buff= NULL;
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +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
strxmov(path, share->normalized_path.str, reg_ext, NullS);
if ((file= my_open(path, O_RDONLY | O_SHARE, MYF(0))) < 0)
table.cc: Try to open an old 5.0 table name if encoded name doesn't exist. This makes SELECT working without having to rename tables with tricky characters in name into 5.1 format (simpliefies upgrade).
2006-01-24 14:33:39 +01:00
{
Bug#17142 Crash if create with encoded name upgrade.result, upgrade.test: Adding test case. table.cc: Don't try to open a table with old-formatted name, if the table name contains "@" character. This is to avoid mixing two different names to each other. Thanks Monty for suggesting this fix.
2006-03-07 10:05:24 +01:00
if (strchr(share->table_name.str, '@'))
goto err_not_open;
table.cc: Try to open an old 5.0 table name if encoded name doesn't exist. This makes SELECT working without having to rename tables with tricky characters in name into 5.1 format (simpliefies upgrade).
2006-01-24 14:33:39 +01:00
/* Try unecoded 5.0 name */
uint length;
strxnmov(path, sizeof(path)-1,
mysql_data_home, "/", share->db.str, "/",
share->table_name.str, reg_ext, NullS);
length= unpack_filename(path, path) - reg_ext_length;
/*
The following is a safety test and should never fail
as the old file name should never be longer than the new one.
*/
DBUG_ASSERT(length <= share->normalized_path.length);
/*
If the old and the new names have the same length,
then table name does not have tricky characters,
so no need to check the old file name.
*/
if (length == share->normalized_path.length ||
((file= my_open(path, O_RDONLY | O_SHARE, MYF(0))) < 0))
goto err_not_open;
/* Unencoded 5.0 table name found */
path[length]= '\0'; // Remove .frm extension
strmov(share->normalized_path.str, path);
share->normalized_path.length= length;
}
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +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
error= 4;
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 (my_read(file,(byte*) head, 64, MYF(MY_NABP)))
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err;
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +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
if (head[0] == (uchar) 254 && head[1] == 1)
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +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
if (head[2] == FRM_VER || head[2] == FRM_VER+1 ||
(head[2] >= FRM_VER+3 && head[2] <= FRM_VER+4))
table_type= 1;
else
{
error= 6; // Unkown .frm version
goto err;
}
}
Merge mysql.com:/home/my/mysql-5.1 into mysql.com:/home/my/mysql-5.1-TDC
2005-11-23 21:58:53 +01:00
else if (memcmp(head, STRING_WITH_LEN("TYPE=")) == 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= 5;
if (memcmp(head+5,"VIEW",4) == 0)
{
share->is_view= 1;
if (db_flags & OPEN_VIEW)
error= 0;
}
goto err;
}
else
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err;
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +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
/* No handling of text based files yet */
if (table_type == 1)
{
root_ptr= my_pthread_getspecific_ptr(MEM_ROOT**, THR_MALLOC);
old_root= *root_ptr;
*root_ptr= &share->mem_root;
error= open_binary_frm(thd, share, head, file);
*root_ptr= old_root;
if (share->db.length == 5 &&
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
!my_strcasecmp(system_charset_info, share->db.str, "mysql"))
{
/*
We can't mark all tables in 'mysql' database as system since we don't
allow to lock such tables for writing with any other tables (even with
other system tables) and some privilege tables need this.
*/
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
if (!my_strcasecmp(system_charset_info, share->table_name.str, "proc"))
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
share->system_table= 1;
else
{
if (!my_strcasecmp(system_charset_info, share->table_name.str,
"general_log"))
Portability compilation fixes.
2006-01-20 13:23:24 +01:00
share->log_table= QUERY_LOG_GENERAL;
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 (!my_strcasecmp(system_charset_info, share->table_name.str,
"slow_log"))
Portability compilation fixes.
2006-01-20 13:23:24 +01:00
share->log_table= QUERY_LOG_SLOW;
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
}
}
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_given= 1;
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +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
if (!error)
thd->status_var.opened_shares++;
Import changeset
2000-07-31 21:29:14 +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
err:
my_close(file, MYF(MY_WME));
Import changeset
2000-07-31 21:29:14 +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
err_not_open:
if (error && !error_given)
Add special error message for .frm files with an un-understood version, like those from 5.0 with new datatypes. (Bug #10435)
2005-05-07 22:57:22 +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
share->error= error;
open_table_error(share, error, (share->open_errno= my_errno), 0);
Add special error message for .frm files with an un-understood version, like those from 5.0 with new datatypes. (Bug #10435)
2005-05-07 22:57:22 +02:00
}
Bug#17181 (mysqlslap test server crash): Moving assignments to table_map_id for thread-safe handling of table shares.
2006-02-23 13:34:03 +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
DBUG_RETURN(error);
}
/*
Read data from a binary .frm file from MySQL 3.23 - 5.0 into TABLE_SHARE
*/
static int open_binary_frm(THD *thd, TABLE_SHARE *share, uchar *head,
File file)
{
int error, errarg= 0;
uint new_frm_ver, field_pack_length, new_field_pack_flag;
uint interval_count, interval_parts, read_length, int_length;
uint db_create_options, keys, key_parts, n_length;
uint key_info_length, com_length, null_bit_pos;
uint extra_rec_buf_length;
uint i,j;
bool use_hash;
char *keynames, *record, *names, *comment_pos;
uchar *disk_buff, *strpos, *null_flags, *null_pos;
ulong pos, record_offset, *rec_per_key, rec_buff_length;
handler *handler_file= 0;
KEY *keyinfo;
KEY_PART_INFO *key_part;
SQL_CRYPT *crypted=0;
Field **field_ptr, *reg_field;
const char **interval_array;
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 legacy_db_type;
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
my_bitmap_map *bitmaps;
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_ENTER("open_binary_frm");
new_field_pack_flag= head[27];
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
new_frm_ver= (head[2] - FRM_VER);
After merge fixes & remove compiler warnings Added lengths for all MYSQL_FIELD string parameters Changed field length to 2 byte in .frm files
2003-02-07 14:47:24 +01:00
field_pack_length= new_frm_ver < 2 ? 11 : 17;
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
disk_buff= 0;
Import changeset
2000-07-31 21:29:14 +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
error= 3;
Import changeset
2000-07-31 21:29:14 +02:00
if (!(pos=get_form_pos(file,head,(TYPELIB*) 0)))
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err; /* purecov: inspected */
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
share->frm_version= head[2];
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
/*
Check if .frm file created by MySQL 5.0. In this case we want to
display CHAR fields as CHAR and not as VARCHAR.
We do it this way as we want to keep the old frm version to enable
MySQL 4.1 to read these files.
*/
if (share->frm_version == FRM_VER_TRUE_VARCHAR -1 && head[33] == 5)
share->frm_version= FRM_VER_TRUE_VARCHAR;
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
#ifdef WITH_PARTITION_STORAGE_ENGINE
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
share->default_part_db_type=
ha_checktype(thd, (enum legacy_db_type) (uint) *(head+61), 0, 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_PRINT("info", ("default_part_db_type = %u", head[61]));
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
#endif
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
legacy_db_type= (enum legacy_db_type) (uint) *(head+3);
share->db_type= ha_checktype(thd, legacy_db_type, 0, 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
share->db_create_options= db_create_options= uint2korr(head+30);
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
share->db_options_in_use= share->db_create_options;
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
share->mysql_version= uint4korr(head+51);
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
share->null_field_first= 0;
Import changeset
2000-07-31 21:29:14 +02:00
if (!head[32]) // New frm file in 3.23
{
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
share->avg_row_length= uint4korr(head+34);
share-> row_type= (row_type) head[40];
share->table_charset= get_charset((uint) head[38],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
share->null_field_first= 1;
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 (!share->table_charset)
{
/* unknown charset in head[38] or pre-3.23 frm */
Print a warning when an old table (with no character set stored) is opened and the default character set is multi-byte, which will result in character column size changes. (Bug #6913)
2005-01-17 21:22:23 +01:00
if (use_mb(default_charset_info))
{
/* Warn that we may be changing the size of character columns */
sql_print_warning("'%s' had no or invalid character set, "
"and default character set is multi-byte, "
"so character column sizes may have changed",
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
share->path);
Print a warning when an old table (with no character set stored) is opened and the default character set is multi-byte, which will result in character column size changes. (Bug #6913)
2005-01-17 21:22:23 +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
share->table_charset= default_charset_info;
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
share->db_record_offset= 1;
Import changeset
2000-07-31 21:29:14 +02:00
if (db_create_options & HA_OPTION_LONG_BLOB_PTR)
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
share->blob_ptr_size= portable_sizeof_char_ptr;
fixed & added comments (according to Igor's review and a bit more)
2004-10-07 00:45:06 +02:00
/* Set temporarily a good value for db_low_byte_first */
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
share->db_low_byte_first= test(legacy_db_type != DB_TYPE_ISAM);
Import changeset
2000-07-31 21:29:14 +02:00
error=4;
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
share->max_rows= uint4korr(head+18);
share->min_rows= uint4korr(head+22);
Import changeset
2000-07-31 21:29:14 +02:00
/* Read keyinformation */
Update for BTREE keys in HEAP tables Split handler->option_flag() to handler->table_flags() and handler->index_flags()
2002-04-12 20:35:46 +02:00
key_info_length= (uint) uint2korr(head+28);
Import changeset
2000-07-31 21:29:14 +02:00
VOID(my_seek(file,(ulong) uint2korr(head+6),MY_SEEK_SET,MYF(0)));
Update for BTREE keys in HEAP tables Split handler->option_flag() to handler->table_flags() and handler->index_flags()
2002-04-12 20:35:46 +02:00
if (read_string(file,(gptr*) &disk_buff,key_info_length))
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err; /* purecov: inspected */
Bitmap post-review fixes
2003-10-30 19:17:57 +01:00
if (disk_buff[0] & 0x80)
>255 keys support in frm
2003-10-21 00:13:17 +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
share->keys= keys= (disk_buff[1] << 7) | (disk_buff[0] & 0x7f);
share->key_parts= key_parts= uint2korr(disk_buff+2);
>255 keys support in frm
2003-10-21 00:13:17 +02:00
}
else
{
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
share->keys= keys= disk_buff[0];
share->key_parts= key_parts= disk_buff[1];
>255 keys support in frm
2003-10-21 00:13:17 +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
share->keys_for_keyread.init(0);
share->keys_in_use.init(keys);
Import changeset
2000-07-31 21:29:14 +02:00
n_length=keys*sizeof(KEY)+key_parts*sizeof(KEY_PART_INFO);
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 (!(keyinfo = (KEY*) alloc_root(&share->mem_root,
n_length + uint2korr(disk_buff+4))))
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err; /* purecov: inspected */
Import changeset
2000-07-31 21:29:14 +02:00
bzero((char*) keyinfo,n_length);
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
share->key_info= keyinfo;
A lot of portability fixes. Added rename table for BDB. This fixes a bug in ALTER TABLE with BDB tables.
2003-01-05 19:18:49 +01:00
key_part= my_reinterpret_cast(KEY_PART_INFO*) (keyinfo+keys);
Import changeset
2000-07-31 21:29:14 +02:00
strpos=disk_buff+6;
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 (!(rec_per_key= (ulong*) alloc_root(&share->mem_root,
Import changeset
2000-07-31 21:29:14 +02:00
sizeof(ulong*)*key_parts)))
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err;
Import changeset
2000-07-31 21:29:14 +02:00
for (i=0 ; i < keys ; i++, keyinfo++)
{
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
keyinfo->table= 0; // Updated in open_frm
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 (new_frm_ver >= 3)
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
{
keyinfo->flags= (uint) uint2korr(strpos) ^ HA_NOSAME;
keyinfo->key_length= (uint) uint2korr(strpos+2);
keyinfo->key_parts= (uint) strpos[4];
keyinfo->algorithm= (enum ha_key_alg) strpos[5];
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
keyinfo->block_size= uint2korr(strpos+6);
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
strpos+=8;
}
else
{
keyinfo->flags= ((uint) strpos[0]) ^ HA_NOSAME;
keyinfo->key_length= (uint) uint2korr(strpos+1);
keyinfo->key_parts= (uint) strpos[3];
keyinfo->algorithm= HA_KEY_ALG_UNDEF;
strpos+=4;
}
Added GIS extension
2002-02-22 12:24:42 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
keyinfo->key_part= key_part;
keyinfo->rec_per_key= rec_per_key;
for (j=keyinfo->key_parts ; j-- ; key_part++)
{
*rec_per_key++=0;
key_part->fieldnr= (uint16) (uint2korr(strpos) & FIELD_NR_MASK);
key_part->offset= (uint) uint2korr(strpos+2)-1;
key_part->key_type= (uint) uint2korr(strpos+5);
// key_part->field= (Field*) 0; // Will be fixed later
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
if (new_frm_ver >= 1)
Import changeset
2000-07-31 21:29:14 +02:00
{
key_part->key_part_flag= *(strpos+4);
key_part->length= (uint) uint2korr(strpos+7);
strpos+=9;
}
else
{
key_part->length= *(strpos+4);
key_part->key_part_flag=0;
if (key_part->length > 128)
{
key_part->length&=127; /* purecov: inspected */
key_part->key_part_flag=HA_REVERSE_SORT; /* purecov: inspected */
}
strpos+=7;
}
key_part->store_length=key_part->length;
}
}
Update for BTREE keys in HEAP tables Split handler->option_flag() to handler->table_flags() and handler->index_flags()
2002-04-12 20:35:46 +02:00
keynames=(char*) key_part;
strpos+= (strmov(keynames, (char *) strpos) - keynames)+1;
Made keyread (key scanning) a key specific attribute. This avoids using fulltext keys for table scanning. This also reverts Sinisa's original fix for this problem.
2002-06-18 23:22:30 +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
share->reclength = uint2korr((head+16));
Import changeset
2000-07-31 21:29:14 +02:00
if (*(head+26) == 1)
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
share->system= 1; /* one-record-database */
Import changeset
2000-07-31 21:29:14 +02:00
#ifdef HAVE_CRYPTED_FRM
else if (*(head+26) == 2)
{
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
crypted= get_crypt_for_frm();
share->crypted= 1;
Import changeset
2000-07-31 21:29:14 +02:00
}
#endif
Store and read engine type string in extra block of .frm.
2005-10-10 20:01:45 +02:00
record_offset= (ulong) (uint2korr(head+6)+
((uint2korr(head+14) == 0xffff ?
uint4korr(head+47) : uint2korr(head+14))));
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 ((n_length= uint4korr(head+55)))
Store and read engine type string in extra block of .frm.
2005-10-10 20:01:45 +02:00
{
/* Read extra data segment */
char *buff, *next_chunk, *buff_end;
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
DBUG_PRINT("info", ("extra segment size is %u bytes", n_length));
Store and read engine type string in extra block of .frm.
2005-10-10 20:01:45 +02:00
if (!(next_chunk= buff= my_malloc(n_length, MYF(MY_WME))))
goto err;
if (my_pread(file, (byte*)buff, n_length, record_offset + share->reclength,
MYF(MY_NABP)))
{
my_free(buff, MYF(0));
goto err;
}
Always save/restore connect string.
2005-10-10 20:53:53 +02:00
share->connect_string.length= uint2korr(buff);
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->connect_string.str= strmake_root(&share->mem_root,
Always save/restore connect string.
2005-10-10 20:53:53 +02:00
next_chunk + 2, share->connect_string.length)))
Store and read engine type string in extra block of .frm.
2005-10-10 20:01:45 +02:00
{
Always save/restore connect string.
2005-10-10 20:53:53 +02:00
my_free(buff, MYF(0));
goto err;
Store and read engine type string in extra block of .frm.
2005-10-10 20:01:45 +02:00
}
Always save/restore connect string.
2005-10-10 20:53:53 +02:00
next_chunk+= share->connect_string.length + 2;
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
buff_end= buff + n_length;
Store and read engine type string in extra block of .frm.
2005-10-10 20:01:45 +02:00
if (next_chunk + 2 < buff_end)
{
uint str_db_type_length= uint2korr(next_chunk);
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
LEX_STRING name= { next_chunk + 2, str_db_type_length };
handlerton *tmp_db_type= ha_resolve_by_name(thd, &name);
if (tmp_db_type != NULL)
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
{
share->db_type= tmp_db_type;
DBUG_PRINT("info", ("setting dbtype to '%.*s' (%d)",
str_db_type_length, next_chunk + 2,
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_legacy_type(share->db_type)));
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
}
Make storage engines "pluggable", handlerton work
2005-11-07 16:25:06 +01:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
else
{
if (!strncmp(next_chunk + 2, "partition", str_db_type_length))
{
/* Use partition 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
share->db_type= &partition_hton;
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
DBUG_PRINT("info", ("setting dbtype to '%.*s' (%d)",
str_db_type_length, next_chunk + 2,
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_legacy_type(share->db_type)));
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
}
}
#endif
Store and read engine type string in extra block of .frm.
2005-10-10 20:01:45 +02:00
next_chunk+= str_db_type_length + 2;
}
BUG#16068: Plugin related bugs Fixed that --plugin_dir option doesn't work. Fixed that parser name is not restored correctly from the table.
2005-12-29 14:28:59 +01:00
if (next_chunk + 5 < buff_end)
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
{
Mikaels fix for compiling without partition storage engine
2005-11-25 10:33:38 +01:00
uint32 partition_info_len = uint4korr(next_chunk);
#ifdef WITH_PARTITION_STORAGE_ENGINE
if ((share->partition_info_len= partition_info_len))
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +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
if (!(share->partition_info=
(uchar*) memdup_root(&share->mem_root, next_chunk + 4,
Mikaels fix for compiling without partition storage engine
2005-11-25 10:33:38 +01:00
partition_info_len + 1)))
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
{
my_free(buff, MYF(0));
goto err;
}
Mikaels fix for compiling without partition storage engine
2005-11-25 10:33:38 +01:00
}
#else
if (partition_info_len)
{
DBUG_PRINT("info", ("WITH_PARTITION_STORAGE_ENGINE is not defined"));
my_free(buff, MYF(0));
goto err;
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
}
Mikaels fix for compiling without partition storage engine
2005-11-25 10:33:38 +01:00
#endif
BUG#16068: Plugin related bugs Fixed that --plugin_dir option doesn't work. Fixed that parser name is not restored correctly from the table.
2005-12-29 14:28:59 +01:00
next_chunk+= 5 + partition_info_len;
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
}
WL 2826: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
#if MYSQL_VERSION_ID < 50200
if (share->mysql_version >= 50106 && share->mysql_version <= 50109)
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: Error handling of ALTER TABLE for partitioning After review changes
2006-04-16 03:49:13 +02:00
Partition state array was here in version 5.1.6 to 5.1.9, this code
makes it possible to load a 5.1.6 table in later versions. Can most
likely be removed at some point in time. Will only be used for
upgrades within 5.1 series of versions. Upgrade to 5.2 can only be
done from newer 5.1 versions.
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: Fourth step Remove partition state from frm file Ensure we can still load tables created in 5.1.6 for the time being
2006-02-01 17:05:22 +01:00
next_chunk+= 4;
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
}
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
else if (share->mysql_version >= 50110)
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
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
{
/* New auto_partitioned indicator introduced in 5.1.11 */
#ifdef WITH_PARTITION_STORAGE_ENGINE
share->auto_partitioned= *next_chunk;
#endif
next_chunk++;
}
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
keyinfo= share->key_info;
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
for (i= 0; i < keys; i++, keyinfo++)
{
if (keyinfo->flags & HA_USES_PARSER)
{
LEX_STRING parser_name;
if (next_chunk >= buff_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
DBUG_PRINT("error",
("fulltext key uses parser that is not defined in .frm"));
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
my_free(buff, MYF(0));
goto err;
}
parser_name.str= next_chunk;
parser_name.length= strlen(next_chunk);
keyinfo->parser= plugin_lock(&parser_name, MYSQL_FTPARSER_PLUGIN);
if (! keyinfo->parser)
{
my_error(ER_PLUGIN_IS_NOT_LOADED, MYF(0), parser_name.str);
fixed reading of free'd memory when opening a table with non-existing plugin
2006-04-06 15:49:00 +02:00
my_free(buff, MYF(0));
Fix of crashed 5.1 tree (wrong merge + wrong pullout patch forced us to create a new 5.1 tree) This is a merge of 5.0 -> 5.1 + some code from old 5.1 tree to get all tests to work and keep the .frm format the same as the old 5.1 tree.
2005-11-05 12:20:35 +01:00
goto err;
}
}
}
Store and read engine type string in extra block of .frm.
