2004-12-11 12:03:51 -08:00
|
|
|
/* Copyright (C) 2004 MySQL AB
|
|
|
|
|
|
|
|
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
|
2006-12-23 20:17:15 +01:00
|
|
|
the Free Software Foundation; version 2 of the License.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
This program is distributed in the hope that it will be useful,
|
|
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
GNU General Public License for more details.
|
|
|
|
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
|
|
along with this program; if not, write to the Free Software
|
|
|
|
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
|
|
|
MySQL Federated Storage Engine
|
|
|
|
|
|
|
|
ha_federated.cc - MySQL Federated Storage Engine
|
|
|
|
Patrick Galbraith and Brian Aker, 2004
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
This is a handler which uses a foreign database as the data file, as
|
2004-12-11 12:03:51 -08:00
|
|
|
opposed to a handler like MyISAM, which uses .MYD files locally.
|
|
|
|
|
|
|
|
How this handler works
|
|
|
|
----------------------------------
|
2005-02-23 00:38:29 -08:00
|
|
|
Normal database files are local and as such: You create a table called
|
|
|
|
'users', a file such as 'users.MYD' is created. A handler reads, inserts,
|
2004-12-11 12:03:51 -08:00
|
|
|
deletes, updates data in this file. The data is stored in particular format,
|
|
|
|
so to read, that data has to be parsed into fields, to write, fields have to
|
2005-02-23 00:38:29 -08:00
|
|
|
be stored in this format to write to this data file.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
With MySQL Federated storage engine, there will be no local files
|
|
|
|
for each table's data (such as .MYD). A foreign database will store
|
|
|
|
the data that would normally be in this file. This will necessitate
|
|
|
|
the use of MySQL client API to read, delete, update, insert this
|
|
|
|
data. The data will have to be retrieve via an SQL call "SELECT *
|
|
|
|
FROM users". Then, to read this data, it will have to be retrieved
|
|
|
|
via mysql_fetch_row one row at a time, then converted from the
|
|
|
|
column in this select into the format that the handler expects.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
The create table will simply create the .frm file, and within the
|
|
|
|
"CREATE TABLE" SQL, there SHALL be any of the following :
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
connection=scheme://username:password@hostname:port/database/tablename
|
|
|
|
connection=scheme://username@hostname/database/tablename
|
|
|
|
connection=scheme://username:password@hostname/database/tablename
|
|
|
|
connection=scheme://username:password@hostname/database/tablename
|
|
|
|
|
|
|
|
- OR -
|
|
|
|
|
|
|
|
As of 5.1 (See worklog #3031), federated now allows you to use a non-url
|
|
|
|
format, taking advantage of mysql.servers:
|
|
|
|
|
|
|
|
connection="connection_one"
|
|
|
|
connection="connection_one/table_foo"
|
2004-12-16 19:08:32 -08:00
|
|
|
|
|
|
|
An example would be:
|
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
connection=mysql://username:password@hostname:port/database/tablename
|
2004-12-16 19:08:32 -08:00
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
or, if we had:
|
|
|
|
|
|
|
|
create server 'server_one' foreign data wrapper 'mysql' options
|
|
|
|
(HOST '127.0.0.1',
|
|
|
|
DATABASE 'db1',
|
|
|
|
USER 'root',
|
|
|
|
PASSWORD '',
|
|
|
|
PORT 3306,
|
|
|
|
SOCKET '',
|
|
|
|
OWNER 'root');
|
|
|
|
|
|
|
|
CREATE TABLE federated.t1 (
|
|
|
|
`id` int(20) NOT NULL,
|
|
|
|
`name` varchar(64) NOT NULL default ''
|
|
|
|
)
|
|
|
|
ENGINE="FEDERATED" DEFAULT CHARSET=latin1
|
|
|
|
CONNECTION='server_one';
|
|
|
|
|
|
|
|
So, this will have been the equivalent of
|
2004-12-16 19:08:32 -08:00
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
CONNECTION="mysql://root@127.0.0.1:3306/db1/t1"
|
2004-12-16 19:08:32 -08:00
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
Then, we can also change the server to point to a new schema:
|
2004-12-16 19:08:32 -08:00
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
ALTER SERVER 'server_one' options(DATABASE 'db2');
|
|
|
|
|
|
|
|
All subsequent calls will now be against db2.t1! Guess what? You don't
|
|
|
|
have to perform an alter table!