2005-10-10 20:01:45 +02:00
my_free(buff, MYF(0));
}
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
share->key_block_size= uint2korr(head+62);
Import changeset
2000-07-31 21:29:14 +02:00
error=4;
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
extra_rec_buf_length= uint2korr(head+59);
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
rec_buff_length= ALIGN_SIZE(share->reclength + 1 + extra_rec_buf_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
share->rec_buff_length= rec_buff_length;
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 (!(record= (char *) alloc_root(&share->mem_root,
rec_buff_length)))
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err; /* purecov: inspected */
Fixes for windows compilation bugs (After review of cs georg:1.1800 by Monty)
2005-01-26 15:19:20 +01:00
share->default_values= (byte *) record;
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 (my_pread(file,(byte*) record, (uint) share->reclength,
Made changes to add federated CONNECTION information to the .frm file (per Monty's patch). Remove references to the "COMMENT" field. WL#2414
2005-09-13 03:02:17 +02:00
record_offset, MYF(MY_NABP)))
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
goto err; /* purecov: inspected */
Import changeset
2000-07-31 21:29:14 +02:00
VOID(my_seek(file,pos,MY_SEEK_SET,MYF(0)));
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
if (my_read(file,(byte*) head,288,MYF(MY_NABP)))
goto err;
fixes/cleanups according to Coverity report
2005-01-24 15:48:25 +01:00
#ifdef HAVE_CRYPTED_FRM
Import changeset
2000-07-31 21:29:14 +02:00
if (crypted)
{
crypted->decode((char*) head+256,288-256);
if (sint2korr(head+284) != 0) // Should be 0
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err; // Wrong password
Import changeset
2000-07-31 21:29:14 +02:00
}
fixes/cleanups according to Coverity report
2005-01-24 15:48:25 +01:00
#endif
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
share->fields= uint2korr(head+258);
pos= uint2korr(head+260); /* Length of all screens */
n_length= uint2korr(head+268);
interval_count= uint2korr(head+270);
interval_parts= uint2korr(head+272);
int_length= uint2korr(head+274);
share->null_fields= uint2korr(head+282);
com_length= uint2korr(head+284);
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
share->comment.length= (int) (head[46]);
Merge myoffice.izhnet.ru:/usr/home/gluh/MySQL/tmp_merge into myoffice.izhnet.ru:/usr/home/gluh/MySQL/5.1
2006-07-18 15:43:55 +02:00
share->comment.str= strmake_root(&share->mem_root, (char*) head+47,
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
share->comment.length);
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
DBUG_PRINT("info",("i_count: %d i_parts: %d index: %d n_length: %d int_length: %d com_length: %d", interval_count,interval_parts, share->keys,n_length,int_length, com_length));
Import changeset
2000-07-31 21:29:14 +02:00
if (!(field_ptr = (Field **)
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
alloc_root(&share->mem_root,
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
(uint) ((share->fields+1)*sizeof(Field*)+
Import changeset
2000-07-31 21:29:14 +02:00
interval_count*sizeof(TYPELIB)+
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
(share->fields+interval_parts+
Import changeset
2000-07-31 21:29:14 +02:00
keys+3)*sizeof(my_string)+
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
(n_length+int_length+com_length)))))
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err; /* purecov: inspected */
Import changeset
2000-07-31 21:29:14 +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
share->field= field_ptr;
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
read_length=(uint) (share->fields * field_pack_length +
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
pos+ (uint) (n_length+int_length+com_length));
Import changeset
2000-07-31 21:29:14 +02:00
if (read_string(file,(gptr*) &disk_buff,read_length))
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err; /* purecov: inspected */
fixes/cleanups according to Coverity report
2005-01-24 15:48:25 +01:00
#ifdef HAVE_CRYPTED_FRM
Import changeset
2000-07-31 21:29:14 +02:00
if (crypted)
{
crypted->decode((char*) disk_buff,read_length);
delete crypted;
crypted=0;
}
fixes/cleanups according to Coverity report
2005-01-24 15:48:25 +01:00
#endif
Import changeset
2000-07-31 21:29:14 +02:00
strpos= disk_buff+pos;
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
share->intervals= (TYPELIB*) (field_ptr+share->fields+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
interval_array= (const char **) (share->intervals+interval_count);
names= (char*) (interval_array+share->fields+interval_parts+keys+3);
Import changeset
2000-07-31 21:29:14 +02:00
if (!interval_count)
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
share->intervals= 0; // For better debugging
memcpy((char*) names, strpos+(share->fields*field_pack_length),
Import changeset
2000-07-31 21:29:14 +02:00
(uint) (n_length+int_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
comment_pos= names+(n_length+int_length);
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
memcpy(comment_pos, disk_buff+read_length-com_length, com_length);
Import changeset
2000-07-31 21:29:14 +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
fix_type_pointers(&interval_array, &share->fieldnames, 1, &names);
fix_type_pointers(&interval_array, share->intervals, interval_count,
Import changeset
2000-07-31 21:29:14 +02:00
&names);
Bug#6202: ENUMs are not case sensitive even if declared BINARY
2004-10-25 14:51:26 +02:00
{
/* Set ENUM and SET lengths */
TYPELIB *interval;
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 (interval= share->intervals;
interval < share->intervals + interval_count;
Bug#6202: ENUMs are not case sensitive even if declared BINARY
2004-10-25 14:51:26 +02:00
interval++)
{
uint count= (uint) (interval->count + 1) * sizeof(uint);
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 (!(interval->type_lengths= (uint *) alloc_root(&share->mem_root,
Bug#6202: ENUMs are not case sensitive even if declared BINARY
2004-10-25 14:51:26 +02:00
count)))
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err;
Bug#6202: ENUMs are not case sensitive even if declared BINARY
2004-10-25 14:51:26 +02:00
for (count= 0; count < interval->count; count++)
interval->type_lengths[count]= strlen(interval->type_names[count]);
interval->type_lengths[count]= 0;
}
}
Import changeset
2000-07-31 21:29:14 +02:00
if (keynames)
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
fix_type_pointers(&interval_array, &share->keynames, 1, &keynames);
Import changeset
2000-07-31 21:29:14 +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
/* Allocate handler */
if (!(handler_file= get_new_handler(share, thd->mem_root,
share->db_type)))
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
goto err;
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
record= (char*) share->default_values-1; /* Fieldstart = 1 */
if (share->null_field_first)
Import changeset
2000-07-31 21:29:14 +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
null_flags= null_pos= (uchar*) record+1;
bit type
2004-12-17 15:06:05 +01:00
null_bit_pos= (db_create_options & HA_OPTION_PACK_RECORD) ? 0 : 1;
Merge bk-internal.mysql.com:/home/bk/mysql-5.0 into mysql.com:/home/my/mysql-5.0
2005-10-08 02:37:23 +02:00
/*
null_bytes below is only correct under the condition that
there are no bit fields. Correct values is set below after the
table struct is initialized
*/
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
share->null_bytes= (share->null_fields + null_bit_pos + 7) / 8;
Import changeset
2000-07-31 21:29:14 +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
#ifndef WE_WANT_TO_SUPPORT_VERY_OLD_FRM_FILES
Import changeset
2000-07-31 21:29:14 +02:00
else
{
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
share->null_bytes= (share->null_fields+7)/8;
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
null_flags= null_pos= (uchar*) (record + 1 +share->reclength -
share->null_bytes);
bit type
2004-12-17 15:06:05 +01:00
null_bit_pos= 0;
Import changeset
2000-07-31 21:29:14 +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
#endif
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
use_hash= share->fields >= MAX_FIELDS_BEFORE_HASH;
Import changeset
2000-07-31 21:29:14 +02:00
if (use_hash)
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
use_hash= !hash_init(&share->name_hash,
Hash now supports several charsets
2002-03-14 18:44:42 +01:00
system_charset_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
share->fields,0,0,
Some optimization in CHARSET_INFO: We don't need separate hash_sort() and hash_caseup() We don't need tosort(). strnxfrm will do the same.
2003-04-01 09:45:16 +02:00
(hash_get_key) get_field_name,0,0);
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
for (i=0 ; i < share->fields; i++, strpos+=field_pack_length, field_ptr++)
Import changeset
2000-07-31 21:29:14 +02:00
{
After merge fixes & remove compiler warnings Added lengths for all MYSQL_FIELD string parameters Changed field length to 2 byte in .frm files
2003-02-07 14:47:24 +01:00
uint pack_flag, interval_nr, unireg_type, recpos, field_length;
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
enum_field_types field_type;
BINARY charset is now used instead of binary_flag
2002-10-25 10:58:32 +02:00
CHARSET_INFO *charset=NULL;
More OpenGIS compatibility: Point, MultiLine, Poligom, etc, are now not aliases for Geometry, but separate field types.
2003-03-27 10:09:09 +01:00
Field::geometry_type geom_type= Field::GEOM_GEOMETRY;
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
LEX_STRING comment;
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
if (new_frm_ver >= 3)
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
{
/* new frm file in 4.1 */
After merge fixes & remove compiler warnings Added lengths for all MYSQL_FIELD string parameters Changed field length to 2 byte in .frm files
2003-02-07 14:47:24 +01:00
field_length= uint2korr(strpos+3);
recpos= uint3korr(strpos+5);
pack_flag= uint2korr(strpos+8);
unireg_type= (uint) strpos[10];
interval_nr= (uint) strpos[12];
uint comment_length=uint2korr(strpos+15);
field_type=(enum_field_types) (uint) strpos[13];
More OpenGIS compatibility: Point, MultiLine, Poligom, etc, are now not aliases for Geometry, but separate field types.
2003-03-27 10:09:09 +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
/* charset and geometry_type share the same byte in frm */
More OpenGIS compatibility: Point, MultiLine, Poligom, etc, are now not aliases for Geometry, but separate field types.
2003-03-27 10:09:09 +01:00
if (field_type == FIELD_TYPE_GEOMETRY)
{
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
More OpenGIS compatibility: Point, MultiLine, Poligom, etc, are now not aliases for Geometry, but separate field types.
2003-03-27 10:09:09 +01:00
geom_type= (Field::geometry_type) strpos[14];
charset= &my_charset_bin;
SCRUM: WL#1163 (Making spatial code optional) Pack of changes to do in sql/ code.
2004-01-15 18:06:22 +01:00
#else
error= 4; // unsupported field type
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err;
SCRUM: WL#1163 (Making spatial code optional) Pack of changes to do in sql/ code.
2004-01-15 18:06:22 +01:00
#endif
More OpenGIS compatibility: Point, MultiLine, Poligom, etc, are now not aliases for Geometry, but separate field types.
2003-03-27 10:09:09 +01:00
}
else
{
Bugs#7278: Don't open a table if it contains columns with non-supported charsets
2004-12-22 15:56:57 +01:00
if (!strpos[14])
charset= &my_charset_bin;
else if (!(charset=get_charset((uint) strpos[14], MYF(0))))
{
error= 5; // Unknown or unavailable charset
errarg= (int) strpos[14];
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err;
Bugs#7278: Don't open a table if it contains columns with non-supported charsets
2004-12-22 15:56:57 +01:00
}
More OpenGIS compatibility: Point, MultiLine, Poligom, etc, are now not aliases for Geometry, but separate field types.
2003-03-27 10:09:09 +01:00
}
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
if (!comment_length)
{
comment.str= (char*) "";
comment.length=0;
}
else
{
comment.str= (char*) comment_pos;
comment.length= comment_length;
comment_pos+= comment_length;
}
}
else
{
After merge fixes & remove compiler warnings Added lengths for all MYSQL_FIELD string parameters Changed field length to 2 byte in .frm files
2003-02-07 14:47:24 +01:00
field_length= (uint) strpos[3];
recpos= uint2korr(strpos+4),
pack_flag= uint2korr(strpos+6);
More fixes for strict mode: More tests. Better error messages. Fixed bug when checking if we updated all needed columns for INSERT. Give an error if we encounter a wrong float value during parsing. Don't print DEFAULT for columns without a default value in SHOW CREATE/SHOW FIELDS. Fixed UPDATE IGNORE when using STRICT mode.
2004-10-02 21:20:08 +02:00
pack_flag&= ~FIELDFLAG_NO_DEFAULT; // Safety for old files
After merge fixes & remove compiler warnings Added lengths for all MYSQL_FIELD string parameters Changed field length to 2 byte in .frm files
2003-02-07 14:47:24 +01:00
unireg_type= (uint) strpos[8];
interval_nr= (uint) strpos[10];
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
/* old frm file */
field_type= (enum_field_types) f_packtype(pack_flag);
table.cc: Bug #6802 MySQL 4.0's VARCHAR(NN) BINARY is interpreted as VARBINARY(NN) in 4.1
2004-11-25 13:18:47 +01:00
if (f_is_binary(pack_flag))
{
/*
Try to choose the best 4.1 type:
- for 4.0 "CHAR(N) BINARY" or "VARCHAR(N) BINARY"
try to find a binary collation for character set.
- for other types (e.g. BLOB) just use my_charset_bin.
*/
if (!f_is_blob(pack_flag))
{
// 3.23 or 4.0 string
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 (!(charset= get_charset_by_csname(share->table_charset->csname,
table.cc: Bug #6802 MySQL 4.0's VARCHAR(NN) BINARY is interpreted as VARBINARY(NN) in 4.1
2004-11-25 13:18:47 +01:00
MY_CS_BINSORT, MYF(0))))
charset= &my_charset_bin;
}
else
charset= &my_charset_bin;
}
else
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
charset= share->table_charset;
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
bzero((char*) &comment, sizeof(comment));
}
UCS2 support in ENUM and SET, which also fixes: Bug #5174 SHOW CREATE TABLE hangs up if the table contains half-with katakana enum values UCS2 values are stored in HEX encoding in FRM file
2004-12-06 17:45:32 +01:00
if (interval_nr && charset->mbminlen > 1)
{
/* Unescape UCS2 intervals from HEX notation */
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
TYPELIB *interval= share->intervals + interval_nr - 1;
Bug#7302: UCS2 data in ENUM field get truncated when new column is added
2004-12-21 14:12:27 +01:00
unhex_type2(interval);
UCS2 support in ENUM and SET, which also fixes: Bug #5174 SHOW CREATE TABLE hangs up if the table contains half-with katakana enum values UCS2 values are stored in HEX encoding in FRM file
2004-12-06 17:45:32 +01:00
}
Moved detection of wrong decimal fields before decimal field was created to avoid ASSERT() for DECIMAL(1)
2005-05-26 13:00:47 +02:00
#ifndef TO_BE_DELETED_ON_PRODUCTION
if (field_type == FIELD_TYPE_NEWDECIMAL && !share->mysql_version)
{
/*
Fix pack length of old decimal values from 5.0.3 -> 5.0.4
The difference is that in the old version we stored precision
in the .frm table while we now store the display_length
*/
uint decimals= f_decimals(pack_flag);
field_length= my_decimal_precision_to_length(field_length,
decimals,
f_is_dec(pack_flag) == 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
sql_print_error("Found incompatible DECIMAL field '%s' in %s; "
"Please do \"ALTER TABLE '%s' FORCE\" to fix it!",
share->fieldnames.type_names[i], share->table_name.str,
share->table_name.str);
Moved detection of wrong decimal fields before decimal field was created to avoid ASSERT() for DECIMAL(1)
2005-05-26 13:00:47 +02:00
push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_ERROR,
ER_CRASHED_ON_USAGE,
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
"Found incompatible DECIMAL field '%s' in %s; "
"Please do \"ALTER TABLE '%s' FORCE\" to fix it!",
share->fieldnames.type_names[i],
share->table_name.str,
share->table_name.str);
Moved detection of wrong decimal fields before decimal field was created to avoid ASSERT() for DECIMAL(1)
2005-05-26 13:00:47 +02:00
share->crashed= 1; // Marker for CHECK TABLE
}
#endif
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
*field_ptr= reg_field=
make_field(share, record+recpos,
After merge fixes & remove compiler warnings Added lengths for all MYSQL_FIELD string parameters Changed field length to 2 byte in .frm files
2003-02-07 14:47:24 +01:00
(uint32) field_length,
bit type
2004-12-17 15:06:05 +01:00
null_pos, null_bit_pos,
Import changeset
2000-07-31 21:29:14 +02:00
pack_flag,
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
field_type,
BINARY charset is now used instead of binary_flag
2002-10-25 10:58:32 +02:00
charset,
More OpenGIS compatibility: Point, MultiLine, Poligom, etc, are now not aliases for Geometry, but separate field types.
2003-03-27 10:09:09 +01:00
geom_type,
After merge fixes & remove compiler warnings Added lengths for all MYSQL_FIELD string parameters Changed field length to 2 byte in .frm files
2003-02-07 14:47:24 +01:00
(Field::utype) MTYP_TYPENR(unireg_type),
Import changeset
2000-07-31 21:29:14 +02:00
(interval_nr ?
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
share->intervals+interval_nr-1 :
Import changeset
2000-07-31 21:29:14 +02:00
(TYPELIB*) 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
share->fieldnames.type_names[i]);
Merge with 4.0
2002-11-21 14:56:48 +01:00
if (!reg_field) // Not supported field type
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
{
error= 4;
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err; /* purecov: inspected */
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
}
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
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
reg_field->field_index= i;
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
reg_field->comment=comment;
Bit type support for non-MyISAM tables.
2005-04-12 09:27:43 +02:00
if (field_type == FIELD_TYPE_BIT && !f_bit_as_char(pack_flag))
Import changeset
2000-07-31 21:29:14 +02:00
{
bit type
2004-12-17 15:06:05 +01:00
if ((null_bit_pos+= field_length & 7) > 7)
Import changeset
2000-07-31 21:29:14 +02:00
{
bit type
2004-12-17 15:06:05 +01:00
null_pos++;
null_bit_pos-= 8;
Import changeset
2000-07-31 21:29:14 +02:00
}
}
bit type
2004-12-17 15:06:05 +01:00
if (!(reg_field->flags & NOT_NULL_FLAG))
{
if (!(null_bit_pos= (null_bit_pos + 1) & 7))
null_pos++;
}
More fixes for strict mode: More tests. Better error messages. Fixed bug when checking if we updated all needed columns for INSERT. Give an error if we encounter a wrong float value during parsing. Don't print DEFAULT for columns without a default value in SHOW CREATE/SHOW FIELDS. Fixed UPDATE IGNORE when using STRICT mode.
2004-10-02 21:20:08 +02:00
if (f_no_default(pack_flag))
reg_field->flags|= NO_DEFAULT_VALUE_FLAG;
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 (reg_field->unireg_check == Field::NEXT_NUMBER)
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
share->found_next_number_field= field_ptr;
if (share->timestamp_field == reg_field)
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
share->timestamp_field_offset= i;
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 (use_hash)
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
(void) my_hash_insert(&share->name_hash,
(byte*) field_ptr); // never fail
Import changeset
2000-07-31 21:29:14 +02:00
}
*field_ptr=0; // End marker
/* Fix key->name and key_part->field */
if (key_parts)
{
fix for the 2361 bug: ALTER TABLE ... DROP PRIMARY KEY drops a non-primary key "PRIMARY" has been replaced by primary_key_name.
2004-01-14 13:01:55 +01:00
uint primary_key=(uint) (find_type((char*) primary_key_name,
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
&share->keynames, 3) - 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
uint ha_option= handler_file->ha_table_flags();
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
keyinfo= share->key_info;
key_part= keyinfo->key_part;
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
for (uint key=0 ; key < share->keys ; key++,keyinfo++)
Import changeset
2000-07-31 21:29:14 +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
uint usable_parts= 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
keyinfo->name=(char*) share->keynames.type_names[key];
Made keyread (key scanning) a key specific attribute. This avoids using fulltext keys for table scanning. This also reverts Sinisa's original fix for this problem.
2002-06-18 23:22:30 +02:00
/* Fix fulltext keys for old .frm files */
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->key_info[key].flags & HA_FULLTEXT)
share->key_info[key].algorithm= HA_KEY_ALG_FULLTEXT;
Made keyread (key scanning) a key specific attribute. This avoids using fulltext keys for table scanning. This also reverts Sinisa's original fix for this problem.
2002-06-18 23:22:30 +02:00
Reverted wrong bug fix (Bug#11228)
2006-06-30 23:37:20 +02:00
if (primary_key >= MAX_KEY && (keyinfo->flags & HA_NOSAME))
{
/*
If the UNIQUE key doesn't have NULL columns and is not a part key
declare this as a primary key.
*/
primary_key=key;
for (i=0 ; i < keyinfo->key_parts ;i++)
{
uint fieldnr= key_part[i].fieldnr;
if (!fieldnr ||
share->field[fieldnr-1]->null_ptr ||
share->field[fieldnr-1]->key_length() !=
key_part[i].length)
{
primary_key=MAX_KEY; // Can't be used
break;
}
}
}
Import changeset
2000-07-31 21:29:14 +02:00
for (i=0 ; i < keyinfo->key_parts ; key_part++,i++)
{
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
Field *field;
Import changeset
2000-07-31 21:29:14 +02:00
if (new_field_pack_flag <= 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
key_part->fieldnr= (uint16) find_field(share->field,
(uint) key_part->offset,
(uint) key_part->length);
if (!key_part->fieldnr)
{
error= 4; // Wrong file
goto err;
}
field= key_part->field= share->field[key_part->fieldnr-1];
if (field->null_ptr)
{
key_part->null_offset=(uint) ((byte*) field->null_ptr -
share->default_values);
key_part->null_bit= field->null_bit;
key_part->store_length+=HA_KEY_NULL_LENGTH;
keyinfo->flags|=HA_NULL_PART_KEY;
keyinfo->extra_length+= HA_KEY_NULL_LENGTH;
keyinfo->key_length+= HA_KEY_NULL_LENGTH;
}
if (field->type() == FIELD_TYPE_BLOB ||
field->real_type() == MYSQL_TYPE_VARCHAR)
{
if (field->type() == FIELD_TYPE_BLOB)
key_part->key_part_flag|= HA_BLOB_PART;
else
key_part->key_part_flag|= HA_VAR_LENGTH_PART;
keyinfo->extra_length+=HA_KEY_BLOB_LENGTH;
key_part->store_length+=HA_KEY_BLOB_LENGTH;
keyinfo->key_length+= HA_KEY_BLOB_LENGTH;
/*
Mark that there may be many matching values for one key
combination ('a', 'a ', 'a '...)
*/
if (!(field->flags & BINARY_FLAG))
keyinfo->flags|= HA_END_SPACE_KEY;
}
if (field->type() == MYSQL_TYPE_BIT)
key_part->key_part_flag|= HA_BIT_PART;
if (i == 0 && key != primary_key)
field->flags |= (((keyinfo->flags & HA_NOSAME) &&
(keyinfo->key_parts == 1)) ?