|
|
|
|
|
|
|
|
This connecton="connection string" is necessary for the handler to be
|
|
|
|
able to connect to the foreign server, either by URL, or by server
|
|
|
|
name.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
|
|
|
|
The basic flow is this:
|
|
|
|
|
|
|
|
SQL calls issues locally ->
|
2005-02-23 00:38:29 -08:00
|
|
|
mysql handler API (data in handler format) ->
|
|
|
|
mysql client API (data converted to SQL calls) ->
|
|
|
|
foreign database -> mysql client API ->
|
2004-12-11 12:03:51 -08:00
|
|
|
convert result sets (if any) to handler format ->
|
|
|
|
handler API -> results or rows affected to local
|
|
|
|
|
|
|
|
What this handler does and doesn't support
|
|
|
|
------------------------------------------
|
2005-02-23 00:38:29 -08:00
|
|
|
* Tables MUST be created on the foreign server prior to any action on those
|
|
|
|
tables via the handler, first version. IMPORTANT: IF you MUST use the
|
2004-12-16 19:08:32 -08:00
|
|
|
federated storage engine type on the REMOTE end, MAKE SURE [ :) ] That
|
2005-02-23 00:38:29 -08:00
|
|
|
the table you connect to IS NOT a table pointing BACK to your ORIGNAL
|
2004-12-16 19:08:32 -08:00
|
|
|
table! You know and have heard the screaching of audio feedback? You
|
2005-02-23 00:38:29 -08:00
|
|
|
know putting two mirror in front of each other how the reflection
|
2004-12-16 19:08:32 -08:00
|
|
|
continues for eternity? Well, need I say more?!
|
|
|
|
* There will not be support for transactions.
|
2005-02-23 00:38:29 -08:00
|
|
|
* There is no way for the handler to know if the foreign database or table
|
2004-12-11 12:03:51 -08:00
|
|
|
has changed. The reason for this is that this database has to work like a
|
2005-02-23 00:38:29 -08:00
|
|
|
data file that would never be written to by anything other than the
|
|
|
|
database. The integrity of the data in the local table could be breached
|
|
|
|
if there was any change to the foreign database.
|
2004-12-16 19:08:32 -08:00
|
|
|
* Support for SELECT, INSERT, UPDATE , DELETE, indexes.
|
2004-12-11 12:03:51 -08:00
|
|
|
* No ALTER TABLE, DROP TABLE or any other Data Definition Language calls.
|
2005-02-23 00:38:29 -08:00
|
|
|
* Prepared statements will not be used in the first implementation, it
|
2004-12-11 12:03:51 -08:00
|
|
|
remains to to be seen whether the limited subset of the client API for the
|
2004-12-16 19:08:32 -08:00
|
|
|
server supports this.
|
2005-02-23 00:38:29 -08:00
|
|
|
* This uses SELECT, INSERT, UPDATE, DELETE and not HANDLER for its
|
|
|
|
implementation.
|
2004-12-16 19:08:32 -08:00
|
|
|
* This will not work with the query cache.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
Method calls
|
|
|
|
|
|
|
|
A two column table, with one record:
|
|
|
|
|
|
|
|
(SELECT)
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
"SELECT * FROM foo"
|
2004-12-11 12:03:51 -08:00
|
|
|
ha_federated::info
|
|
|
|
ha_federated::scan_time:
|
|
|
|
ha_federated::rnd_init: share->select_query SELECT * FROM foo
|
|
|
|
ha_federated::extra
|
|
|
|
|
|
|
|
<for every row of data retrieved>
|
|
|
|
ha_federated::rnd_next
|
|
|
|
ha_federated::convert_row_to_internal_format
|
|
|
|
ha_federated::rnd_next
|
|
|
|
</for every row of data retrieved>
|
|
|
|
|
|
|
|
ha_federated::rnd_end
|
|
|
|
ha_federated::extra
|
|
|
|
ha_federated::reset
|
|
|
|
|
|
|
|
(INSERT)
|
|
|
|
|
|
|
|
"INSERT INTO foo (id, ts) VALUES (2, now());"
|
|
|
|
|
|
|
|
ha_federated::write_row
|
|
|
|
|
|
|
|
ha_federated::reset
|
|
|
|
|
|
|
|
(UPDATE)
|
|
|
|
|
|
|
|
"UPDATE foo SET ts = now() WHERE id = 1;"
|
|
|
|
|
|
|
|
ha_federated::index_init
|
|
|
|
ha_federated::index_read
|
|
|
|
ha_federated::index_read_idx
|
|
|
|
ha_federated::rnd_next
|
|
|
|
ha_federated::convert_row_to_internal_format
|
|
|
|
ha_federated::update_row
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
ha_federated::extra
|
|
|
|
ha_federated::extra
|
|
|
|
ha_federated::extra
|
|
|
|
ha_federated::external_lock
|
|
|
|
ha_federated::reset
|
|
|
|
|
|
|
|
|
|
|
|
How do I use this handler?
|
|
|
|
--------------------------
|
|
|
|
First of all, you need to build this storage engine:
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
./configure --with-federated-storage-engine
|
|
|
|
make
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
Next, to use this handler, it's very simple. You must
|
2004-12-11 12:03:51 -08:00
|
|
|
have two databases running, either both on the same host, or
|
|
|
|
on different hosts.
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
One the server that will be connecting to the foreign
|
2004-12-11 12:03:51 -08:00
|
|
|
host (client), you create your table as such:
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
CREATE TABLE test_table (
|
2004-12-11 12:03:51 -08:00
|
|
|
id int(20) NOT NULL auto_increment,
|
|
|
|
name varchar(32) NOT NULL default '',
|
|
|
|
other int(20) NOT NULL default '0',
|
|
|
|
PRIMARY KEY (id),
|
|
|
|
KEY name (name),
|
2005-02-23 00:38:29 -08:00
|
|
|
KEY other_key (other))
|
|
|
|
ENGINE="FEDERATED"
|
|
|
|
DEFAULT CHARSET=latin1
|
2007-03-24 01:18:19 -07:00
|
|
|
CONNECTION='mysql://root@127.0.0.1:9306/federated/test_federated';
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
Notice the "COMMENT" and "ENGINE" field? This is where you
|
|
|
|
respectively set the engine type, "FEDERATED" and foreign
|
|
|
|
host information, this being the database your 'client' database
|
|
|
|
will connect to and use as the "data file". Obviously, the foreign
|
|
|
|
database is running on port 9306, so you want to start up your other
|
|
|
|
database so that it is indeed on port 9306, and your federated
|
|
|
|
database on a port other than that. In my setup, I use port 5554
|
|
|
|
for federated, and port 5555 for the foreign database.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
Then, on the foreign database:
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
CREATE TABLE test_table (
|
2004-12-11 12:03:51 -08:00
|
|
|
id int(20) NOT NULL auto_increment,
|
|
|
|
name varchar(32) NOT NULL default '',
|
|
|
|
other int(20) NOT NULL default '0',
|
|
|
|
PRIMARY KEY (id),
|
|
|
|
KEY name (name),
|
2005-02-23 00:38:29 -08:00
|
|
|
KEY other_key (other))
|
2004-12-11 12:03:51 -08:00
|
|
|
ENGINE="<NAME>" <-- whatever you want, or not specify
|
|
|
|
DEFAULT CHARSET=latin1 ;
|
|
|
|
|
|
|
|
This table is exactly the same (and must be exactly the same),
|
2005-02-23 00:38:29 -08:00
|
|
|
except that it is not using the federated handler and does
|
2004-12-11 12:03:51 -08:00
|
|
|
not need the URL.
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
How to see the handler in action
|
|
|
|
--------------------------------
|
|
|
|
|
|
|
|
When developing this handler, I compiled the federated database with
|
|
|
|
debugging:
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
./configure --with-federated-storage-engine
|
2004-12-11 12:03:51 -08:00
|
|
|
--prefix=/home/mysql/mysql-build/federated/ --with-debug
|
|
|
|
|
|
|
|
Once compiled, I did a 'make install' (not for the purpose of installing
|
|
|
|
the binary, but to install all the files the binary expects to see in the
|
2005-02-23 00:38:29 -08:00
|
|
|
diretory I specified in the build with --prefix,
|
|
|
|
"/home/mysql/mysql-build/federated".
|
|
|
|
|
|
|
|
Then, I started the foreign server:
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
/usr/local/mysql/bin/mysqld_safe
|
2004-12-11 12:03:51 -08:00
|
|
|
--user=mysql --log=/tmp/mysqld.5555.log -P 5555
|
|
|
|
|
|
|
|
Then, I went back to the directory containing the newly compiled mysqld,
|
|
|
|
<builddir>/sql/, started up gdb:
|
|
|
|
|
|
|
|
gdb ./mysqld
|
|
|
|
|
|
|
|
Then, withn the (gdb) prompt:
|
|
|
|
(gdb) run --gdb --port=5554 --socket=/tmp/mysqld.5554 --skip-innodb --debug
|
|
|
|
|
|
|
|
Next, I open several windows for each:
|
|
|
|
|
|
|
|
1. Tail the debug trace: tail -f /tmp/mysqld.trace|grep ha_fed
|
2005-02-23 00:38:29 -08:00
|
|
|
2. Tail the SQL calls to the foreign database: tail -f /tmp/mysqld.5555.log
|
2004-12-11 12:03:51 -08:00
|
|
|
3. A window with a client open to the federated server on port 5554
|
|
|
|
4. A window with a client open to the federated server on port 5555
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
I would create a table on the client to the foreign server on port
|
2004-12-11 12:03:51 -08:00
|
|
|
5555, and then to the federated server on port 5554. At this point,
|
|
|
|
I would run whatever queries I wanted to on the federated server,
|
2005-02-23 00:38:29 -08:00
|
|
|
just always remembering that whatever changes I wanted to make on
|
2004-12-11 12:03:51 -08:00
|
|
|
the table, or if I created new tables, that I would have to do that
|
2005-02-23 00:38:29 -08:00
|
|
|
on the foreign server.