UNIQUE_KEY_FLAG : MULTIPLE_KEY_FLAG);
if (i == 0)
field->key_start.set_bit(key);
if (field->key_length() == key_part->length &&
!(field->flags & BLOB_FLAG))
{
if (handler_file->index_flags(key, i, 0) & HA_KEYREAD_ONLY)
{
share->keys_for_keyread.set_bit(key);
field->part_of_key.set_bit(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
field->part_of_key_not_clustered.set_bit(key);
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 (handler_file->index_flags(key, i, 1) & HA_READ_ORDER)
field->part_of_sortkey.set_bit(key);
}
if (!(key_part->key_part_flag & HA_REVERSE_SORT) &&
usable_parts == i)
usable_parts++; // For FILESORT
field->flags|= PART_KEY_FLAG;
if (key == primary_key)
{
field->flags|= PRI_KEY_FLAG;
/*
If this field is part of the primary key and all keys contains
the primary key, then we can use any key to find this column
*/
if (ha_option & HA_PRIMARY_KEY_IN_READ_INDEX)
field->part_of_key= share->keys_in_use;
}
if (field->key_length() != key_part->length)
{
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
#ifndef TO_BE_DELETED_ON_PRODUCTION
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 (field->type() == FIELD_TYPE_NEWDECIMAL)
{
/*
Fix a fatal error in decimal key handling that causes crashes
on Innodb. We fix it by reducing the key length so that
InnoDB never gets a too big key when searching.
This allows the end user to do an ALTER TABLE to fix the
error.
*/
keyinfo->key_length-= (key_part->length - field->key_length());
key_part->store_length-= (uint16)(key_part->length -
field->key_length());
key_part->length= (uint16)field->key_length();
sql_print_error("Found wrong key definition in %s; "
"Please do \"ALTER TABLE '%s' FORCE \" to fix it!",
share->table_name.str,
share->table_name.str);
push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_ERROR,
ER_CRASHED_ON_USAGE,
"Found wrong key definition in %s; "
"Please do \"ALTER TABLE '%s' FORCE\" to fix "
"it!",
share->table_name.str,
share->table_name.str);
share->crashed= 1; // Marker for CHECK TABLE
goto to_be_deleted;
}
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
#endif
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
key_part->key_part_flag|= HA_PART_KEY_SEG;
}
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
to_be_deleted:
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 the field can be NULL, don't optimize away the test
key_part_column = expression from the WHERE clause
as we need to test for NULL = NULL.
*/
if (field->real_maybe_null())
key_part->key_part_flag|= HA_PART_KEY_SEG;
Import changeset
2000-07-31 21:29:14 +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
keyinfo->usable_key_parts= usable_parts; // Filesort
Fix max_key_length and max_unique_length to take into account varchar and null extra lengths This fixes a bug found by bdb.test Fixes a portability issue with bdb tables
2004-12-31 10:56:50 +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
set_if_bigger(share->max_key_length,keyinfo->key_length+
Fix max_key_length and max_unique_length to take into account varchar and null extra lengths This fixes a bug found by bdb.test Fixes a portability issue with bdb tables
2004-12-31 10:56:50 +01:00
keyinfo->key_parts);
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
share->total_key_length+= keyinfo->key_length;
Bug#10400 - Improperly-defined MERGE table crashes with INSERT ... ON DUPLICATE KEY UPDATE After review version. Added a condition for MERGE tables. These do not have unique indexes. But every key could be a unique key on the underlying MyISAM table. So get the maximum key length for MERGE tables instead of the maximum unique key length. This is used for buffer allocation in write_record().
2005-05-18 19:40:39 +02:00
/*
MERGE tables do not have unique indexes. But every key could be
an unique index on the underlying MyISAM table. (Bug #10400)
*/
if ((keyinfo->flags & HA_NOSAME) ||
(ha_option & HA_ANY_INDEX_MAY_BE_UNIQUE))
Resolve merge from 4.1
2005-05-18 22:54:36 +02:00
set_if_bigger(share->max_unique_length,keyinfo->key_length);
Import changeset
2000-07-31 21:29:14 +02:00
}
bugs #701 and #783 - old .frm files were read in incorrectly in 4.1 (each field was assigned a charset, BLOBs were converted to TEXT)
2003-07-28 16:58:52 +02:00
if (primary_key < MAX_KEY &&
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
(share->keys_in_use.is_set(primary_key)))
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
share->primary_key= primary_key;
Automatic primary key for BDB tables
2000-10-16 01:29:48 +02:00
/*
If we are using an integer as the primary key then allow the user to
refer to it as '_rowid'
*/
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->key_info[primary_key].key_parts == 1)
Import changeset
2000-07-31 21:29:14 +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
Field *field= share->key_info[primary_key].key_part[0].field;
Import changeset
2000-07-31 21:29:14 +02:00
if (field && field->result_type() == INT_RESULT)
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
{
/* note that fieldnr here (and rowid_field_offset) starts from 1 */
share->rowid_field_offset= (share->key_info[primary_key].key_part[0].
fieldnr);
}
Import changeset
2000-07-31 21:29:14 +02:00
}
}
else
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
share->primary_key = MAX_KEY; // we do not have a primary key
Import changeset
2000-07-31 21:29:14 +02:00
}
Give an error if you use a BLOB(X) where X > 255 Fixes for MIT pthreads
2000-10-22 00:19:05 +02:00
else
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
share->primary_key= MAX_KEY;
Import changeset
2000-07-31 21:29:14 +02:00
x_free((gptr) disk_buff);
Automatic primary key for BDB tables
2000-10-16 01:29:48 +02:00
disk_buff=0;
Import changeset
2000-07-31 21:29:14 +02:00
if (new_field_pack_flag <= 1)
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 file format with default as not null */
uint null_length= (share->null_fields+7)/8;
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
bfill(share->default_values + (null_flags - (uchar*) record),
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
null_length, 255);
Import changeset
2000-07-31 21:29:14 +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
if (share->found_next_number_field)
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
{
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
/*
We must have a table object for find_ref_key to calculate field offset
*/
TABLE tmp_table;
tmp_table.record[0]= share->default_values;
reg_field= *share->found_next_number_field;
reg_field->table= &tmp_table;
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 ((int) (share->next_number_index= (uint)
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
find_ref_key(share->key_info, share->keys, reg_field,
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
&share->next_number_key_offset)) < 0)
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
{
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
reg_field->unireg_check= Field::NONE; /* purecov: inspected */
share->found_next_number_field= 0;
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
}
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
reg_field->flags |= AUTO_INCREMENT_FLAG;
reg_field->table= 0;
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
}
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 (share->blob_fields)
Import changeset
2000-07-31 21:29:14 +02:00
{
Field **ptr;
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
uint i, *save;
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
/* Store offsets to blob fields to find them fast */
if (!(share->blob_field= save=
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
(uint*) alloc_root(&share->mem_root,
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
(uint) (share->blob_fields* sizeof(uint)))))
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err;
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
for (i=0, ptr= share->field ; *ptr ; ptr++, i++)
Import changeset
2000-07-31 21:29:14 +02:00
{
if ((*ptr)->flags & BLOB_FLAG)
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
(*save++)= i;
Import changeset
2000-07-31 21:29:14 +02:00
}
}
Merge bk-internal.mysql.com:/home/bk/mysql-5.0 into mysql.com:/home/my/mysql-5.0
2005-10-08 02:37:23 +02:00
/*
the correct null_bytes can now be set, since bitfields have been taken
into account
*/
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
share->null_bytes= (null_pos - (uchar*) null_flags +
Merge bk-internal.mysql.com:/home/bk/mysql-5.0 into mysql.com:/home/my/mysql-5.0
2005-10-08 02:37:23 +02:00
(null_bit_pos + 7) / 8);
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
share->last_null_bit_pos= null_bit_pos;
Corrected last fix for Bug #13152 bit fields and fields that follow become corrupted when dumped from NDB tables
2005-09-19 15:52:57 +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
share->db_low_byte_first= handler_file->low_byte_first();
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
share->column_bitmap_size= bitmap_buffer_size(share->fields);
if (!(bitmaps= (my_bitmap_map*) alloc_root(&share->mem_root,
share->column_bitmap_size)))
goto err;
bitmap_init(&share->all_set, bitmaps, share->fields, FALSE);
bitmap_set_all(&share->all_set);
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
delete handler_file;
#ifndef DBUG_OFF
if (use_hash)
(void) hash_check(&share->name_hash);
#endif
DBUG_RETURN (0);
err:
share->error= error;
share->open_errno= my_errno;
share->errarg= errarg;
x_free((gptr) disk_buff);
delete crypted;
delete handler_file;
hash_free(&share->name_hash);
open_table_error(share, error, share->open_errno, errarg);
DBUG_RETURN(error);
} /* open_binary_frm */
/*
Open a table based on a TABLE_SHARE
SYNOPSIS
open_table_from_share()
thd Thread handler
share Table definition
alias Alias for table
db_stat open flags (for example HA_OPEN_KEYFILE|
HA_OPEN_RNDFILE..) can be 0 (example in
ha_example_table)
prgflag READ_ALL etc..
ha_open_flags HA_OPEN_ABORT_IF_LOCKED etc..
outparam result table
RETURN VALUES
0 ok
1 Error (see open_table_error)
2 Error (see open_table_error)
3 Wrong data in .frm file
4 Error (see open_table_error)
5 Error (see open_table_error: charset unavailable)
7 Table definition has changed in engine
*/
int open_table_from_share(THD *thd, TABLE_SHARE *share, const char *alias,
uint db_stat, uint prgflag, uint ha_open_flags,
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
TABLE *outparam, bool is_create_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
{
int error;
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
uint records, i, bitmap_size;
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_reported= FALSE;
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
byte *record, *bitmaps;
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
Field **field_ptr;
DBUG_ENTER("open_table_from_share");
DBUG_PRINT("enter",("name: '%s.%s' form: 0x%lx", share->db.str,
share->table_name.str, outparam));
error= 1;
bzero((char*) outparam, sizeof(*outparam));
outparam->in_use= thd;
outparam->s= share;
outparam->db_stat= db_stat;
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
outparam->write_row_record= 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
init_sql_alloc(&outparam->mem_root, TABLE_ALLOC_BLOCK_SIZE, 0);
if (!(outparam->alias= my_strdup(alias, MYF(MY_WME))))
goto err;
outparam->quick_keys.init();
outparam->used_keys.init();
outparam->keys_in_use_for_query.init();
/* Allocate handler */
if (!(outparam->file= get_new_handler(share, &outparam->mem_root,
share->db_type)))
goto err;
error= 4;
outparam->reginfo.lock_type= TL_UNLOCK;
outparam->current_lock= F_UNLCK;
records=0;
if ((db_stat & HA_OPEN_KEYFILE) || (prgflag & DELAYED_OPEN))
records=1;
if (prgflag & (READ_ALL+EXTRA_RECORD))
records++;
if (!(record= (byte*) alloc_root(&outparam->mem_root,
share->rec_buff_length * records)))
goto err; /* purecov: inspected */
if (records == 0)
{
/* We are probably in hard repair, and the buffers should not be used */
outparam->record[0]= outparam->record[1]= share->default_values;
}
else
{
outparam->record[0]= record;
if (records > 1)
outparam->record[1]= record+ share->rec_buff_length;
else
outparam->record[1]= outparam->record[0]; // Safety
}
#ifdef HAVE_purify
/*
We need this because when we read var-length rows, we are not updating
bytes after end of varchar
*/
if (records > 1)
{
memcpy(outparam->record[0], share->default_values, share->rec_buff_length);
if (records > 2)
memcpy(outparam->record[1], share->default_values,
share->rec_buff_length);
}
#endif
if (!(field_ptr = (Field **) alloc_root(&outparam->mem_root,
(uint) ((share->fields+1)*
sizeof(Field*)))))
goto err; /* purecov: inspected */
outparam->field= field_ptr;
Fix byte * vs. char * mismatches that break Win32 compilation.
2005-11-29 09:07:21 +01:00
record= (byte*) outparam->record[0]-1; /* Fieldstart = 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
if (share->null_field_first)
outparam->null_flags= (uchar*) record+1;
else
outparam->null_flags= (uchar*) (record+ 1+ share->reclength -
share->null_bytes);
/* Setup copy of fields from share, but use the right alias and record */
for (i=0 ; i < share->fields; i++, field_ptr++)
{
if (!((*field_ptr)= share->field[i]->clone(&outparam->mem_root, outparam)))
goto err;
}
(*field_ptr)= 0; // End marker
if (share->found_next_number_field)
outparam->found_next_number_field=
outparam->field[(uint) (share->found_next_number_field - share->field)];
if (share->timestamp_field)
outparam->timestamp_field= (Field_timestamp*) outparam->field[share->timestamp_field_offset];
/* Fix key->name and key_part->field */
if (share->key_parts)
{
KEY *key_info, *key_info_end;
KEY_PART_INFO *key_part;
uint n_length;
n_length= share->keys*sizeof(KEY) + share->key_parts*sizeof(KEY_PART_INFO);
if (!(key_info= (KEY*) alloc_root(&outparam->mem_root, n_length)))
goto err;
outparam->key_info= key_info;
key_part= (my_reinterpret_cast(KEY_PART_INFO*) (key_info+share->keys));
memcpy(key_info, share->key_info, sizeof(*key_info)*share->keys);
memcpy(key_part, share->key_info[0].key_part, (sizeof(*key_part) *
share->key_parts));
for (key_info_end= key_info + share->keys ;
key_info < key_info_end ;
key_info++)
{
KEY_PART_INFO *key_part_end;
key_info->table= outparam;
key_info->key_part= key_part;
for (key_part_end= key_part+ key_info->key_parts ;
key_part < key_part_end ;
key_part++)
{
Field *field= key_part->field= outparam->field[key_part->fieldnr-1];
if (field->key_length() != key_part->length &&
!(field->flags & BLOB_FLAG))
{
/*
We are using only a prefix of the column as a key:
Create a new field for the key part that matches the index
*/
field= key_part->field=field->new_field(&outparam->mem_root,
After merge fix.
2006-07-06 20:12:33 +02:00
outparam, 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
field->field_length= key_part->length;
}
}
}
}
#ifdef WITH_PARTITION_STORAGE_ENGINE
if (share->partition_info_len)
{
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
MEM_ROOT **root_ptr, *old_root;
bool tmp;
root_ptr= my_pthread_getspecific_ptr(MEM_ROOT**, THR_MALLOC);
old_root= *root_ptr;
*root_ptr= &outparam->mem_root;
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
tmp= mysql_unpack_partition(thd, share->partition_info,
share->partition_info_len,
(uchar*)share->part_state,
share->part_state_len,
outparam, is_create_table,
share->default_part_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
outparam->part_info->is_auto_partitioned= share->auto_partitioned;
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
DBUG_PRINT("info", ("autopartitioned: %u", share->auto_partitioned));
if (!tmp)
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/bug16002
2006-06-14 15:12:07 +02:00
tmp= fix_partition_func(thd, outparam, is_create_table);
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
*root_ptr= old_root;
if (tmp)
BUG#16000: .par file not removed plus errors to error log for normal errors
2006-06-20 19:24:30 +02:00
{
if (is_create_table)
{
/*
During CREATE/ALTER TABLE it is ok to receive errors here.
It is not ok if it happens during the opening of an frm
file as part of a normal query.
*/
error_reported= 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
goto err;
BUG#16000: .par file not removed plus errors to error log for normal errors
2006-06-20 19:24:30 +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
}
#endif
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
/* Allocate bitmaps */
bitmap_size= share->column_bitmap_size;
if (!(bitmaps= (byte*) alloc_root(&outparam->mem_root, bitmap_size*3)))
goto err;
bitmap_init(&outparam->def_read_set,
(my_bitmap_map*) bitmaps, share->fields, FALSE);
bitmap_init(&outparam->def_write_set,
(my_bitmap_map*) (bitmaps+bitmap_size), share->fields, FALSE);
bitmap_init(&outparam->tmp_set,
(my_bitmap_map*) (bitmaps+bitmap_size*2), share->fields, FALSE);
outparam->default_column_bitmaps();
fixed & added comments (according to Igor's review and a bit more)
2004-10-07 00:45:06 +02:00
/* The table struct is now initialized; Open the 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
error= 2;
Automatic primary key for BDB tables
2000-10-16 01:29:48 +02:00
if (db_stat)
{
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
int ha_err;
if ((ha_err= (outparam->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
ha_open(outparam, share->normalized_path.str,
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
(db_stat & HA_READ_ONLY ? O_RDONLY : O_RDWR),
(db_stat & HA_OPEN_TEMPORARY ? HA_OPEN_TMP_TABLE :
((db_stat & HA_WAIT_IF_LOCKED) ||
(specialflag & SPECIAL_WAIT_IF_LOCKED)) ?
HA_OPEN_WAIT_IF_LOCKED :
(db_stat & (HA_ABORT_IF_LOCKED | HA_GET_INFO)) ?
HA_OPEN_ABORT_IF_LOCKED :
HA_OPEN_IGNORE_IF_LOCKED) | ha_open_flags))))
Automatic primary key for BDB tables
2000-10-16 01:29:48 +02:00
{
/* Set a flag if the table is crashed and it can be auto. repaired */
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
share->crashed= ((ha_err == HA_ERR_CRASHED_ON_USAGE) &&
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
outparam->file->auto_repair() &&
!(ha_open_flags & HA_OPEN_FOR_REPAIR));
WL1424 Multiple MySQL Servers: SHOW TABLES etc. should detect new and delete old tables.
2004-09-13 14:46:38 +02:00
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
if (ha_err == HA_ERR_NO_SUCH_TABLE)
WL1424 Multiple MySQL Servers: SHOW TABLES etc. should detect new and delete old tables.
2004-09-13 14:46:38 +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
/*
The table did not exists in storage engine, use same error message
as if the .frm file didn't exist
*/
WL1424 Multiple MySQL Servers: SHOW TABLES etc. should detect new and delete old tables.
2004-09-13 14:46:38 +02:00
error= 1;
my_errno= ENOENT;
}
WL#1895 - Print message to error log in case of detected MyISAM corruption Changed my_error() to print error messages, which come from arbitrary registered ranges of error messages. Messages can be unregistered (and should be at end of the program). Added registration of handler error messages. Added a new mi_print_error() macro and a new mi_report_error() function, which supply error messages with a table name. Added calls to mi_print_error() or mi_report_error() at all places in MyISAM, where table corruption is detected.
2004-12-23 20:11:38 +01:00
else
{
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
outparam->file->print_error(ha_err, MYF(0));
WL#1895 - Print message to error log in case of detected MyISAM corruption Changed my_error() to print error messages, which come from arbitrary registered ranges of error messages. Messages can be unregistered (and should be at end of the program). Added registration of handler error messages. Added a new mi_print_error() macro and a new mi_report_error() function, which supply error messages with a table name. Added calls to mi_print_error() or mi_report_error() at all places in MyISAM, where table corruption is detected.
2004-12-23 20:11:38 +01:00
error_reported= 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 (ha_err == HA_ERR_TABLE_DEF_CHANGED)
error= 7;
WL#1895 - Print message to error log in case of detected MyISAM corruption Changed my_error() to print error messages, which come from arbitrary registered ranges of error messages. Messages can be unregistered (and should be at end of the program). Added registration of handler error messages. Added a new mi_print_error() macro and a new mi_report_error() function, which supply error messages with a table name. Added calls to mi_print_error() or mi_report_error() at all places in MyISAM, where table corruption is detected.
2004-12-23 20:11:38 +01:00
}
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
goto err; /* purecov: inspected */
Automatic primary key for BDB tables
2000-10-16 01:29:48 +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 defined(HAVE_purify) && !defined(DBUG_OFF)
bzero((char*) bitmaps, bitmap_size*3);
#endif
Removed wrong warnings in test suite (This was because select_insert / select_create results was not freed. Added thd to openfrm() for initialization of TABLE->in_use. This fixed a bug in BDB handling where table->in_use was used early
2004-09-17 02:08:23 +02:00
thd->status_var.opened_tables++;
WL#1012: All changes as one single changeset. This includes both code and test cases.
2005-12-22 06:39:02 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN (0);
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
err:
WL#1895 - Print message to error log in case of detected MyISAM corruption Changed my_error() to print error messages, which come from arbitrary registered ranges of error messages. Messages can be unregistered (and should be at end of the program). Added registration of handler error messages. Added a new mi_print_error() macro and a new mi_report_error() function, which supply error messages with a table name. Added calls to mi_print_error() or mi_report_error() at all places in MyISAM, where table corruption is detected.
2004-12-23 20:11:38 +01:00
if (! error_reported)
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
open_table_error(share, error, my_errno, 0);
Automatic repair of MyISAM tables + portability fixes
2000-10-10 23:06:37 +02:00
delete outparam->file;
Make storage engines "pluggable", handlerton work
2005-11-07 16:25:06 +01:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
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 (outparam->part_info)
free_items(outparam->part_info->item_free_list);
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
#endif
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
outparam->file= 0; // For easier error checking
Fixed problem with auto-repair of MyISAM tables Fixed bug in ISAM and MyISAM when updating from multiple-processes
2000-11-21 02:43:34 +01:00
outparam->db_stat=0;
Fix accesses to uninitialized memory (found by valgrind)
2005-01-13 02:02:49 +01:00
free_root(&outparam->mem_root, MYF(0)); // Safe to call on bzero'd root
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_free((char*) outparam->alias, MYF(MY_ALLOW_ZERO_PTR));
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN (error);
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
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
/*
Free information allocated by openfrm
SYNOPSIS
closefrm()
table TABLE object to free
free_share Is 1 if we also want to free table_share
*/
Import changeset
2000-07-31 21:29:14 +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
int closefrm(register TABLE *table, bool free_share)
Import changeset
2000-07-31 21:29:14 +02:00
{
int error=0;
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
uint idx;
KEY *key_info;
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_ENTER("closefrm");
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
DBUG_PRINT("enter", ("table: 0x%lx", (long) 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
Import changeset
2000-07-31 21:29:14 +02:00
if (table->db_stat)
error=table->file->close();
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= table->key_info;
for (idx= table->s->keys; idx; idx--, key_info++)
{
if (key_info->flags & HA_USES_PARSER)
monty review: fixes after reapplying plugin patch from "crashed" 5.1 tree
2005-11-06 14:26:37 +01:00
{
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
plugin_unlock(key_info->parser);
monty review: fixes after reapplying plugin patch from "crashed" 5.1 tree
2005-11-06 14:26:37 +01:00
key_info->flags= 0;
}
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
}
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_free((char*) table->alias, MYF(MY_ALLOW_ZERO_PTR));
table->alias= 0;
if (table->field)
Import changeset
2000-07-31 21:29:14 +02:00
{
for (Field **ptr=table->field ; *ptr ; ptr++)
delete *ptr;
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->field= 0;
Import changeset
2000-07-31 21:29:14 +02:00
}
delete table->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
table->file= 0; /* For easier errorchecking */
Make storage engines "pluggable", handlerton work
2005-11-07 16:25:06 +01:00
#ifdef WITH_PARTITION_STORAGE_ENGINE
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->part_info)
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +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
free_items(table->part_info->item_free_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
table->part_info->item_free_list= 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->part_info= 0;
Patch for push of wl1354 Partitioning
2005-07-18 13:31:02 +02:00
}
#endif
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 (free_share)
{
if (table->s->tmp_table == NO_TMP_TABLE)
release_table_share(table->s, RELEASE_NORMAL);
else
free_table_share(table->s);
}
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
free_root(&table->mem_root, MYF(0));
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(error);
}
/* Deallocate temporary blob storage */
void free_blobs(register TABLE *table)
{
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
uint *ptr, *end;
for (ptr= table->s->blob_field, end=ptr + table->s->blob_fields ;
ptr != end ;
ptr++)
((Field_blob*) table->field[*ptr])->free();
Import changeset
2000-07-31 21:29:14 +02:00
}
/* Find where a form starts */
/* if formname is NullS then only formnames is read */
ulong get_form_pos(File file, uchar *head, TYPELIB *save_names)
{
uint a_length,names,length;
uchar *pos,*buf;
ulong ret_value=0;
DBUG_ENTER("get_form_pos");
names=uint2korr(head+8);
a_length=(names+2)*sizeof(my_string); /* Room for two extra */
if (!save_names)
a_length=0;
else
save_names->type_names=0; /* Clear if error */
if (names)
{
length=uint2korr(head+4);
VOID(my_seek(file,64L,MY_SEEK_SET,MYF(0)));
if (!(buf= (uchar*) my_malloc((uint) length+a_length+names*4,
MYF(MY_WME))) ||
my_read(file,(byte*) buf+a_length,(uint) (length+names*4),
MYF(MY_NABP)))
{ /* purecov: inspected */
x_free((gptr) buf); /* purecov: inspected */
DBUG_RETURN(0L); /* purecov: inspected */
}
pos= buf+a_length+length;
ret_value=uint4korr(pos);
}
if (! save_names)
many warnings (practically safe but annoying) corrected
2006-01-03 17:54:54 +01:00
{
if (names)
my_free((gptr) buf,MYF(0));
}
Import changeset
2000-07-31 21:29:14 +02:00
else if (!names)
bzero((char*) save_names,sizeof(save_names));
else
{
char *str;
str=(char *) (buf+a_length);
fix_type_pointers((const char ***) &buf,save_names,1,&str);
}
DBUG_RETURN(ret_value);
}
/* Read string from a file with malloc */
int read_string(File file, gptr *to, uint length)
{
DBUG_ENTER("read_string");
x_free((gptr) *to);
if (!(*to= (gptr) my_malloc(length+1,MYF(MY_WME))) ||
my_read(file,(byte*) *to,length,MYF(MY_NABP)))
{
x_free((gptr) *to); /* purecov: inspected */
*to= 0; /* purecov: inspected */
DBUG_RETURN(1); /* purecov: inspected */
}
*((char*) *to+length)= '\0';
DBUG_RETURN (0);
} /* read_string */
/* Add a new form to a form file */
ulong make_new_entry(File file, uchar *fileinfo, TYPELIB *formnames,
const char *newname)
{
uint i,bufflength,maxlength,n_length,length,names;
ulong endpos,newpos;
char buff[IO_SIZE];
uchar *pos;
DBUG_ENTER("make_new_entry");
FLUSH TABLE table_list Fixes for RENAME TABLE Portability fixes
2000-08-21 23:18:32 +02:00
length=(uint) strlen(newname)+1;
Import changeset
2000-07-31 21:29:14 +02:00
n_length=uint2korr(fileinfo+4);
maxlength=uint2korr(fileinfo+6);
names=uint2korr(fileinfo+8);
newpos=uint4korr(fileinfo+10);
if (64+length+n_length+(names+1)*4 > maxlength)
{ /* Expand file */
newpos+=IO_SIZE;
int4store(fileinfo+10,newpos);
FLUSH TABLE table_list Fixes for RENAME TABLE Portability fixes
2000-08-21 23:18:32 +02:00
endpos=(ulong) my_seek(file,0L,MY_SEEK_END,MYF(0));/* Copy from file-end */
Import changeset
2000-07-31 21:29:14 +02:00
bufflength= (uint) (endpos & (IO_SIZE-1)); /* IO_SIZE is a power of 2 */
while (endpos > maxlength)
{
VOID(my_seek(file,(ulong) (endpos-bufflength),MY_SEEK_SET,MYF(0)));
if (my_read(file,(byte*) buff,bufflength,MYF(MY_NABP+MY_WME)))
DBUG_RETURN(0L);
VOID(my_seek(file,(ulong) (endpos-bufflength+IO_SIZE),MY_SEEK_SET,
MYF(0)));
if ((my_write(file,(byte*) buff,bufflength,MYF(MY_NABP+MY_WME))))
DBUG_RETURN(0);
endpos-=bufflength; bufflength=IO_SIZE;
}
bzero(buff,IO_SIZE); /* Null new block */
VOID(my_seek(file,(ulong) maxlength,MY_SEEK_SET,MYF(0)));
if (my_write(file,(byte*) buff,bufflength,MYF(MY_NABP+MY_WME)))
DBUG_RETURN(0L);
maxlength+=IO_SIZE; /* Fix old ref */
int2store(fileinfo+6,maxlength);
for (i=names, pos= (uchar*) *formnames->type_names+n_length-1; i-- ;
pos+=4)
{
endpos=uint4korr(pos)+IO_SIZE;
int4store(pos,endpos);
}
}
if (n_length == 1 )
{ /* First name */
length++;
VOID(strxmov(buff,"/",newname,"/",NullS));
}
else
VOID(strxmov(buff,newname,"/",NullS)); /* purecov: inspected */
VOID(my_seek(file,63L+(ulong) n_length,MY_SEEK_SET,MYF(0)));
if (my_write(file,(byte*) buff,(uint) length+1,MYF(MY_NABP+MY_WME)) ||
(names && my_write(file,(byte*) (*formnames->type_names+n_length-1),
names*4, MYF(MY_NABP+MY_WME))) ||
my_write(file,(byte*) fileinfo+10,(uint) 4,MYF(MY_NABP+MY_WME)))
DBUG_RETURN(0L); /* purecov: inspected */
int2store(fileinfo+8,names+1);
int2store(fileinfo+4,n_length+length);
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
VOID(my_chsize(file, newpos, 0, MYF(MY_WME)));/* Append file with '\0' */
Import changeset
2000-07-31 21:29:14 +02:00
DBUG_RETURN(newpos);
} /* make_new_entry */
/* error message when opening a form 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
void open_table_error(TABLE_SHARE *share, int error, int db_errno, int errarg)
Import changeset
2000-07-31 21:29:14 +02:00
{
int err_no;
char buff[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
myf errortype= ME_ERROR+ME_WAITTANG;
DBUG_ENTER("open_table_error");
Import changeset
2000-07-31 21:29:14 +02:00
switch (error) {
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
case 7:
Import changeset
2000-07-31 21:29:14 +02:00
case 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
if (db_errno == ENOENT)
my_error(ER_NO_SUCH_TABLE, MYF(0), share->db.str, share->table_name.str);
else
Import changeset
2000-07-31 21:29:14 +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
strxmov(buff, share->normalized_path.str, reg_ext, NullS);
Merge neptunus.(none):/home/msvensson/mysql/bug15020/my50-bug15020 into neptunus.(none):/home/msvensson/mysql/bug15020/my51-bug15020
2006-01-19 11:43:19 +01:00
my_error((db_errno == EMFILE) ? ER_CANT_OPEN_FILE : ER_FILE_NOT_FOUND,
errortype, buff, db_errno);
Import changeset
2000-07-31 21:29:14 +02:00
}
break;
case 2:
{
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= 0;
const char *datext= "";
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 (share->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 ((file= get_new_handler(share, current_thd->mem_root,
share->db_type)))
{
if (!(datext= *file->bas_ext()))
datext= "";
}
}
err_no= (db_errno == ENOENT) ? ER_FILE_NOT_FOUND : (db_errno == EAGAIN) ?
Import changeset
2000-07-31 21:29:14 +02:00
ER_FILE_USED : ER_CANT_OPEN_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
strxmov(buff, share->normalized_path.str, datext, NullS);
my_error(err_no,errortype, buff, db_errno);
delete file;
Import changeset
2000-07-31 21:29:14 +02:00
break;
}
Bugs#7278: Don't open a table if it contains columns with non-supported charsets
2004-12-22 15:56:57 +01:00
case 5:
{
const char *csname= get_charset_name((uint) errarg);
char tmp[10];
if (!csname || csname[0] =='?')
{
my_snprintf(tmp, sizeof(tmp), "#%d", errarg);
csname= tmp;
}
my_printf_error(ER_UNKNOWN_COLLATION,
"Unknown collation '%s' in table '%-.64s' definition",
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
MYF(0), csname, share->table_name.str);
Import changeset
2000-07-31 21:29:14 +02:00
break;
}
Add special error message for .frm files with an un-understood version, like those from 5.0 with new datatypes. (Bug #10435)
2005-05-07 22:57:22 +02:00
case 6:
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(buff, share->normalized_path.str, reg_ext, NullS);
Add special error message for .frm files with an un-understood version, like those from 5.0 with new datatypes. (Bug #10435)
2005-05-07 22:57:22 +02:00
my_printf_error(ER_NOT_FORM_FILE,
"Table '%-.64s' was created with a different version "
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
"of MySQL and cannot be read",
MYF(0), buff);
Add special error message for .frm files with an un-understood version, like those from 5.0 with new datatypes. (Bug #10435)
2005-05-07 22:57:22 +02:00
break;
Import changeset
2000-07-31 21:29:14 +02:00
default: /* Better wrong error than none */
case 4:
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(buff, share->normalized_path.str, reg_ext, NullS);
my_error(ER_NOT_FORM_FILE, errortype, buff, 0);
Import changeset
2000-07-31 21:29:14 +02:00
break;
}
DBUG_VOID_RETURN;
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
} /* open_table_error */
Import changeset
2000-07-31 21:29:14 +02:00
/*
** fix a str_type to a array type
fixed & added comments (according to Igor's review and a bit more)
2004-10-07 00:45:06 +02:00
** typeparts separated with some char. differents types are separated
Import changeset
2000-07-31 21:29:14 +02:00
** with a '\0'
*/
static void
fix_type_pointers(const char ***array, TYPELIB *point_to_type, uint types,
char **names)
{
char *type_name, *ptr;
char chr;
ptr= *names;
while (types--)
{
point_to_type->name=0;
point_to_type->type_names= *array;
if ((chr= *ptr)) /* Test if empty type */
{
while ((type_name=strchr(ptr+1,chr)) != NullS)
{
*((*array)++) = ptr+1;
*type_name= '\0'; /* End string */
ptr=type_name;
}
Portability fix Cleanup typos (like SKIPP -> SKIP)
2001-11-06 23:13:29 +01:00
ptr+=2; /* Skip end mark and last 0 */
Import changeset
2000-07-31 21:29:14 +02:00
}
else
ptr++;
point_to_type->count= (uint) (*array - point_to_type->type_names);
point_to_type++;
*((*array)++)= NullS; /* End of type */
}
*names=ptr; /* Update end */
return;
} /* fix_type_pointers */
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
TYPELIB *typelib(MEM_ROOT *mem_root, List<String> &strings)
Import changeset
2000-07-31 21:29:14 +02:00
{
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
TYPELIB *result= (TYPELIB*) alloc_root(mem_root, sizeof(TYPELIB));
Import changeset
2000-07-31 21:29:14 +02:00
if (!result)
return 0;
result->count=strings.elements;
result->name="";
Bug#6202: ENUMs are not case sensitive even if declared BINARY
2004-10-25 14:51:26 +02:00
uint nbytes= (sizeof(char*) + sizeof(uint)) * (result->count + 1);
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
if (!(result->type_names= (const char**) alloc_root(mem_root, nbytes)))
Import changeset
2000-07-31 21:29:14 +02:00
return 0;
Bug#6202: ENUMs are not case sensitive even if declared BINARY
2004-10-25 14:51:26 +02:00
result->type_lengths= (uint*) (result->type_names + result->count + 1);
Import changeset
2000-07-31 21:29:14 +02:00
List_iterator<String> it(strings);
String *tmp;
for (uint i=0; (tmp=it++) ; i++)
Bug#6202: ENUMs are not case sensitive even if declared BINARY
2004-10-25 14:51:26 +02:00
{
result->type_names[i]= tmp->ptr();
result->type_lengths[i]= tmp->length();
}
result->type_names[result->count]= 0; // End marker
result->type_lengths[result->count]= 0;
Import changeset
2000-07-31 21:29:14 +02:00
return result;
}
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
/*
Search after a field with given start & length
If an exact field isn't found, return longest field with starts
at right position.
NOTES
This is needed because in some .frm fields 'fieldnr' was saved wrong
RETURN
0 error
# field number +1
*/
Import changeset
2000-07-31 21:29:14 +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 uint find_field(Field **fields, uint start, uint length)
Import changeset
2000-07-31 21:29:14 +02:00
{
Field **field;
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
uint i, pos;
Import changeset
2000-07-31 21:29:14 +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
pos= 0;
for (field= fields, i=1 ; *field ; i++,field++)
Import changeset
2000-07-31 21:29:14 +02:00
{
if ((*field)->offset() == start)
{
if ((*field)->key_length() == length)
return (i);
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 (!pos || fields[pos-1]->pack_length() <
Import changeset
2000-07-31 21:29:14 +02:00
(*field)->pack_length())
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
pos= i;
Import changeset
2000-07-31 21:29:14 +02:00
}
}
return (pos);
}
fixed & added comments (according to Igor's review and a bit more)
2004-10-07 00:45:06 +02:00
/* Check that the integer is in the internal */
Import changeset
2000-07-31 21:29:14 +02:00
int set_zone(register int nr, int min_zone, int max_zone)
{
if (nr<=min_zone)
return (min_zone);
if (nr>=max_zone)
return (max_zone);
return (nr);
} /* set_zone */
/* Adjust number to next larger disk buffer */
ulong next_io_size(register ulong pos)
{
reg2 ulong offset;
if ((offset= pos & (IO_SIZE-1)))
return pos-offset+IO_SIZE;
return pos;
} /* next_io_size */
proper fix for the bug #2077: accented characters in enum/defaul values are reported incorrectly. note: bar asked me to use res.charset in ::sql_type() functions to be more consistent.
2003-12-16 14:39:33 +01:00
/*
Store an SQL quoted string.
SYNOPSIS
append_unescaped()
res result String
pos string to be quoted
length it's length
NOTE
This function works correctly with utf8 or single-byte charset strings.
May fail with some multibyte charsets though.
*/
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
proper fix for the bug #2077: accented characters in enum/defaul values are reported incorrectly. note: bar asked me to use res.charset in ::sql_type() functions to be more consistent.
2003-12-16 14:39:33 +01:00
void append_unescaped(String *res, const char *pos, uint length)
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
const char *end= pos+length;
res->append('\'');
for (; pos != end ; pos++)
Import changeset
2000-07-31 21:29:14 +02:00
{
Code style fixes. Initialize LOG_error_log before get_options to not use an uninitalized mutex in case of an error from handle_options()
2004-08-27 23:49:54 +02:00
#if defined(USE_MB) && MYSQL_VERSION_ID < 40100
table.cc: Bug #4558 Escape handling error for ENUM values in SJIS encoding
2004-08-27 09:09:28 +02:00
uint mblen;
if (use_mb(default_charset_info) &&
(mblen= my_ismbchar(default_charset_info, pos, end)))
{
res->append(pos, mblen);
pos+= mblen;
continue;
}
#endif
Import changeset
2000-07-31 21:29:14 +02:00
switch (*pos) {
case 0: /* Must be escaped for 'mysql' */
res->append('\\');
res->append('0');
break;
case '\n': /* Must be escaped for logs */
res->append('\\');
res->append('n');
break;
case '\r':
fixed & added comments (according to Igor's review and a bit more)
2004-10-07 00:45:06 +02:00
res->append('\\'); /* This gives better readability */
Import changeset
2000-07-31 21:29:14 +02:00
res->append('r');
break;
case '\\':
res->append('\\'); /* Because of the sql syntax */
res->append('\\');
break;
case '\'':
res->append('\''); /* Because of the sql syntax */
res->append('\'');
break;
default:
res->append(*pos);
break;
}
}
Extension of .frm file (not yet ready for push)
2002-06-02 20:22:20 +02:00
res->append('\'');
Import changeset
2000-07-31 21:29:14 +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
Import changeset
2000-07-31 21:29:14 +02:00
/* Create a .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
File create_frm(THD *thd, const char *name, const char *db,
Merge mysql.com:/home/my/mysql-4.1 into mysql.com:/home/my/mysql-5.0
2005-07-28 16:09:54 +02:00
const char *table, uint reclength, uchar *fileinfo,
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
HA_CREATE_INFO *create_info, uint keys)
Import changeset
2000-07-31 21:29:14 +02:00
{
register File 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
uint key_length;
Import changeset
2000-07-31 21:29:14 +02:00
ulong length;
char fill[IO_SIZE];
Fixes for bugs reported by Stefano Di Paola (stefano.dipaola@wisec.it)
2005-03-03 19:51:29 +01:00
int create_flags= O_RDWR | O_TRUNC;
if (create_info->options & HA_LEX_CREATE_TMP_TABLE)
create_flags|= O_EXCL | O_NOFOLLOW;
Import changeset
2000-07-31 21:29:14 +02:00
After merge fixes Return NULL if a time argument is given to date_add(). (Warning will be shown after Dimitri's timezone patch is pushed)
2004-06-21 09:21:20 +02:00
/* Fix this when we have new .frm files; Current limit is 4G rows (QQ) */
Fix handling of maximum value for MAX_ROWS on 64-bit platforms. (Bug #14155)
2005-11-24 02:05:59 +01:00
if (create_info->max_rows > UINT_MAX32)
create_info->max_rows= UINT_MAX32;
if (create_info->min_rows > UINT_MAX32)
create_info->min_rows= UINT_MAX32;
Fix for bug #15602: 5.0.17 test case 'create' failure.
2005-12-09 16:55:59 +01:00
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 ((file= my_create(name, CREATE_MODE, create_flags, MYF(0))) >= 0)
Import changeset
2000-07-31 21:29:14 +02:00
{
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 key_length, tmp_key_length;
uint tmp;
Import changeset
2000-07-31 21:29:14 +02:00
bzero((char*) fileinfo,64);
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
/* header */
fileinfo[0]=(uchar) 254;
fileinfo[1]= 1;
fileinfo[2]= FRM_VER+3+ test(create_info->varchar);
Finalize storage engine plugins Give BerkeleyDB savepoints Remove "enum db_type" from most of the code
2005-12-21 19:18:40 +01:00
fileinfo[3]= (uchar) ha_legacy_type(
ha_checktype(thd,ha_legacy_type(create_info->db_type),0,0));
Import changeset
2000-07-31 21:29:14 +02:00
fileinfo[4]=1;
int2store(fileinfo+6,IO_SIZE); /* Next block starts here */
key_length=keys*(7+NAME_LEN+MAX_REF_PARTS*9)+16;
Store and read engine type string in extra block of .frm.
2005-10-10 20:01:45 +02:00
length= next_io_size((ulong) (IO_SIZE+key_length+reclength+
create_info->extra_size));
Import changeset
2000-07-31 21:29:14 +02:00
int4store(fileinfo+10,length);
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
tmp_key_length= (key_length < 0xffff) ? key_length : 0xffff;
int2store(fileinfo+14,tmp_key_length);
Import changeset
2000-07-31 21:29:14 +02:00
int2store(fileinfo+16,reclength);
int4store(fileinfo+18,create_info->max_rows);
int4store(fileinfo+22,create_info->min_rows);
fileinfo[27]=2; // Use long pack-fields
create_info->table_options|=HA_OPTION_LONG_BLOB_PTR; // Use portable blob pointers
int2store(fileinfo+30,create_info->table_options);
fileinfo[32]=0; // No filename anymore
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
fileinfo[33]=5; // Mark for 5.0 frm file
Import changeset
2000-07-31 21:29:14 +02:00
int4store(fileinfo+34,create_info->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
fileinfo[38]= (create_info->default_table_charset ?
create_info->default_table_charset->number : 0);
Import changeset
2000-07-31 21:29:14 +02:00
fileinfo[40]= (uchar) create_info->row_type;
This patch is to further remove the RAID code. We removed support for people creating tables with RAID. This patch remove most of the source for this.
2006-02-12 02:21:01 +01:00
/* Next few bytes were for RAID support */
fileinfo[41]= 0;
fileinfo[42]= 0;
fileinfo[43]= 0;
fileinfo[44]= 0;
fileinfo[45]= 0;
fileinfo[46]= 0;
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
int4store(fileinfo+47, key_length);
tmp= MYSQL_VERSION_ID; // Store to avoid warning from int4store
int4store(fileinfo+51, tmp);
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
int4store(fileinfo+55, create_info->extra_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
/*
59-60 is reserved for extra_rec_buf_length,
61 for default_part_db_type
*/
int2store(fileinfo+62, create_info->key_block_size);
Import changeset
2000-07-31 21:29:14 +02:00
bzero(fill,IO_SIZE);
for (; length > IO_SIZE ; length-= IO_SIZE)
{
if (my_write(file,(byte*) fill,IO_SIZE,MYF(MY_WME | MY_NABP)))
{
VOID(my_close(file,MYF(0)));
VOID(my_delete(name,MYF(0)));
return(-1);
}
}
}
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
else
{
if (my_errno == ENOENT)
my_error(ER_BAD_DB_ERROR,MYF(0),db);
else
my_error(ER_CANT_CREATE_TABLE,MYF(0),table,my_errno);
}
Import changeset
2000-07-31 21:29:14 +02:00
return (file);
} /* create_frm */
void update_create_info_from_table(HA_CREATE_INFO *create_info, TABLE *table)
{
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_SHARE *share= table->s;
Fixed that SHOW CREATE TABLE shows all attributes
2001-09-30 04:47:35 +02:00
DBUG_ENTER("update_create_info_from_table");
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= share->max_rows;
create_info->min_rows= share->min_rows;
create_info->table_options= share->db_create_options;
create_info->avg_row_length= share->avg_row_length;
create_info->row_type= share->row_type;
create_info->default_table_charset= share->table_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
create_info->table_charset= 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
Fixed that SHOW CREATE TABLE shows all attributes
2001-09-30 04:47:35 +02:00
DBUG_VOID_RETURN;
Update copyright Fixed memory leak on shutdown (Affects the embedded version & MyODBC)
2001-12-06 13:10:51 +01:00
}
Import changeset
2000-07-31 21:29:14 +02:00
int
rename_file_ext(const char * from,const char * to,const char * ext)
{
char from_b[FN_REFLEN],to_b[FN_REFLEN];
VOID(strxmov(from_b,from,ext,NullS));
VOID(strxmov(to_b,to,ext,NullS));
return (my_rename(from_b,to_b,MYF(MY_WME)));
}
new version of help
2003-05-29 23:47:31 +02:00
/*
Allocate string field in MEM_ROOT and return it as String
SYNOPSIS
get_field()
mem MEM_ROOT for allocating
field Field for retrieving of string
res result String
RETURN VALUES
After merge fixes + bugs from last merge
2003-08-19 15:00:12 +02:00
1 string is empty
0 all ok
new version of help
2003-05-29 23:47:31 +02:00
*/
bool get_field(MEM_ROOT *mem, Field *field, String *res)
{
After merge fixes + bugs from last merge
2003-08-19 15:00:12 +02:00
char buff[MAX_FIELD_WIDTH], *to;
new version of help
2003-05-29 23:47:31 +02:00
String str(buff,sizeof(buff),&my_charset_bin);
After merge fixes + bugs from last merge
2003-08-19 15:00:12 +02:00
uint length;
::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
field->val_str(&str);
After merge fixes + bugs from last merge
2003-08-19 15:00:12 +02:00
if (!(length= str.length()))
Code cleanups during review of pushed code
2005-03-15 15:07:28 +01:00
{
res->length(0);
After merge fixes + bugs from last merge
2003-08-19 15:00:12 +02:00
return 1;
Code cleanups during review of pushed code
2005-03-15 15:07:28 +01:00
}
if (!(to= strmake_root(mem, str.ptr(), length)))
length= 0; // Safety fix
After merge fixes + bugs from last merge
2003-08-19 15:00:12 +02:00
res->set(to, length, ((Field_str*)field)->charset());
return 0;
new version of help
2003-05-29 23:47:31 +02:00
}
After merge fixes + bugs from last merge
2003-08-19 15:00:12 +02:00
Import changeset
2000-07-31 21:29:14 +02:00
/*
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
Allocate string field in MEM_ROOT and return it as NULL-terminated string
SYNOPSIS
get_field()
mem MEM_ROOT for allocating
field Field for retrieving of string
RETURN VALUES
NullS string is empty
# pointer to NULL-terminated string value of field
Import changeset
2000-07-31 21:29:14 +02:00
*/
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
char *get_field(MEM_ROOT *mem, Field *field)
Import changeset
2000-07-31 21:29:14 +02:00
{
Fixes after merge with 4.0 Cleaned up embedded library access and query cache handling Changed min stack size to 128K (to allow longer MyISAM keys) Fixed wrong priority for XOR (should be less than NEG to get -1^1 to work)
2003-12-19 15:25:50 +01:00
char buff[MAX_FIELD_WIDTH], *to;
item_func.cc, item_cmpfunc.cc, unireg.cc, table.cc, sql_base.cc: some default_charset_info were removed
2003-03-07 10:39:53 +01:00
String str(buff,sizeof(buff),&my_charset_bin);
After merge fixes + bugs from last merge
2003-08-19 15:00:12 +02:00
uint length;
::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
field->val_str(&str);
Fixes after merge with 4.0 Cleaned up embedded library access and query cache handling Changed min stack size to 128K (to allow longer MyISAM keys) Fixed wrong priority for XOR (should be less than NEG to get -1^1 to work)
2003-12-19 15:25:50 +01:00
length= str.length();
code cleanup after some reasoning
2003-12-09 20:49:48 +01:00
if (!length || !(to= (char*) alloc_root(mem,length+1)))
Import changeset
2000-07-31 21:29:14 +02:00
return NullS;
memcpy(to,str.ptr(),(uint) length);
to[length]=0;
return to;
}
Fixed handling of lower_case_table_names in SHOW TABLE STATUS, mysql_list_fields() and mysql_table_dump(). This fixes some Errcode 13 errors on Windows when deleting tables.
2003-01-29 17:56:34 +01:00
/*
Check if database name is valid
SYNPOSIS
check_db_name()
name Name of database
NOTES
If lower_case_table_names is set then database is converted to lower case
RETURN
0 ok
1 error
*/
bool check_db_name(char *name)
Import changeset
2000-07-31 21:29:14 +02:00
{
fixed Bug #2985 "Table truncated when creating another table name with Spaces" added to check_db_name, check_table_name and check_column_name test for end space
2004-03-05 19:13:33 +01:00
char *start=name;
merge with 4.0
2004-03-16 21:41:30 +01:00
/* Used to catch empty names and names with end space */
bool last_char_is_space= TRUE;
Fixed handling of lower_case_table_names in SHOW TABLE STATUS, mysql_list_fields() and mysql_table_dump(). This fixes some Errcode 13 errors on Windows when deleting tables.
2003-01-29 17:56:34 +01:00
don't modify constant strings
2004-04-06 12:31:25 +02:00
if (lower_case_table_names && name != any_db)
After merge fixes & remove compiler warnings Added lengths for all MYSQL_FIELD string parameters Changed field length to 2 byte in .frm files
2003-02-07 14:47:24 +01:00
my_casedn_str(files_charset_info, name);
Fixed handling of lower_case_table_names in SHOW TABLE STATUS, mysql_list_fields() and mysql_table_dump(). This fixes some Errcode 13 errors on Windows when deleting tables.
2003-01-29 17:56:34 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
while (*name)
{
#if defined(USE_MB) && defined(USE_MB_IDENT)
Bug#18004 Connecting crashes server when default charset is UCS2 table.cc: Fixing to use system_charset_info instead of default_charset_info. Crash happened because the "ctype" array is empty in UCS2, and thus cannot be used with my_isspace(). The reason why UCS2 appeared in this context was because of of default_charset_info variable incorrectly substituted to my_isspace(). As functions check_db_name(), check_table_name() and check_column_name() always get values in utf8, system_charset_info must be used instead. ctype_ucs2_def.test, ctype_ucs2_def-master.opt, ctype_ucs2_def.result: new file
2006-03-20 11:43:02 +01:00
last_char_is_space= my_isspace(system_charset_info, *name);
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
if (use_mb(system_charset_info))
Import changeset
2000-07-31 21:29:14 +02:00
{
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
int len=my_ismbchar(system_charset_info, name,
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
name+system_charset_info->mbmaxlen);
Import changeset
2000-07-31 21:29:14 +02:00
if (len)
{
name += len;
continue;
}
}
fixed Bug #2985 "Table truncated when creating another table name with Spaces" added to check_db_name, check_table_name and check_column_name test for end space
2004-03-05 19:13:33 +01:00
#else
extra correction of check_db_name, check_table_name and check_column_name
2004-03-12 14:55:33 +01:00
last_char_is_space= *name==' ';
Import changeset
2000-07-31 21:29:14 +02:00
#endif
name++;
}
extra correction of check_db_name, check_table_name and check_column_name
2004-03-12 14:55:33 +01:00
return last_char_is_space || (uint) (name - start) > NAME_LEN;
Import changeset
2000-07-31 21:29:14 +02:00
}
/*
Allow anything as a table name, as long as it doesn't contain an
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
' ' at the end
Import changeset
2000-07-31 21:29:14 +02:00
returns 1 on error
*/
bool check_table_name(const char *name, uint length)
{
const char *end= name+length;
Check for empty table/column names
2003-03-14 16:08:42 +01:00
if (!length || length > NAME_LEN)
return 1;
fixed Bug #2985 "Table truncated when creating another table name with Spaces" added to check_db_name, check_table_name and check_column_name test for end space
2004-03-05 19:13:33 +01:00
#if defined(USE_MB) && defined(USE_MB_IDENT)
extra correction of check_db_name, check_table_name and check_column_name
2004-03-12 14:55:33 +01:00
bool last_char_is_space= FALSE;
fixed Bug #2985 "Table truncated when creating another table name with Spaces" added to check_db_name, check_table_name and check_column_name test for end space
2004-03-05 19:13:33 +01:00
#else
if (name[length-1]==' ')
return 1;
#endif
Import changeset
2000-07-31 21:29:14 +02:00
while (name != end)
{
#if defined(USE_MB) && defined(USE_MB_IDENT)
Bug#18004 Connecting crashes server when default charset is UCS2 table.cc: Fixing to use system_charset_info instead of default_charset_info. Crash happened because the "ctype" array is empty in UCS2, and thus cannot be used with my_isspace(). The reason why UCS2 appeared in this context was because of of default_charset_info variable incorrectly substituted to my_isspace(). As functions check_db_name(), check_table_name() and check_column_name() always get values in utf8, system_charset_info must be used instead. ctype_ucs2_def.test, ctype_ucs2_def-master.opt, ctype_ucs2_def.result: new file
2006-03-20 11:43:02 +01:00
last_char_is_space= my_isspace(system_charset_info, *name);
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
if (use_mb(system_charset_info))
Import changeset
2000-07-31 21:29:14 +02:00
{
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
int len=my_ismbchar(system_charset_info, name, end);
Import changeset
2000-07-31 21:29:14 +02:00
if (len)
{
name += len;
continue;
}
}
#endif
name++;
}
fixed Bug #2985 "Table truncated when creating another table name with Spaces" added to check_db_name, check_table_name and check_column_name test for end space
2004-03-05 19:13:33 +01:00
#if defined(USE_MB) && defined(USE_MB_IDENT)
extra correction of check_db_name, check_table_name and check_column_name
2004-03-12 14:55:33 +01:00
return last_char_is_space;
#else
Import changeset
2000-07-31 21:29:14 +02:00
return 0;
extra correction of check_db_name, check_table_name and check_column_name
2004-03-12 14:55:33 +01:00
#endif
Import changeset
2000-07-31 21:29:14 +02:00
}
Fixes to newly pushed code
2004-03-12 14:56:28 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
bool check_column_name(const char *name)
{
Check for empty table/column names
2003-03-14 16:08:42 +01:00
const char *start= name;
merge with 4.0
2004-03-16 21:41:30 +01:00
bool last_char_is_space= TRUE;
extra correction of check_db_name, check_table_name and check_column_name
2004-03-12 14:55:33 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
while (*name)
{
#if defined(USE_MB) && defined(USE_MB_IDENT)
Bug#18004 Connecting crashes server when default charset is UCS2 table.cc: Fixing to use system_charset_info instead of default_charset_info. Crash happened because the "ctype" array is empty in UCS2, and thus cannot be used with my_isspace(). The reason why UCS2 appeared in this context was because of of default_charset_info variable incorrectly substituted to my_isspace(). As functions check_db_name(), check_table_name() and check_column_name() always get values in utf8, system_charset_info must be used instead. ctype_ucs2_def.test, ctype_ucs2_def-master.opt, ctype_ucs2_def.result: new file
2006-03-20 11:43:02 +01:00
last_char_is_space= my_isspace(system_charset_info, *name);
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
if (use_mb(system_charset_info))
Import changeset
2000-07-31 21:29:14 +02:00
{
New ctype functions/macros to support many charsets at a time
2002-03-12 18:37:58 +01:00
int len=my_ismbchar(system_charset_info, name,
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
name+system_charset_info->mbmaxlen);
Import changeset
2000-07-31 21:29:14 +02:00
if (len)
{
name += len;
continue;
}
}
fixed Bug #2985 "Table truncated when creating another table name with Spaces" added to check_db_name, check_table_name and check_column_name test for end space
2004-03-05 19:13:33 +01:00
#else
extra correction of check_db_name, check_table_name and check_column_name
2004-03-12 14:55:33 +01:00
last_char_is_space= *name==' ';
Import changeset
2000-07-31 21:29:14 +02:00
#endif
if (*name == NAMES_SEP_CHAR)
return 1;
name++;
}
Check for empty table/column names
2003-03-14 16:08:42 +01:00
/* Error if empty or too long column name */
merge with 4.0
2004-03-16 21:41:30 +01:00
return last_char_is_space || (uint) (name - start) > NAME_LEN;
Import changeset
2000-07-31 21:29:14 +02:00
}
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
/*
Checks whether a table is intact. Should be done *just* after the table has
been opened.
Synopsis
table_check_intact()
table - the table to check
table_f_count - expected number of columns in the table
table_def - expected structure of the table (column name and type)
last_create_time- the table->file->create_time of the table in memory
we have checked last time
error_num - ER_XXXX from the error messages file. When 0 no error
is sent to the client in case types does not match.
If different col number either
ER_COL_COUNT_DOESNT_MATCH_PLEASE_UPDATE or
ER_COL_COUNT_DOESNT_MATCH_CORRUPTED is used
RETURNS
0 - OK
1 - There was an error
*/
my_bool
table_check_intact(TABLE *table, uint table_f_count,
TABLE_FIELD_W_TYPE *table_def, time_t *last_create_time,
int error_num)
{
uint i;
my_bool error= FALSE;
my_bool fields_diff_count;
DBUG_ENTER("table_check_intact");
DBUG_PRINT("info",("table=%s expected_count=%d",table->alias, table_f_count));
DBUG_PRINT("info",("last_create_time=%d", *last_create_time));
if ((fields_diff_count= (table->s->fields != table_f_count)) ||
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
(*last_create_time != table->file->stats.create_time))
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
{
DBUG_PRINT("info", ("I am suspecting, checking table"));
if (fields_diff_count)
{
// previous MySQL version
error= TRUE;
if (MYSQL_VERSION_ID > table->s->mysql_version)
reduce error log spamming by old privilege tables. If they're old, report it once, don't whine about every missing column.
2006-05-30 18:18:13 +02:00
{
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
my_error(ER_COL_COUNT_DOESNT_MATCH_PLEASE_UPDATE, MYF(0), table->alias,
table_f_count, table->s->fields, table->s->mysql_version,
MYSQL_VERSION_ID);
reduce error log spamming by old privilege tables. If they're old, report it once, don't whine about every missing column.
2006-05-30 18:18:13 +02:00
sql_print_error(ER(ER_COL_COUNT_DOESNT_MATCH_PLEASE_UPDATE),
table->alias, table_f_count, table->s->fields,
table->s->mysql_version, MYSQL_VERSION_ID);
DBUG_RETURN(error);
}
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
else if (MYSQL_VERSION_ID == table->s->mysql_version)
reduce error log spamming by old privilege tables. If they're old, report it once, don't whine about every missing column.
2006-05-30 18:18:13 +02:00
{
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
my_error(ER_COL_COUNT_DOESNT_MATCH_CORRUPTED,MYF(0), table->alias,
table_f_count, table->s->fields);
reduce error log spamming by old privilege tables. If they're old, report it once, don't whine about every missing column.
2006-05-30 18:18:13 +02:00
sql_print_error(ER(ER_COL_COUNT_DOESNT_MATCH_CORRUPTED), table->alias,
table_f_count, table->s->fields);
}
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
else
Fixed some issues found by valgrind (one testcase, one memory leak and some accesses to not initialized memory)
2006-06-06 01:47:30 +02:00
{
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
/*
moving from newer mysql to older one -> let's say not an error but
Fixed some issues found by valgrind (one testcase, one memory leak and some accesses to not initialized memory)
2006-06-06 01:47:30 +02:00
will check the definition afterwards. If a column was added at the
end then we don't care much since it's not in the middle.
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
*/
error= FALSE;
Fixed some issues found by valgrind (one testcase, one memory leak and some accesses to not initialized memory)
2006-06-06 01:47:30 +02:00
}
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
}
//definitely something has changed
char buffer[255];
Fixed some issues found by valgrind (one testcase, one memory leak and some accesses to not initialized memory)
2006-06-06 01:47:30 +02:00
for (i=0 ; i < table_f_count; i++, table_def++)
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
{
String sql_type(buffer, sizeof(buffer), system_charset_info);
sql_type.length(0);
/*
name changes are not fatal, we use sequence numbers => no prob for us
but this can show tampered table or broken table.
*/
Fixed some issues found by valgrind (one testcase, one memory leak and some accesses to not initialized memory)
2006-06-06 01:47:30 +02:00
if (i < table->s->fields)
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
{
Fixed some issues found by valgrind (one testcase, one memory leak and some accesses to not initialized memory)
2006-06-06 01:47:30 +02:00
Field *field= table->field[i];
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
if (strncmp(field->field_name, table_def->name.str,
table_def->name.length))
{
sql_print_error("(%s) Expected field %s at position %d, found %s",
table->alias, table_def->name.str, i,
field->field_name);
}
/*
IF the type does not match than something is really wrong
Check up to length - 1. Why?
1. datetime -> datetim -> the same
2. int(11) -> int(11 -> the same
3. set('one','two') -> set('one','two'
so for sets if the same prefix is there it's ok if more are
added as part of the set. The same is valid for enum. So a new
table running on a old server will be valid.
*/
field->sql_type(sql_type);
fix for bug #17619 Scheduler race conditions - Scheduler is either initialized at server start or never. Starting & stopping is now suspending & resuming. - The scheduler has clear OO interface - Now all calls to the scheduler are synchronous - GLOBAL event_scheduler uses thd::sys_var_tmp (see set_var.cc) - External API is encapsulated into class Events - Includes fixes for all comments of Kostja's review of 19.05.2005 Starting to merge into 5.1-release (5.1.10) and push
2006-05-22 20:46:13 +02:00
if (strncmp(sql_type.c_ptr_safe(), table_def->type.str,
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
table_def->type.length - 1))
{
sql_print_error("(%s) Expected field %s at position %d to have type "
"%s, found %s", table->alias, table_def->name.str,
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
i, table_def->type.str, sql_type.c_ptr_safe());
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
error= TRUE;
}
else if (table_def->cset.str && !field->has_charset())
{
sql_print_error("(%s) Expected field %s at position %d to have "
"character set '%s' but found no such", table->alias,
table_def->name.str, i, table_def->cset.str);
error= TRUE;
}
else if (table_def->cset.str &&
strcmp(field->charset()->csname, table_def->cset.str))
{
sql_print_error("(%s) Expected field %s at position %d to have "
"character set '%s' but found '%s'", table->alias,
table_def->name.str, i, table_def->cset.str,
field->charset()->csname);
error= TRUE;
}
}
else
{
sql_print_error("(%s) Expected field %s at position %d to have type %s "
Fixed some issues found by valgrind (one testcase, one memory leak and some accesses to not initialized memory)
2006-06-06 01:47:30 +02:00
" but no field found.", table->alias,
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
table_def->name.str, i, table_def->type.str);
error= TRUE;
}
}
if (!error)
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
*last_create_time= table->file->stats.create_time;
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
else if (!fields_diff_count && error_num)
my_error(error_num,MYF(0), table->alias, table_f_count, table->s->fields);
}
else
{
post-merge fixes of bug 16548
2006-02-14 20:10:51 +01:00
DBUG_PRINT("info", ("Table seems ok without thorough checking."));
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
*last_create_time= table->file->stats.create_time;
- final fixes for bug#16431 (Events: An event which alters itself disappears) - fix for bug#16423 (Events: SHOW CREATE EVENT doesn't work) - this Changeset commits makes CREATE/UPDATE/DELETE EVENT real DDL statements by committing the currently open transaction before they are executed. - this Changeset also fixes a trailing space problem since the very early days of the internal cron - adds sophisticated checking of whether mysql.event was tampered accidentally or with purpose by an user. - adds a lot of inline function documentation - documents everything left uncodumented - INTERVAL_XXXX to XXX in I_S.EVENTS.INTERVAL_FIELD WL#1034 (Internal CRON)
2006-02-14 16:20:48 +01:00
}
DBUG_RETURN(error);
}
A fix and a test case for Bug#6513 "Test Suite: Values inserted by using cursor is interpreted latin1 character and Bug#9819 "Cursors: Mysql Server Crash while fetching from table with 5 million records." A fix for a possible memory leak when fetching into an SP cursor in a long loop. The patch uses a common implementation of cursors in the binary protocol and in stored procedures and implements materialized cursors. For implementation details, see comments in sql_cursor.cc
2005-09-22 00:11:21 +02:00
/*
Create Item_field for each column in the table.
SYNPOSIS
st_table::fill_item_list()
item_list a pointer to an empty list used to store items
DESCRIPTION
Create Item_field object for each column in the table and
initialize it with the corresponding Field. New items are
created in the current THD memory root.
RETURN VALUE
0 success
1 out of memory
*/
bool st_table::fill_item_list(List<Item> *item_list) const
{
/*
All Item_field's created using a direct pointer to a field
are fixed in Item_field constructor.
*/
for (Field **ptr= field; *ptr; ptr++)
{
Item_field *item= new Item_field(*ptr);
if (!item || item_list->push_back(item))
return TRUE;
}
return FALSE;
}
/*
Reset an existing list of Item_field items to point to the
Fields of this table.
SYNPOSIS
st_table::fill_item_list()
item_list a non-empty list with Item_fields
DESCRIPTION
This is a counterpart of fill_item_list used to redirect
Item_fields to the fields of a newly created table.
The caller must ensure that number of items in the item_list
is the same as the number of columns in the table.
*/
void st_table::reset_item_list(List<Item> *item_list) const
{
List_iterator_fast<Item> it(*item_list);
for (Field **ptr= field; *ptr; ptr++)
{
Item_field *item_field= (Item_field*) it++;
DBUG_ASSERT(item_field != 0);
item_field->reset_field(*ptr);
}
}
Import changeset
2000-07-31 21:29:14 +02:00
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
/*
calculate md5 of query
SYNOPSIS
st_table_list::calc_md5()
buffer buffer for md5 writing
*/
Cleanup of new code pushed into 5.0 since last pull Merged the different find_xxxx_table_in_list functions to one + some inline functions
2004-09-03 20:43:04 +02:00
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
void st_table_list::calc_md5(char *buffer)
{
my_MD5_CTX context;
Cleanup of new code pushed into 5.0 since last pull Merged the different find_xxxx_table_in_list functions to one + some inline functions
2004-09-03 20:43:04 +02:00
uchar digest[16];
my_MD5Init(&context);
my_MD5Update(&context,(uchar *) query.str, query.length);
my_MD5Final(digest, &context);
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
sprintf((char *) buffer,
"%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x",
digest[0], digest[1], digest[2], digest[3],
digest[4], digest[5], digest[6], digest[7],
digest[8], digest[9], digest[10], digest[11],
digest[12], digest[13], digest[14], digest[15]);
}
/*
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
set underlying TABLE for table place holder of VIEW
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
Post review changes
2005-02-23 17:51:10 +01:00
DESCRIPTION
Replace all views that only uses one table with the table itself.
This allows us to treat the view as a simple table and even update
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
it (it is a kind of optimisation)
Post review changes
2005-02-23 17:51:10 +01:00
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
SYNOPSIS
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
st_table_list::set_underlying_merge()
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
*/
Cleanup of new code pushed into 5.0 since last pull Merged the different find_xxxx_table_in_list functions to one + some inline functions
2004-09-03 20:43:04 +02:00
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
void st_table_list::set_underlying_merge()
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
{
Post review changes
2005-02-23 17:51:10 +01:00
TABLE_LIST *tbl;
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
if ((tbl= merge_underlying_list))
merge
2004-11-21 19:08:12 +01:00
{
Post review changes
2005-02-23 17:51:10 +01:00
/* This is a view. Process all tables of view */
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
DBUG_ASSERT(view && effective_algorithm == VIEW_ALGORITHM_MERGE);
Post review changes
2005-02-23 17:51:10 +01:00
do
{
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
if (tbl->merge_underlying_list) // This is a view
Post review changes
2005-02-23 17:51:10 +01:00
{
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
DBUG_ASSERT(tbl->view &&
tbl->effective_algorithm == VIEW_ALGORITHM_MERGE);
Post review changes
2005-02-23 17:51:10 +01:00
/*
This is the only case where set_ancestor is called on an object
that may not be a view (in which case ancestor is 0)
*/
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
tbl->merge_underlying_list->set_underlying_merge();
Post review changes
2005-02-23 17:51:10 +01:00
}
} while ((tbl= tbl->next_local));
Many files: Fixed bug #8528. Representation for single-table views was made similar to representation for multi-table views. view.test: Added test case for bug #8528. view.result: Added test case for bug #8528. Fixed other test cases.
2005-05-11 01:31:13 +02:00
if (!multitable_view)
Post review changes
2005-02-23 17:51:10 +01:00
{
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
table= merge_underlying_list->table;
schema_table= merge_underlying_list->schema_table;
Post review changes
2005-02-23 17:51:10 +01:00
}
merge
2004-11-21 19:08:12 +01:00
}
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
}
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
/*
setup fields of placeholder of merged VIEW
SYNOPSIS
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
st_table_list::setup_underlying()
new behaviour of CHECK option build, for mor efficience and more correct: check option build only according most top VIEW CHECK OPTION TYPE (BUG#5993)
2004-10-21 17:10:59 +02:00
thd - thread handler
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
fixed & added comments (according to Igor's review and a bit more)
2004-10-07 00:45:06 +02:00
DESCRIPTION
It is:
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
- preparing translation table for view columns
fixed & added comments (according to Igor's review and a bit more)
2004-10-07 00:45:06 +02:00
If there are underlying view(s) procedure first will be called for them.
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
RETURN
Post review changes
2005-02-23 17:51:10 +01:00
FALSE - OK
TRUE - error
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
*/
Cleanup of new code pushed into 5.0 since last pull Merged the different find_xxxx_table_in_list functions to one + some inline functions
2004-09-03 20:43:04 +02:00
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
bool st_table_list::setup_underlying(THD *thd)
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
{
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
DBUG_ENTER("st_table_list::setup_underlying");
if (!field_translation && merge_underlying_list)
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
{
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
Field_translator *transl;
SELECT_LEX *select= &view->select_lex;
Item *item;
TABLE_LIST *tbl;
List_iterator_fast<Item> it(select->item_list);
uint field_count= 0;
if (check_stack_overrun(thd, STACK_MIN_SIZE, (char *)&field_count))
{
Post review changes
2005-02-23 17:51:10 +01:00
DBUG_RETURN(TRUE);
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
}
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
for (tbl= merge_underlying_list; tbl; tbl= tbl->next_local)
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
{
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
if (tbl->merge_underlying_list &&
tbl->setup_underlying(thd))
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
{
DBUG_RETURN(TRUE);
}
}
Post review changes
2005-02-23 17:51:10 +01:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
/* Create view fields translation table */
if (!(transl=
Rename: - current_arena to stmt_arena: the thread may have more than one 'current' arenas: one for runtime data, and one for the parsed tree of a statement. Only one of them is active at any moment. - set_item_arena -> set_query_arena, because Item_arena was renamed to Query_arena a while ago - set_n_backup_item_arena -> set_n_backup_active_arena; the active arena is the arena thd->mem_root and thd->free_list are currently pointing at. - restore_backup_item_arena -> restore_active_arena (with the same rationale) - change_arena_if_needed -> activate_stmt_arena_if_needed; this method sets thd->stmt_arena active if it's not done yet.
2005-09-02 15:21:19 +02:00
(Field_translator*)(thd->stmt_arena->
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
alloc(select->item_list.elements *
sizeof(Field_translator)))))
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
{
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
DBUG_RETURN(TRUE);
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
}
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
while ((item= it++))
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
{
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
transl[field_count].name= item->name;
transl[field_count++].item= item;
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
}
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
field_translation= transl;
field_translation_end= transl + field_count;
/* TODO: use hash for big number of fields */
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
/* full text function moving to current select */
if (view->select_lex.ftfunc_list->elements)
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
{
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
Item_func_match *ifm;
SELECT_LEX *current_select= thd->lex->current_select;
List_iterator_fast<Item_func_match>
li(*(view->select_lex.ftfunc_list));
while ((ifm= li++))
current_select->ftfunc_list->push_front(ifm);
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
}
}
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
DBUG_RETURN(FALSE);
}
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
Post review changes
2005-02-23 17:51:10 +01:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
/*
Prepare where expression of view
error message interception (Bug#4598) error handling of view initialization fixed (Bug#4599)
2004-07-21 03:26:20 +02:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
SYNOPSIS
st_table_list::prep_where()
thd - thread handler
conds - condition of this JOIN
no_where_clause - do not build WHERE or ON outer qwery do not need it
(it is INSERT), we do not need conds if this flag is set
fixed using VIEW fields (BUG#4617)
2004-07-23 08:20:58 +02:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
NOTE: have to be called befor CHECK OPTION preparation, because it makes
fix_fields for view WHERE clause
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
RETURN
FALSE - OK
TRUE - error
*/
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
bool st_table_list::prep_where(THD *thd, Item **conds,
bool no_where_clause)
{
DBUG_ENTER("st_table_list::prep_where");
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
for (TABLE_LIST *tbl= merge_underlying_list; tbl; tbl= tbl->next_local)
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
{
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
if (tbl->view && tbl->prep_where(thd, conds, no_where_clause))
{
DBUG_RETURN(TRUE);
}
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
}
view with WHERE in nested join (BUG#5511)
2004-09-10 21:39:04 +02:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
if (where)
{
if (!where->fixed && where->fix_fields(thd, &where))
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
{
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
DBUG_RETURN(TRUE);
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
}
postmerge fix
2004-10-04 12:45:35 +02:00
/*
check that it is not VIEW in which we insert with INSERT SELECT
(in this case we can't add view WHERE condition to main SELECT_LEX)
*/
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
if (!no_where_clause && !where_processed)
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
{
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
TABLE_LIST *tbl= this;
Rename: - current_arena to stmt_arena: the thread may have more than one 'current' arenas: one for runtime data, and one for the parsed tree of a statement. Only one of them is active at any moment. - set_item_arena -> set_query_arena, because Item_arena was renamed to Query_arena a while ago - set_n_backup_item_arena -> set_n_backup_active_arena; the active arena is the arena thd->mem_root and thd->free_list are currently pointing at. - restore_backup_item_arena -> restore_active_arena (with the same rationale) - change_arena_if_needed -> activate_stmt_arena_if_needed; this method sets thd->stmt_arena active if it's not done yet.
2005-09-02 15:21:19 +02:00
Query_arena *arena= thd->stmt_arena, backup;
arena= thd->activate_stmt_arena_if_needed(&backup); // For easier test
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
postmerge fix
2004-10-04 12:45:35 +02:00
/* Go up to join tree and try to find left join */
for (; tbl; tbl= tbl->embedding)
CHECK OPTIONs added (WL#1983)
2004-09-03 14:18:40 +02:00
{
postmerge fix
2004-10-04 12:45:35 +02:00
if (tbl->outer_join)
{
/*
Store WHERE condition to ON expression for outer join, because
fixed & added comments (according to Igor's review and a bit more)
2004-10-07 00:45:06 +02:00
we can't use WHERE to correctly execute left joins on VIEWs and
postmerge fix
2004-10-04 12:45:35 +02:00
this expression will not be moved to WHERE condition (i.e. will
be clean correctly for PS/SP)
*/
tbl->on_expr= and_conds(tbl->on_expr, where);
break;
}
CHECK OPTIONs added (WL#1983)
2004-09-03 14:18:40 +02:00
}
postmerge fix
2004-10-04 12:45:35 +02:00
if (tbl == 0)
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
*conds= and_conds(*conds, where);
if (arena)
Rename: - current_arena to stmt_arena: the thread may have more than one 'current' arenas: one for runtime data, and one for the parsed tree of a statement. Only one of them is active at any moment. - set_item_arena -> set_query_arena, because Item_arena was renamed to Query_arena a while ago - set_n_backup_item_arena -> set_n_backup_active_arena; the active arena is the arena thd->mem_root and thd->free_list are currently pointing at. - restore_backup_item_arena -> restore_active_arena (with the same rationale) - change_arena_if_needed -> activate_stmt_arena_if_needed; this method sets thd->stmt_arena active if it's not done yet.
2005-09-02 15:21:19 +02:00
thd->restore_active_arena(arena, &backup);
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
where_processed= TRUE;
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
}
}
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
DBUG_RETURN(FALSE);
}
/*
Prepare check option expression of table
SYNOPSIS
st_table_list::prep_check_option()
thd - thread handler
check_opt_type - WITH CHECK OPTION type (VIEW_CHECK_NONE,
VIEW_CHECK_LOCAL, VIEW_CHECK_CASCADED)
we use this parameter instead of direct check of
effective_with_check to change type of underlying
views to VIEW_CHECK_CASCADED if outer view have
such option and prevent processing of underlying
view check options if outer view have just
VIEW_CHECK_LOCAL option.
NOTE
This method build check options for every call
(usual execution or every SP/PS call)
This method have to be called after WHERE preparation
(st_table_list::prep_where)
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
RETURN
FALSE - OK
TRUE - error
*/
bool st_table_list::prep_check_option(THD *thd, uint8 check_opt_type)
{
DBUG_ENTER("st_table_list::prep_check_option");
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
for (TABLE_LIST *tbl= merge_underlying_list; tbl; tbl= tbl->next_local)
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
{
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
/* see comment of check_opt_type parameter */
if (tbl->view &&
tbl->prep_check_option(thd,
((check_opt_type == VIEW_CHECK_CASCADED) ?
VIEW_CHECK_CASCADED :
VIEW_CHECK_NONE)))
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
{
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
DBUG_RETURN(TRUE);
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
}
}
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
if (check_opt_type)
full text function moving to current select (BUG#4822)
2004-08-25 15:14:42 +02:00
{
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
Item *item= 0;
if (where)
{
DBUG_ASSERT(where->fixed);
item= where->copy_andor_structure(thd);
}
if (check_opt_type == VIEW_CHECK_CASCADED)
{
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
for (TABLE_LIST *tbl= merge_underlying_list; tbl; tbl= tbl->next_local)
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
{
if (tbl->check_option)
item= and_conds(item, tbl->check_option);
}
}
if (item)
thd->change_item_tree(&check_option, item);
}
if (check_option)
{
const char *save_where= thd->where;
thd->where= "check option";
if (!check_option->fixed &&
check_option->fix_fields(thd, &check_option) ||
check_option->check_cols(1))
{
DBUG_RETURN(TRUE);
}
thd->where= save_where;
full text function moving to current select (BUG#4822)
2004-08-25 15:14:42 +02:00
}
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
DBUG_RETURN(FALSE);
}
full text function moving to current select (BUG#4822)
2004-08-25 15:14:42 +02:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
/*
Hide errors which show view underlying table information
SYNOPSIS
st_table_list::hide_view_error()
thd thread handler
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
*/
void st_table_list::hide_view_error(THD *thd)
{
Fix for bug#7212: information_schema: "Can't find file" errors if storage engine gone(after review)
2004-12-16 14:31:36 +01:00
/* Hide "Unknown column" or "Unknown function" error */
if (thd->net.last_errno == ER_BAD_FIELD_ERROR ||
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
thd->net.last_errno == ER_SP_DOES_NOT_EXIST ||
thd->net.last_errno == ER_PROCACCESS_DENIED_ERROR ||
BUG#18681: View privileges are broken The check for view security was lacking several points : 1. Check with the right set of permissions : for each table ref that participates in a view there were the right credentials to use in it's security_ctx member, but these weren't used for checking the credentials. This makes hard enforcing the SQL SECURITY DEFINER|INVOKER property consistently. 2. Because of the above the security checking for views was just ruled out in explicit ways in several places. 3. The security was checked only for the columns of the tables that are brought into the query from a view. So if there is no column reference outside of the view definition it was not detecting the lack of access to the tables in the view in SQL SECURITY INVOKER mode. The fix below tries to fix the above 3 points.
2006-05-26 10:47:53 +02:00
thd->net.last_errno == ER_COLUMNACCESS_DENIED_ERROR ||
thd->net.last_errno == ER_TABLEACCESS_DENIED_ERROR)
error message interception (Bug#4598) error handling of view initialization fixed (Bug#4599)
2004-07-21 03:26:20 +02:00
{
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
TABLE_LIST *top= top_table();
error message interception (Bug#4598) error handling of view initialization fixed (Bug#4599)
2004-07-21 03:26:20 +02:00
thd->clear_error();
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
my_error(ER_VIEW_INVALID, MYF(0), top->view_db.str, top->view_name.str);
error message interception (Bug#4598) error handling of view initialization fixed (Bug#4599)
2004-07-21 03:26:20 +02:00
}
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
else if (thd->net.last_errno == ER_NO_DEFAULT_FOR_FIELD)
{
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
TABLE_LIST *top= top_table();
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
thd->clear_error();
// TODO: make correct error message
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
my_error(ER_NO_DEFAULT_FOR_VIEW_FIELD, MYF(0),
top->view_db.str, top->view_name.str);
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
}
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
}
- stackoverflow check added for view of view processing - fixed bug in join view processing - postreview fixes (BUG#9398 & BUG#8703)
2005-04-03 00:23:45 +02:00
/*
Find underlying base tables (TABLE_LIST) which represent given
table_to_find (TABLE)
SYNOPSIS
st_table_list::find_underlying_table()
table_to_find table to find
RETURN
0 table is not found
found table reference
*/
st_table_list *st_table_list::find_underlying_table(TABLE *table_to_find)
{
/* is this real table and table which we are looking for? */
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
if (table == table_to_find && merge_underlying_list == 0)
- stackoverflow check added for view of view processing - fixed bug in join view processing - postreview fixes (BUG#9398 & BUG#8703)
2005-04-03 00:23:45 +02:00
return this;
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
for (TABLE_LIST *tbl= merge_underlying_list; tbl; tbl= tbl->next_local)
- stackoverflow check added for view of view processing - fixed bug in join view processing - postreview fixes (BUG#9398 & BUG#8703)
2005-04-03 00:23:45 +02:00
{
TABLE_LIST *result;
if ((result= tbl->find_underlying_table(table_to_find)))
return result;
}
return 0;
}
postreview fixes
2004-11-08 00:54:23 +01:00
/*
cleunup items belonged to view fields translation table
SYNOPSIS
st_table_list::cleanup_items()
*/
void st_table_list::cleanup_items()
{
if (!field_translation)
return;
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
for (Field_translator *transl= field_translation;
transl < field_translation_end;
transl++)
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
transl->item->walk(&Item::cleanup_processor, 0, 0);
postreview fixes
2004-11-08 00:54:23 +01:00
}
post review changes: CHECK OPTION moved to one function view name added to error messages
2004-09-29 15:35:01 +02:00
/*
check CHECK OPTION condition
SYNOPSIS
check_option()
ignore_failure ignore check option fail
RETURN
VIEW_CHECK_OK OK
VIEW_CHECK_ERROR FAILED
VIEW_CHECK_SKIP FAILED, but continue
*/
int st_table_list::view_check_option(THD *thd, bool ignore_failure)
{
if (check_option && check_option->val_int() == 0)
{
issue correct error message in case of view presence for duplicated table on update (BUG#10773) frequently used command sequence replaced with inline function
2005-08-02 21:54:49 +02:00
TABLE_LIST *view= top_table();
post review changes: CHECK OPTION moved to one function view name added to error messages
2004-09-29 15:35:01 +02:00
if (ignore_failure)
{
push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_ERROR,
ER_VIEW_CHECK_FAILED, ER(ER_VIEW_CHECK_FAILED),
Many files: Fixed bug #8528. Representation for single-table views was made similar to representation for multi-table views. view.test: Added test case for bug #8528. view.result: Added test case for bug #8528. Fixed other test cases.
2005-05-11 01:31:13 +02:00
view->view_db.str, view->view_name.str);
post review changes: CHECK OPTION moved to one function view name added to error messages
2004-09-29 15:35:01 +02:00
return(VIEW_CHECK_SKIP);
}
else
{
Many files: Fixed bug #8528. Representation for single-table views was made similar to representation for multi-table views. view.test: Added test case for bug #8528. view.result: Added test case for bug #8528. Fixed other test cases.
2005-05-11 01:31:13 +02:00
my_error(ER_VIEW_CHECK_FAILED, MYF(0), view->view_db.str, view->view_name.str);
post review changes: CHECK OPTION moved to one function view name added to error messages
2004-09-29 15:35:01 +02:00
return(VIEW_CHECK_ERROR);
}
}
return(VIEW_CHECK_OK);
}
support of join view updateability (WL#1809)
2004-09-15 22:42:56 +02:00
/*
spelling fixed/comments added (postreview fixes)
2004-11-25 08:28:32 +01:00
Find table in underlying tables by mask and check that only this
post merge
2004-11-21 18:33:49 +01:00
table belong to given mask
support of join view updateability (WL#1809)
2004-09-15 22:42:56 +02:00
SYNOPSIS
st_table_list::check_single_table()
table reference on variable where to store found table
(should be 0 on call, to find table, or point to table for
unique test)
map bit mask of tables
Many files: Fixed bug #8528. Representation for single-table views was made similar to representation for multi-table views. view.test: Added test case for bug #8528. view.result: Added test case for bug #8528. Fixed other test cases.
2005-05-11 01:31:13 +02:00
view view for which we are looking table
support of join view updateability (WL#1809)
2004-09-15 22:42:56 +02:00
RETURN
post merge
2004-11-21 18:33:49 +01:00
FALSE table not found or found only one
TRUE found several tables
support of join view updateability (WL#1809)
2004-09-15 22:42:56 +02:00
*/
Many files: Fixed bug #8528. Representation for single-table views was made similar to representation for multi-table views. view.test: Added test case for bug #8528. view.result: Added test case for bug #8528. Fixed other test cases.
2005-05-11 01:31:13 +02:00
bool st_table_list::check_single_table(st_table_list **table, table_map map,
st_table_list *view)
support of join view updateability (WL#1809)
2004-09-15 22:42:56 +02:00
{
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
for (TABLE_LIST *tbl= merge_underlying_list; tbl; tbl= tbl->next_local)
support of join view updateability (WL#1809)
2004-09-15 22:42:56 +02:00
{
if (tbl->table)
{
if (tbl->table->map & map)
{
if (*table)
post merge
2004-11-21 18:33:49 +01:00
return TRUE;
Code cleanups during code reviews Ensure we get error if INSERT IGNORE ... SELECT fails Fixed wrong key_part->key_length usage in index_merge
2005-06-01 15:35:09 +02:00
*table= tbl;
tbl->check_option= view->check_option;
support of join view updateability (WL#1809)
2004-09-15 22:42:56 +02:00
}
}
Code cleanups during code reviews Ensure we get error if INSERT IGNORE ... SELECT fails Fixed wrong key_part->key_length usage in index_merge
2005-06-01 15:35:09 +02:00
else if (tbl->check_single_table(table, map, view))
return TRUE;
support of join view updateability (WL#1809)
2004-09-15 22:42:56 +02:00
}
post merge
2004-11-21 18:33:49 +01:00
return FALSE;
support of join view updateability (WL#1809)
2004-09-15 22:42:56 +02:00
}
/*
Set insert_values buffer
SYNOPSIS
set_insert_values()
mem_root memory pool for allocating
RETURN
FALSE - OK
TRUE - out of memory
*/
bool st_table_list::set_insert_values(MEM_ROOT *mem_root)
{
if (table)
{
if (!table->insert_values &&
!(table->insert_values= (byte *)alloc_root(mem_root,
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->rec_buff_length)))
support of join view updateability (WL#1809)
2004-09-15 22:42:56 +02:00
return TRUE;
}
else
{
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
DBUG_ASSERT(view && merge_underlying_list);
for (TABLE_LIST *tbl= merge_underlying_list; tbl; tbl= tbl->next_local)
support of join view updateability (WL#1809)
2004-09-15 22:42:56 +02:00
if (tbl->set_insert_values(mem_root))
return TRUE;
}
return FALSE;
}
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
/*
Test if this is a leaf with respect to name resolution.
SYNOPSIS
st_table_list::is_leaf_for_name_resolution()
DESCRIPTION
A table reference is a leaf with respect to name resolution if
it is either a leaf node in a nested join tree (table, view,
schema table, subquery), or an inner node that represents a
NATURAL/USING join, or a nested join with materialized join
columns.
RETURN
TRUE if a leaf, FALSE otherwise.
*/
bool st_table_list::is_leaf_for_name_resolution()
{
return (view || is_natural_join || is_join_columns_complete ||
!nested_join);
}
/*
Retrieve the first (left-most) leaf in a nested join tree with
respect to name resolution.
SYNOPSIS
st_table_list::first_leaf_for_name_resolution()
DESCRIPTION
Given that 'this' is a nested table reference, recursively walk
down the left-most children of 'this' until we reach a leaf
table reference with respect to name resolution.
IMPLEMENTATION
The left-most child of a nested table reference is the last element
in the list of children because the children are inserted in
reverse order.
RETURN
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
If 'this' is a nested table reference - the left-most child of
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
the tree rooted in 'this',
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
else return 'this'
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
*/
TABLE_LIST *st_table_list::first_leaf_for_name_resolution()
{
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
TABLE_LIST *cur_table_ref;
NESTED_JOIN *cur_nested_join;
LINT_INIT(cur_table_ref);
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
if (is_leaf_for_name_resolution())
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
return this;
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
DBUG_ASSERT(nested_join);
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
for (cur_nested_join= nested_join;
cur_nested_join;
cur_nested_join= cur_table_ref->nested_join)
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
{
List_iterator_fast<TABLE_LIST> it(cur_nested_join->join_list);
cur_table_ref= it++;
/*
Fix for BUG#13597 - columns in ON condition not resolved if references a table in a nested right join. The problem was in that when finding the last table reference in a nested join tree, the procedure doing the iteration over the right-most branches of a join tree was testing for RIGHT JOINs the table reference that represents the join, and not the second operand of the JOIN. Currently the information whether a join is LEFT/RIGHT is stored not on the join object itself, but on one of its operands.
2005-09-30 09:39:17 +02:00
If the current nested join is a RIGHT JOIN, the operands in
'join_list' are in reverse order, thus the first operand is
already at the front of the list. Otherwise the first operand
is in the end of the list of join operands.
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
*/
if (!(cur_table_ref->outer_join & JOIN_TYPE_RIGHT))
{
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
TABLE_LIST *next;
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
while ((next= it++))
cur_table_ref= next;
}
if (cur_table_ref->is_leaf_for_name_resolution())
break;
}
return cur_table_ref;
}
/*
Retrieve the last (right-most) leaf in a nested join tree with
respect to name resolution.
SYNOPSIS
st_table_list::last_leaf_for_name_resolution()
DESCRIPTION
Given that 'this' is a nested table reference, recursively walk
down the right-most children of 'this' until we reach a leaf
table reference with respect to name resolution.
IMPLEMENTATION
The right-most child of a nested table reference is the first
element in the list of children because the children are inserted
in reverse order.
RETURN
- If 'this' is a nested table reference - the right-most child of
the tree rooted in 'this',
- else - 'this'
*/
TABLE_LIST *st_table_list::last_leaf_for_name_resolution()
{
TABLE_LIST *cur_table_ref= this;
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
NESTED_JOIN *cur_nested_join;
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
if (is_leaf_for_name_resolution())
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
return this;
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
DBUG_ASSERT(nested_join);
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
for (cur_nested_join= nested_join;
cur_nested_join;
cur_nested_join= cur_table_ref->nested_join)
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
{
Fix for BUG#13597 - columns in ON condition not resolved if references a table in a nested right join. The problem was in that when finding the last table reference in a nested join tree, the procedure doing the iteration over the right-most branches of a join tree was testing for RIGHT JOINs the table reference that represents the join, and not the second operand of the JOIN. Currently the information whether a join is LEFT/RIGHT is stored not on the join object itself, but on one of its operands.
2005-09-30 09:39:17 +02:00
cur_table_ref= cur_nested_join->join_list.head();
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
/*
Fix for BUG#13597 - columns in ON condition not resolved if references a table in a nested right join. The problem was in that when finding the last table reference in a nested join tree, the procedure doing the iteration over the right-most branches of a join tree was testing for RIGHT JOINs the table reference that represents the join, and not the second operand of the JOIN. Currently the information whether a join is LEFT/RIGHT is stored not on the join object itself, but on one of its operands.
2005-09-30 09:39:17 +02:00
If the current nested is a RIGHT JOIN, the operands in
'join_list' are in reverse order, thus the last operand is in the
end of the list.
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
*/
if ((cur_table_ref->outer_join & JOIN_TYPE_RIGHT))
{
List_iterator_fast<TABLE_LIST> it(cur_nested_join->join_list);
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
TABLE_LIST *next;
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
cur_table_ref= it++;
while ((next= it++))
cur_table_ref= next;
}
if (cur_table_ref->is_leaf_for_name_resolution())
break;
}
return cur_table_ref;
}
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
/*
Register access mode which we need for underlying tables
SYNOPSIS
register_want_access()
want_access Acess which we require
*/
void st_table_list::register_want_access(ulong want_access)
{
/* Remove SHOW_VIEW_ACL, because it will be checked during making view */
want_access&= ~SHOW_VIEW_ACL;
if (belong_to_view)
{
grant.want_privilege= want_access;
if (table)
table->grant.want_privilege= want_access;
}
for (TABLE_LIST *tbl= merge_underlying_list; tbl; tbl= tbl->next_local)
tbl->register_want_access(want_access);
}
/*
fixe of typos, comments & layout
2005-10-27 23:56:44 +02:00
Load security context information for this view
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
SYNOPSIS
st_table_list::prepare_view_securety_context()
thd [in] thread handler
RETURN
FALSE OK
TRUE Error
*/
#ifndef NO_EMBEDDED_ACCESS_CHECKS
bool st_table_list::prepare_view_securety_context(THD *thd)
{
DBUG_ENTER("st_table_list::prepare_view_securety_context");
DBUG_PRINT("enter", ("table: %s", alias));
DBUG_ASSERT(!prelocking_placeholder && view);
if (view_suid)
{
DBUG_PRINT("info", ("This table is suid view => load contest"));
DBUG_ASSERT(view && view_sctx);
if (acl_getroot_no_password(view_sctx,
definer.user.str,
definer.host.str,
definer.host.str,
thd->db))
{
Bug #14875: Bad view DEFINER makes SHOW CREATE VIEW fail When reading a view definition from a .frm file it was throwing a SQL error if the DEFINER user is not defined. Changed it to a warning to match the (documented) case when a view with undefined DEFINER user is created.
2006-05-26 10:49:39 +02:00
if (thd->lex->sql_command == SQLCOM_SHOW_CREATE)
{
push_warning_printf(thd, MYSQL_ERROR::WARN_LEVEL_NOTE,
ER_NO_SUCH_USER,
ER(ER_NO_SUCH_USER),
definer.user.str, definer.host.str);
}
else
{
my_error(ER_NO_SUCH_USER, MYF(0), definer.user.str, definer.host.str);
DBUG_RETURN(TRUE);
}
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
}
}
DBUG_RETURN(FALSE);
}
#endif
/*
Find security context of current view
SYNOPSIS
st_table_list::find_view_security_context()
thd [in] thread handler
*/
#ifndef NO_EMBEDDED_ACCESS_CHECKS
Security_context *st_table_list::find_view_security_context(THD *thd)
{
Security_context *sctx;
TABLE_LIST *upper_view= this;
DBUG_ENTER("st_table_list::find_view_security_context");
DBUG_ASSERT(view);
while (upper_view && !upper_view->view_suid)
{
DBUG_ASSERT(!upper_view->prelocking_placeholder);
upper_view= upper_view->referencing_view;
}
if (upper_view)
{
DBUG_PRINT("info", ("Securety context of view %s will be used",
upper_view->alias));
sctx= upper_view->view_sctx;
DBUG_ASSERT(sctx);
}
else
{
DBUG_PRINT("info", ("Current global context will be used"));
sctx= thd->security_ctx;
}
DBUG_RETURN(sctx);
}
#endif
/*
Prepare security context and load underlying tables priveleges for view
SYNOPSIS
st_table_list::prepare_security()
thd [in] thread handler
RETURN
FALSE OK
TRUE Error
*/
bool st_table_list::prepare_security(THD *thd)
{
List_iterator_fast<TABLE_LIST> tb(*view_tables);
TABLE_LIST *tbl;
Added missing DBUG_ENTER (bug in last push)
2005-11-01 20:59:24 +01:00
DBUG_ENTER("st_table_list::prepare_security");
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
#ifndef NO_EMBEDDED_ACCESS_CHECKS
Security_context *save_security_ctx= thd->security_ctx;
DBUG_ASSERT(!prelocking_placeholder);
if (prepare_view_securety_context(thd))
DBUG_RETURN(TRUE);
thd->security_ctx= find_view_security_context(thd);
while ((tbl= tb++))
{
DBUG_ASSERT(tbl->referencing_view);
char *db, *table_name;
if (tbl->view)
{
db= tbl->view_db.str;
table_name= tbl->view_name.str;
}
else
{
db= tbl->db;
table_name= tbl->table_name;
}
fill_effective_table_privileges(thd, &tbl->grant, db, table_name);
if (tbl->table)
tbl->table->grant= grant;
}
thd->security_ctx= save_security_ctx;
#else
while ((tbl= tb++))
tbl->grant.privilege= ~NO_ACCESS;
#endif
found problem removed
2005-11-02 14:17:57 +01:00
DBUG_RETURN(FALSE);
support of view underlying tables and SP functions security check added (BUG#9505) (WL#2787)
2005-10-27 23:18:23 +02:00
}
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
Natural_join_column::Natural_join_column(Field_translator *field_param,
TABLE_LIST *tab)
{
DBUG_ASSERT(tab->field_translation);
view_field= field_param;
table_field= NULL;
table_ref= tab;
is_common= FALSE;
}
Natural_join_column::Natural_join_column(Field *field_param,
TABLE_LIST *tab)
{
WL#2486 - natural and using join according to SQL:2003 - Corrected problem with N-way nested natural joins in PS mode. - Code cleanup - More asserts to check consistency of name resolution contexts - Fixed potential memory leak of name resolution contexts
2005-08-19 14:22:30 +02:00
DBUG_ASSERT(tab->table == field_param->table);
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
table_field= field_param;
view_field= NULL;
table_ref= tab;
is_common= FALSE;
}
const char *Natural_join_column::name()
{
if (view_field)
{
DBUG_ASSERT(table_field == NULL);
return view_field->name;
}
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
return table_field->field_name;
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
Item *Natural_join_column::create_item(THD *thd)
{
if (view_field)
{
DBUG_ASSERT(table_field == NULL);
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
return create_view_field(thd, table_ref, &view_field->item,
view_field->name);
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
return new Item_field(thd, &thd->lex->current_select->context, table_field);
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
Field *Natural_join_column::field()
{
if (view_field)
{
DBUG_ASSERT(table_field == NULL);
return NULL;
}
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
return table_field;
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
const char *Natural_join_column::table_name()
{
return table_ref->alias;
}
const char *Natural_join_column::db_name()
{
if (view_field)
return table_ref->view_db.str;
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
Fix for BUG#17523: natural join and information schema. The cause of the bug was an ASSERT that checked the consistency of TABLE_SHARE::db and TABLE_LIST::db and failed for I_S tables. The fix relaxes the requirement for consistency for I_S.
2006-02-22 09:04:10 +01:00
/*
Test that TABLE_LIST::db is the same as st_table_share::db to
ensure consistency. An exception are I_S schema tables, which
are inconsistent in this respect.
*/
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
DBUG_ASSERT(!strcmp(table_ref->db,
Merge mysql.com:/extern/mysql/5.0/bug17476/mysql-5.0 into mysql.com:/extern/mysql/5.1/generic/mysql-5.1-new
2006-03-03 12:03:27 +01:00
table_ref->table->s->db.str) ||
Fix for BUG#17523: natural join and information schema. The cause of the bug was an ASSERT that checked the consistency of TABLE_SHARE::db and TABLE_LIST::db and failed for I_S tables. The fix relaxes the requirement for consistency for I_S.
2006-02-22 09:04:10 +01:00
(table_ref->schema_table &&
Merge mysql.com:/extern/mysql/5.0/bug17476/mysql-5.0 into mysql.com:/extern/mysql/5.1/generic/mysql-5.1-new
2006-03-03 12:03:27 +01:00
table_ref->table->s->db.str[0] == 0));
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
return table_ref->db;
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
GRANT_INFO *Natural_join_column::grant()
{
if (view_field)
return &(table_ref->grant);
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
return &(table_ref->table->grant);
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
void Field_iterator_view::set(TABLE_LIST *table)
{
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
DBUG_ASSERT(table->field_translation);
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
view= table;
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
ptr= table->field_translation;
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
array_end= table->field_translation_end;
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
}
const char *Field_iterator_table::name()
{
return (*ptr)->field_name;
}
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
Item *Field_iterator_table::create_item(THD *thd)
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
{
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
return new Item_field(thd, &thd->lex->current_select->context, *ptr);
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
}
const char *Field_iterator_view::name()
{
fixed merged view fields names (BUG#5147) support of merged VIEW over several tables added (WL#1809)
2004-09-14 18:28:29 +02:00
return ptr->name;
VIEW two TABLE_LIST copy eliminated
2004-07-16 00:15:55 +02:00
}
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
Item *Field_iterator_view::create_item(THD *thd)
{
return create_view_field(thd, view, &ptr->item, ptr->name);
}
Item *create_view_field(THD *thd, TABLE_LIST *view, Item **field_ref,
const char *name)
{
bool save_wrapper= thd->lex->select_lex.no_wrap_view_item;
Item *field= *field_ref;
DBUG_ENTER("create_view_field");
if (view->schema_table_reformed)
{
/*
WL#2486 - Natural/using join according to SQL:2003. Post-review fixes according to Monty's review.
2005-11-28 20:57:50 +01:00
Translation table items are always Item_fields and already fixed
('mysql_schema_table' function). So we can return directly the
field. This case happens only for 'show & where' commands.
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
*/
DBUG_ASSERT(field && field->fixed);
DBUG_RETURN(field);
}
DBUG_ASSERT(field);
thd->lex->current_select->no_wrap_view_item= TRUE;
if (!field->fixed)
{
if (field->fix_fields(thd, field_ref))
{
thd->lex->current_select->no_wrap_view_item= save_wrapper;
DBUG_RETURN(0);
}
field= *field_ref;
}
thd->lex->current_select->no_wrap_view_item= save_wrapper;
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 (save_wrapper)
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
{
DBUG_RETURN(field);
}
Item *item= new Item_direct_view_ref(&view->view->select_lex.context,
Fix for BUG#13410 - qualified reference to a view column in the HAVING clause cannot be resolved. The problem was then when a column reference was resolved to a view column, the new Item created for this column contained the name of the view, and not the view alias.
2005-10-01 08:35:30 +02:00
field_ref, view->alias,
Name resolution context added (BUG#6443)
2005-07-01 06:05:42 +02:00
name);
DBUG_RETURN(item);
}
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
void Field_iterator_natural_join::set(TABLE_LIST *table_ref)
{
DBUG_ASSERT(table_ref->join_columns);
WL#2486 - Natural/using join according to SQL:2003. Post-review fixes according to Monty's review.
2005-11-28 20:57:50 +01:00
column_ref_it.init(*(table_ref->join_columns));
cur_column_ref= column_ref_it++;
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
WL#2486 - natural and using join according to SQL:2003 - Corrected problem with N-way nested natural joins in PS mode. - Code cleanup - More asserts to check consistency of name resolution contexts - Fixed potential memory leak of name resolution contexts
2005-08-19 14:22:30 +02:00
void Field_iterator_natural_join::next()
{
WL#2486 - Natural/using join according to SQL:2003. Post-review fixes according to Monty's review.
2005-11-28 20:57:50 +01:00
cur_column_ref= column_ref_it++;
Cleanup during review of new pushed code
2005-08-22 00:13:37 +02:00
DBUG_ASSERT(!cur_column_ref || ! cur_column_ref->table_field ||
cur_column_ref->table_ref->table ==
cur_column_ref->table_field->table);
WL#2486 - natural and using join according to SQL:2003 - Corrected problem with N-way nested natural joins in PS mode. - Code cleanup - More asserts to check consistency of name resolution contexts - Fixed potential memory leak of name resolution contexts
2005-08-19 14:22:30 +02:00
}
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
void Field_iterator_table_ref::set_field_iterator()
{
DBUG_ENTER("Field_iterator_table_ref::set_field_iterator");
/*
If the table reference we are iterating over is a natural join, or it is
an operand of a natural join, and TABLE_LIST::join_columns contains all
the columns of the join operand, then we pick the columns from
TABLE_LIST::join_columns, instead of the orginial container of the
columns of the join operator.
*/
if (table_ref->is_join_columns_complete)
{
/* Necesary, but insufficient conditions. */
DBUG_ASSERT(table_ref->is_natural_join ||
table_ref->nested_join ||
table_ref->join_columns &&
/* This is a merge view. */
((table_ref->field_translation &&
table_ref->join_columns->elements ==
(ulong)(table_ref->field_translation_end -
table_ref->field_translation)) ||
/* This is stored table or a tmptable view. */
(!table_ref->field_translation &&
table_ref->join_columns->elements ==
table_ref->table->s->fields)));
field_it= &natural_join_it;
DBUG_PRINT("info",("field_it for '%s' is Field_iterator_natural_join",
WL#2486 - natural and using join Fixed debug printout.
2005-08-18 10:28:50 +02:00
table_ref->alias));
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
/* This is a merge view, so use field_translation. */
else if (table_ref->field_translation)
{
DBUG_ASSERT(table_ref->view &&
table_ref->effective_algorithm == VIEW_ALGORITHM_MERGE);
field_it= &view_field_it;
DBUG_PRINT("info", ("field_it for '%s' is Field_iterator_view",
WL#2486 - natural and using join Fixed debug printout.
2005-08-18 10:28:50 +02:00
table_ref->alias));
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
/* This is a base table or stored view. */
else
{
DBUG_ASSERT(table_ref->table || table_ref->view);
field_it= &table_field_it;
DBUG_PRINT("info", ("field_it for '%s' is Field_iterator_table",
WL#2486 - natural and using join Fixed debug printout.
2005-08-18 10:28:50 +02:00
table_ref->alias));
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
field_it->set(table_ref);
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
DBUG_VOID_RETURN;
}
void Field_iterator_table_ref::set(TABLE_LIST *table)
{
DBUG_ASSERT(table);
first_leaf= table->first_leaf_for_name_resolution();
last_leaf= table->last_leaf_for_name_resolution();
DBUG_ASSERT(first_leaf && last_leaf);
table_ref= first_leaf;
set_field_iterator();
}
void Field_iterator_table_ref::next()
{
/* Move to the next field in the current table reference. */
field_it->next();
/*
If all fields of the current table reference are exhausted, move to
the next leaf table reference.
*/
if (field_it->end_of_fields() && table_ref != last_leaf)
{
table_ref= table_ref->next_name_resolution_table;
DBUG_ASSERT(table_ref);
set_field_iterator();
}
}
const char *Field_iterator_table_ref::table_name()
{
if (table_ref->view)
return table_ref->view_name.str;
else if (table_ref->is_natural_join)
return natural_join_it.column_ref()->table_name();
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
DBUG_ASSERT(!strcmp(table_ref->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
table_ref->table->s->table_name.str));
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
return table_ref->table_name;
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
const char *Field_iterator_table_ref::db_name()
{
if (table_ref->view)
return table_ref->view_db.str;
else if (table_ref->is_natural_join)
return natural_join_it.column_ref()->db_name();
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
Fix for BUG#17523: natural join and information schema. The cause of the bug was an ASSERT that checked the consistency of TABLE_SHARE::db and TABLE_LIST::db and failed for I_S tables. The fix relaxes the requirement for consistency for I_S.
2006-02-22 09:04:10 +01:00
/*
Test that TABLE_LIST::db is the same as st_table_share::db to
ensure consistency. An exception are I_S schema tables, which
are inconsistent in this respect.
*/
Merge mysql.com:/extern/mysql/5.0/bug17476/mysql-5.0 into mysql.com:/extern/mysql/5.1/generic/mysql-5.1-new
2006-03-03 12:03:27 +01:00
DBUG_ASSERT(!strcmp(table_ref->db, table_ref->table->s->db.str) ||
Fix for BUG#17523: natural join and information schema. The cause of the bug was an ASSERT that checked the consistency of TABLE_SHARE::db and TABLE_LIST::db and failed for I_S tables. The fix relaxes the requirement for consistency for I_S.
2006-02-22 09:04:10 +01:00
(table_ref->schema_table &&
Merge mysql.com:/extern/mysql/5.0/bug17476/mysql-5.0 into mysql.com:/extern/mysql/5.1/generic/mysql-5.1-new
2006-03-03 12:03:27 +01:00
table_ref->table->s->db.str[0] == 0));
Fix for BUG#17523: natural join and information schema. The cause of the bug was an ASSERT that checked the consistency of TABLE_SHARE::db and TABLE_LIST::db and failed for I_S tables. The fix relaxes the requirement for consistency for I_S.
2006-02-22 09:04:10 +01:00
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
return table_ref->db;
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
GRANT_INFO *Field_iterator_table_ref::grant()
{
if (table_ref->view)
return &(table_ref->grant);
else if (table_ref->is_natural_join)
return natural_join_it.column_ref()->grant();
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
return &(table_ref->table->grant);
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
/*
Create new or return existing column reference to a column of a
natural/using join.
SYNOPSIS
Field_iterator_table_ref::get_or_create_column_ref()
Fix for BUG#15229. The cause of this bug was a design flaw due to which the list of natural join columns was incorrectly computed and stored for nested joins that are not natural joins, but are operands (possibly indirect) of nested joins. The patch corrects the flaw in a such a way, that the result columns of a table reference are materialized only if it is a leaf table (that is, only if it is a view, stored table, or natural/using join).
2006-03-02 10:50:15 +01:00
parent_table_ref the parent table reference over which the
iterator is iterating
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
DESCRIPTION
Fix for BUG#15229. The cause of this bug was a design flaw due to which the list of natural join columns was incorrectly computed and stored for nested joins that are not natural joins, but are operands (possibly indirect) of nested joins. The patch corrects the flaw in a such a way, that the result columns of a table reference are materialized only if it is a leaf table (that is, only if it is a view, stored table, or natural/using join).
2006-03-02 10:50:15 +01:00
Create a new natural join column for the current field of the
iterator if no such column was created, or return an already
created natural join column. The former happens for base tables or
views, and the latter for natural/using joins. If a new field is
created, then the field is added to 'parent_table_ref' if it is
given, or to the original table referene of the field if
parent_table_ref == NULL.
NOTES
This method is designed so that when a Field_iterator_table_ref
walks through the fields of a table reference, all its fields
are created and stored as follows:
- If the table reference being iterated is a stored table, view or
natural/using join, store all natural join columns in a list
attached to that table reference.
- If the table reference being iterated is a nested join that is
not natural/using join, then do not materialize its result
fields. This is OK because for such table references
Field_iterator_table_ref iterates over the fields of the nested
table references (recursively). In this way we avoid the storage
of unnecessay copies of result columns of nested joins.
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
RETURN
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
# Pointer to a column of a natural join (or its operand)
NULL No memory to allocate the column
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
*/
Natural_join_column *
Fix for BUG#15229. The cause of this bug was a design flaw due to which the list of natural join columns was incorrectly computed and stored for nested joins that are not natural joins, but are operands (possibly indirect) of nested joins. The patch corrects the flaw in a such a way, that the result columns of a table reference are materialized only if it is a leaf table (that is, only if it is a view, stored table, or natural/using join).
2006-03-02 10:50:15 +01:00
Field_iterator_table_ref::get_or_create_column_ref(TABLE_LIST *parent_table_ref)
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
{
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
Natural_join_column *nj_col;
Fix for BUG#15229. The cause of this bug was a design flaw due to which the list of natural join columns was incorrectly computed and stored for nested joins that are not natural joins, but are operands (possibly indirect) of nested joins. The patch corrects the flaw in a such a way, that the result columns of a table reference are materialized only if it is a leaf table (that is, only if it is a view, stored table, or natural/using join).
2006-03-02 10:50:15 +01:00
bool is_created= TRUE;
uint field_count;
TABLE_LIST *add_table_ref= parent_table_ref ?
parent_table_ref : table_ref;
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +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
LINT_INIT(field_count);
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
if (field_it == &table_field_it)
WL#2486 - natural and using join according to SQL:2003 - Corrected problem with N-way nested natural joins in PS mode. - Code cleanup - More asserts to check consistency of name resolution contexts - Fixed potential memory leak of name resolution contexts
2005-08-19 14:22:30 +02:00
{
/* The field belongs to a stored table. */
Field *field= table_field_it.field();
nj_col= new Natural_join_column(field, table_ref);
Fix for BUG#15229. The cause of this bug was a design flaw due to which the list of natural join columns was incorrectly computed and stored for nested joins that are not natural joins, but are operands (possibly indirect) of nested joins. The patch corrects the flaw in a such a way, that the result columns of a table reference are materialized only if it is a leaf table (that is, only if it is a view, stored table, or natural/using join).
2006-03-02 10:50:15 +01:00
field_count= table_ref->table->s->fields;
WL#2486 - natural and using join according to SQL:2003 - Corrected problem with N-way nested natural joins in PS mode. - Code cleanup - More asserts to check consistency of name resolution contexts - Fixed potential memory leak of name resolution contexts
2005-08-19 14:22:30 +02:00
}
else if (field_it == &view_field_it)
{
/* The field belongs to a merge view or information schema table. */
Field_translator *translated_field= view_field_it.field_translator();
nj_col= new Natural_join_column(translated_field, table_ref);
Fix for BUG#15229. The cause of this bug was a design flaw due to which the list of natural join columns was incorrectly computed and stored for nested joins that are not natural joins, but are operands (possibly indirect) of nested joins. The patch corrects the flaw in a such a way, that the result columns of a table reference are materialized only if it is a leaf table (that is, only if it is a view, stored table, or natural/using join).
2006-03-02 10:50:15 +01:00
field_count= table_ref->field_translation_end -
table_ref->field_translation;
WL#2486 - natural and using join according to SQL:2003 - Corrected problem with N-way nested natural joins in PS mode. - Code cleanup - More asserts to check consistency of name resolution contexts - Fixed potential memory leak of name resolution contexts
2005-08-19 14:22:30 +02:00
}
else
{
/*
The field belongs to a NATURAL join, therefore the column reference was
already created via one of the two constructor calls above. In this case
we just return the already created column reference.
*/
Fix for BUG#15229. The cause of this bug was a design flaw due to which the list of natural join columns was incorrectly computed and stored for nested joins that are not natural joins, but are operands (possibly indirect) of nested joins. The patch corrects the flaw in a such a way, that the result columns of a table reference are materialized only if it is a leaf table (that is, only if it is a view, stored table, or natural/using join).
2006-03-02 10:50:15 +01:00
DBUG_ASSERT(table_ref->is_join_columns_complete);
is_created= FALSE;
WL#2486 - natural and using join according to SQL:2003 - Corrected problem with N-way nested natural joins in PS mode. - Code cleanup - More asserts to check consistency of name resolution contexts - Fixed potential memory leak of name resolution contexts
2005-08-19 14:22:30 +02:00
nj_col= natural_join_it.column_ref();
DBUG_ASSERT(nj_col);
}
Cleanup during review of new pushed code
2005-08-22 00:13:37 +02:00
DBUG_ASSERT(!nj_col->table_field ||
nj_col->table_ref->table == nj_col->table_field->table);
Fix for BUG#15229. The cause of this bug was a design flaw due to which the list of natural join columns was incorrectly computed and stored for nested joins that are not natural joins, but are operands (possibly indirect) of nested joins. The patch corrects the flaw in a such a way, that the result columns of a table reference are materialized only if it is a leaf table (that is, only if it is a view, stored table, or natural/using join).
2006-03-02 10:50:15 +01:00
/*
If the natural join column was just created add it to the list of
natural join columns of either 'parent_table_ref' or to the table
reference that directly contains the original field.
*/
if (is_created)
{
/* Make sure not all columns were materialized. */
DBUG_ASSERT(!add_table_ref->is_join_columns_complete);
if (!add_table_ref->join_columns)
{
/* Create a list of natural join columns on demand. */
if (!(add_table_ref->join_columns= new List<Natural_join_column>))
return NULL;
add_table_ref->is_join_columns_complete= FALSE;
}
add_table_ref->join_columns->push_back(nj_col);
/*
If new fields are added to their original table reference, mark if
all fields were added. We do it here as the caller has no easy way
of knowing when to do it.
If the fields are being added to parent_table_ref, then the caller
must take care to mark when all fields are created/added.
*/
if (!parent_table_ref &&
add_table_ref->join_columns->elements == field_count)
add_table_ref->is_join_columns_complete= TRUE;
}
Cleanups and optimization during review of new code
2005-08-18 02:12:42 +02:00
return nj_col;
Implementation of WL#2486 - "Process NATURAL and USING joins according to SQL:2003". * Some of the main problems fixed by the patch: - in "select *" queries the * expanded correctly according to ANSI for arbitrary natural/using joins - natural/using joins are correctly transformed into JOIN ... ON for any number/nesting of the joins. - column references are correctly resolved against natural joins of any nesting and combined with arbitrary other joins. * This patch also contains a fix for name resolution of items inside the ON condition of JOIN ... ON - in this case items must be resolved only against the JOIN operands. To support such 'local' name resolution, the patch introduces a stack of name resolution contexts used at parse time. NOTICE: - This patch is not complete in the sense that - there are 2 test cases that still do not pass - one in join.test, one in select.test. Both are marked with a comment "TODO: WL#2486". - it does not include a new test specific for the task
2005-08-12 16:57:19 +02:00
}
WL#2486 - Natural/using join according to SQL:2003. Post-review fixes according to Monty's review.
2005-11-28 20:57:50 +01:00
/*
Return an existing reference to a column of a natural/using join.
SYNOPSIS
Field_iterator_table_ref::get_natural_column_ref()
DESCRIPTION
The method should be called in contexts where it is expected that
all natural join columns are already created, and that the column
being retrieved is a Natural_join_column.
RETURN
# Pointer to a column of a natural join (or its operand)
NULL No memory to allocate the column
*/
Natural_join_column *
Field_iterator_table_ref::get_natural_column_ref()
{
Natural_join_column *nj_col;
DBUG_ASSERT(field_it == &natural_join_it);
/*
The field belongs to a NATURAL join, therefore the column reference was
already created via one of the two constructor calls above. In this case
we just return the already created column reference.
*/
nj_col= natural_join_it.column_ref();
DBUG_ASSERT(nj_col &&
(!nj_col->table_field ||
nj_col->table_ref->table == nj_col->table_field->table));
return nj_col;
}
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
/*****************************************************************************
Functions to handle column usage bitmaps (read_set, write_set etc...)
*****************************************************************************/
/* Reset all columns bitmaps */
void st_table::clear_column_bitmaps()
{
/*
Reset column read/write usage. It's identical to:
bitmap_clear_all(&table->def_read_set);
bitmap_clear_all(&table->def_write_set);
*/
bzero((char*) def_read_set.bitmap, s->column_bitmap_size*2);
column_bitmaps_set(&def_read_set, &def_write_set);
}
/*
Tell handler we are going to call position() and rnd_pos() later.
NOTES:
This is needed for handlers that uses the primary key to find the
row. In this case we have to extend the read bitmap with the primary
key fields.
*/
void st_table::prepare_for_position()
{
DBUG_ENTER("st_table::prepare_for_position");
if ((file->ha_table_flags() & HA_PRIMARY_KEY_IN_READ_INDEX) &&
s->primary_key < MAX_KEY)
{
mark_columns_used_by_index_no_reset(s->primary_key, read_set);
/* signal change */
file->column_bitmaps_signal();
}
DBUG_VOID_RETURN;
}
/*
Mark that only fields from one key is used
NOTE:
This changes the bitmap to use the tmp bitmap
After this, you can't access any other columns in the table until
bitmaps are reset, for example with st_table::clear_column_bitmaps()
or st_table::restore_column_maps_after_mark_index()
*/
void st_table::mark_columns_used_by_index(uint index)
{
MY_BITMAP *bitmap= &tmp_set;
DBUG_ENTER("st_table::mark_columns_used_by_index");
(void) file->extra(HA_EXTRA_KEYREAD);
bitmap_clear_all(bitmap);
mark_columns_used_by_index_no_reset(index, bitmap);
column_bitmaps_set(bitmap, bitmap);
DBUG_VOID_RETURN;
}
/*
Restore to use normal column maps after key read
NOTES
This reverse the change done by mark_columns_used_by_index
WARNING
For this to work, one must have the normal table maps in place
when calling mark_columns_used_by_index
*/
void st_table::restore_column_maps_after_mark_index()
{
DBUG_ENTER("st_table::restore_column_maps_after_mark_index");
key_read= 0;
(void) file->extra(HA_EXTRA_NO_KEYREAD);
default_column_bitmaps();
file->column_bitmaps_signal();
DBUG_VOID_RETURN;
}
/*
mark columns used by key, but don't reset other fields
*/
void st_table::mark_columns_used_by_index_no_reset(uint index,
MY_BITMAP *bitmap)
{
KEY_PART_INFO *key_part= key_info[index].key_part;
KEY_PART_INFO *key_part_end= (key_part +
key_info[index].key_parts);
for (;key_part != key_part_end; key_part++)
bitmap_set_bit(bitmap, key_part->fieldnr-1);
}
/*
Mark auto-increment fields as used fields in both read and write maps
NOTES
This is needed in insert & update as the auto-increment field is
always set and sometimes read.
*/
void st_table::mark_auto_increment_column()
{
DBUG_ASSERT(found_next_number_field);
/*
We must set bit in read set as update_auto_increment() is using the
store() to check overflow of auto_increment values
*/
bitmap_set_bit(read_set, found_next_number_field->field_index);
bitmap_set_bit(write_set, found_next_number_field->field_index);
if (s->next_number_key_offset)
mark_columns_used_by_index_no_reset(s->next_number_index, read_set);
file->column_bitmaps_signal();
}
/*
Mark columns needed for doing an delete of a row
DESCRIPTON
Some table engines don't have a cursor on the retrieve rows
so they need either to use the primary key or all columns to
be able to delete a row.
If the engine needs this, the function works as follows:
- If primary key exits, mark the primary key columns to be read.
- If not, mark all columns to be read
If the engine has HA_REQUIRES_KEY_COLUMNS_FOR_DELETE, we will
mark all key columns as 'to-be-read'. This allows the engine to
loop over the given record to find all keys and doesn't have to
retrieve the row again.
*/
void st_table::mark_columns_needed_for_delete()
{
if (triggers)
After merge fixes for patch solving bug#18437 "Wrong values inserted with a before update trigger on NDB table". Two main changes: - We use TABLE::read_set/write_set bitmaps for marking fields used by statement instead of Field::query_id in 5.1. - Now when we mark columns used by statement we take into account columns used by table's triggers instead of marking all columns as used if table has triggers.
2006-07-06 11:33:23 +02:00
triggers->mark_fields_used(TRG_EVENT_DELETE);
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_REQUIRES_KEY_COLUMNS_FOR_DELETE)
{
Field **reg_field;
for (reg_field= field ; *reg_field ; reg_field++)
{
if ((*reg_field)->flags & PART_KEY_FLAG)
bitmap_set_bit(read_set, (*reg_field)->field_index);
}
file->column_bitmaps_signal();
}
if (file->ha_table_flags() & HA_PRIMARY_KEY_REQUIRED_FOR_DELETE)
{
/*
If the handler has no cursor capabilites, we have to read either
the primary key, the hidden primary key or all columns to be
able to do an delete
*/
if (s->primary_key == MAX_KEY)
file->use_hidden_primary_key();
else
{
mark_columns_used_by_index_no_reset(s->primary_key, read_set);
file->column_bitmaps_signal();
}
}
}
/*
Mark columns needed for doing an update of a row
DESCRIPTON
Some engines needs to have all columns in an update (to be able to
build a complete row). If this is the case, we mark all not
updated columns to be read.
If this is no the case, we do like in the delete case and mark
if neeed, either the primary key column or all columns to be read.
(see mark_columns_needed_for_delete() for details)
If the engine has HA_REQUIRES_KEY_COLUMNS_FOR_DELETE, we will
mark all USED key columns as 'to-be-read'. This allows the engine to
loop over the given record to find all changed keys and doesn't have to
retrieve the row again.
*/
void st_table::mark_columns_needed_for_update()
{
DBUG_ENTER("mark_columns_needed_for_update");
if (triggers)
After merge fixes for patch solving bug#18437 "Wrong values inserted with a before update trigger on NDB table". Two main changes: - We use TABLE::read_set/write_set bitmaps for marking fields used by statement instead of Field::query_id in 5.1. - Now when we mark columns used by statement we take into account columns used by table's triggers instead of marking all columns as used if table has triggers.
2006-07-06 11:33:23 +02:00
triggers->mark_fields_used(TRG_EVENT_UPDATE);
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_REQUIRES_KEY_COLUMNS_FOR_DELETE)
{
/* Mark all used key columns for read */
Field **reg_field;
for (reg_field= field ; *reg_field ; reg_field++)
{
/* Merge keys is all keys that had a column refered to in the query */
if (merge_keys.is_overlapping((*reg_field)->part_of_key))
bitmap_set_bit(read_set, (*reg_field)->field_index);
}
file->column_bitmaps_signal();
}
if (file->ha_table_flags() & HA_PRIMARY_KEY_REQUIRED_FOR_DELETE)
{
/*
If the handler has no cursor capabilites, we have to read either
the primary key, the hidden primary key or all columns to be
able to do an update
*/
if (s->primary_key == MAX_KEY)
file->use_hidden_primary_key();
else
{
mark_columns_used_by_index_no_reset(s->primary_key, read_set);
file->column_bitmaps_signal();
}
}
DBUG_VOID_RETURN;
}
/*
Mark columns the handler needs for doing an insert
For now, this is used to mark fields used by the trigger
as changed.
*/
void st_table::mark_columns_needed_for_insert()
{
if (triggers)
{
After merge fixes for patch solving bug#18437 "Wrong values inserted with a before update trigger on NDB table". Two main changes: - We use TABLE::read_set/write_set bitmaps for marking fields used by statement instead of Field::query_id in 5.1. - Now when we mark columns used by statement we take into account columns used by table's triggers instead of marking all columns as used if table has triggers.
2006-07-06 11:33:23 +02:00
/*
We don't need to mark columns which are used by ON DELETE and
ON UPDATE triggers, which may be invoked in case of REPLACE or
INSERT ... ON DUPLICATE KEY UPDATE, since before doing actual
row replacement or update write_record() will mark all table
fields as used.
*/
triggers->mark_fields_used(TRG_EVENT_INSERT);
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 (found_next_number_field)
mark_auto_increment_column();
}
A fix and a test case for Bug#19399 "res 'Lost Connection' when dropping/creating tables". The bug could lead to a crash when multi-delete statements were prepared and used with temporary tables. The bug was caused by lack of clean-up of multi-delete tables before re-execution of a prepared statement. In a statement like DELETE t1 FROM t1, t2 WHERE ... the first table list (t1) is moved to lex->auxilliary_table_list and excluded from lex->query_tables or select_lex->tables. Thus it was unaccessible to reinit_stmt_before_use and not cleaned up before re-execution of a prepared statement.
2006-07-06 21:59:04 +02:00
/*
Cleanup this table for re-execution.
SYNOPSIS
st_table_list::reinit_before_use()
*/
Post-merge fixes for Bug#19399 "Stored Procedures 'Lost Connection' when dropping/creating tables"
2006-07-11 21:39:51 +02:00
void st_table_list::reinit_before_use(THD *thd)
A fix and a test case for Bug#19399 "res 'Lost Connection' when dropping/creating tables". The bug could lead to a crash when multi-delete statements were prepared and used with temporary tables. The bug was caused by lack of clean-up of multi-delete tables before re-execution of a prepared statement. In a statement like DELETE t1 FROM t1, t2 WHERE ... the first table list (t1) is moved to lex->auxilliary_table_list and excluded from lex->query_tables or select_lex->tables. Thus it was unaccessible to reinit_stmt_before_use and not cleaned up before re-execution of a prepared statement.
2006-07-06 21:59:04 +02:00
{
/*
Reset old pointers to TABLEs: they are not valid since the tables
were closed in the end of previous prepare or execute call.
*/
table= 0;
Post-merge fixes for Bug#19399 "Stored Procedures 'Lost Connection' when dropping/creating tables"
2006-07-11 21:39:51 +02:00
/* Reset is_schema_table_processed value(needed for I_S tables */
is_schema_table_processed= FALSE;
TABLE_LIST *embedded; /* The table at the current level of nesting. */
TABLE_LIST *embedding= this; /* The parent nested table reference. */
do
{
embedded= embedding;
if (embedded->prep_on_expr)
embedded->on_expr= embedded->prep_on_expr->copy_andor_structure(thd);
embedding= embedded->embedding;
}
while (embedding &&
embedding->nested_join->join_list.head() == embedded);
A fix and a test case for Bug#19399 "res 'Lost Connection' when dropping/creating tables". The bug could lead to a crash when multi-delete statements were prepared and used with temporary tables. The bug was caused by lack of clean-up of multi-delete tables before re-execution of a prepared statement. In a statement like DELETE t1 FROM t1, t2 WHERE ... the first table list (t1) is moved to lex->auxilliary_table_list and excluded from lex->query_tables or select_lex->tables. Thus it was unaccessible to reinit_stmt_before_use and not cleaned up before re-execution of a prepared statement.
2006-07-06 21:59:04 +02:00
}
WL#2486 - Natural/using join according to SQL:2003. Post-review fixes according to Monty's review.
2005-11-28 20:57:50 +01:00
Import changeset
2000-07-31 21:29:14 +02:00
/*****************************************************************************
** Instansiate templates
*****************************************************************************/
WL#2286 - Compile MySQL w/YASSL support Fix GCC 4.0 link failure. Better CXX_VERSION guessing.
2005-06-22 11:08:28 +02:00
#ifdef HAVE_EXPLICIT_TEMPLATE_INSTANTIATION
Import changeset
2000-07-31 21:29:14 +02:00
template class List<String>;
template class List_iterator<String>;
#endif
Reference in a new issue Copy permalink