|
|
|
|
|
|
|
|
Another thing to look for is 'show variables' to show you that you have
|
2004-12-11 12:03:51 -08:00
|
|
|
support for federated handler support:
|
|
|
|
|
|
|
|
show variables like '%federat%'
|
|
|
|
|
|
|
|
and:
|
|
|
|
|
|
|
|
show storage engines;
|
|
|
|
|
|
|
|
Both should display the federated storage handler.
|
|
|
|
|
|
|
|
|
|
|
|
Testing
|
|
|
|
-------
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
There is a test for MySQL Federated Storage Handler in ./mysql-test/t,
|
2004-12-11 12:03:51 -08:00
|
|
|
federatedd.test It starts both a slave and master database using
|
|
|
|
the same setup that the replication tests use, with the exception that
|
|
|
|
it turns off replication, and sets replication to ignore the test tables.
|
2005-02-23 00:38:29 -08:00
|
|
|
After ensuring that you actually do have support for the federated storage
|
|
|
|
handler, numerous queries/inserts/updates/deletes are run, many derived
|
2004-12-11 12:03:51 -08:00
|
|
|
from the MyISAM tests, plus som other tests which were meant to reveal
|
|
|
|
any issues that would be most likely to affect this handler. All tests
|
|
|
|
should work! ;)
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
To run these tests, go into ./mysql-test (based in the directory you
|
2004-12-11 12:03:51 -08:00
|
|
|
built the server in)
|
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
./mysql-test-run federated
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
To run the test, or if you want to run the test and have debug info:
|
|
|
|
|
|
|
|
./mysql-test-run --debug federated
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
This will run the test in debug mode, and you can view the trace and
|
2004-12-11 12:03:51 -08:00
|
|
|
log files in the ./mysql-test/var/log directory
|
|
|
|
|
|
|
|
ls -l mysql-test/var/log/
|
|
|
|
-rw-r--r-- 1 patg patg 17 4 Dec 12:27 current_test
|
|
|
|
-rw-r--r-- 1 patg patg 692 4 Dec 12:52 manager.log
|
|
|
|
-rw-rw---- 1 patg patg 21246 4 Dec 12:51 master-bin.000001
|
|
|
|
-rw-rw---- 1 patg patg 68 4 Dec 12:28 master-bin.index
|
|
|
|
-rw-r--r-- 1 patg patg 1620 4 Dec 12:51 master.err
|
|
|
|
-rw-rw---- 1 patg patg 23179 4 Dec 12:51 master.log
|
|
|
|
-rw-rw---- 1 patg patg 16696550 4 Dec 12:51 master.trace
|
|
|
|
-rw-r--r-- 1 patg patg 0 4 Dec 12:28 mysqltest-time
|
|
|
|
-rw-r--r-- 1 patg patg 2024051 4 Dec 12:51 mysqltest.trace
|
|
|
|
-rw-rw---- 1 patg patg 94992 4 Dec 12:51 slave-bin.000001
|
|
|
|
-rw-rw---- 1 patg patg 67 4 Dec 12:28 slave-bin.index
|
|
|
|
-rw-rw---- 1 patg patg 249 4 Dec 12:52 slave-relay-bin.000003
|
|
|
|
-rw-rw---- 1 patg patg 73 4 Dec 12:28 slave-relay-bin.index
|
|
|
|
-rw-r--r-- 1 patg patg 1349 4 Dec 12:51 slave.err
|
|
|
|
-rw-rw---- 1 patg patg 96206 4 Dec 12:52 slave.log
|
|
|
|
-rw-rw---- 1 patg patg 15706355 4 Dec 12:51 slave.trace
|
|
|
|
-rw-r--r-- 1 patg patg 0 4 Dec 12:51 warnings
|
|
|
|
|
|
|
|
Of course, again, you can tail the trace log:
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
tail -f mysql-test/var/log/master.trace |grep ha_fed
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
As well as the slave query log:
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
tail -f mysql-test/var/log/slave.log
|
|
|
|
|
|
|
|
Files that comprise the test suit
|
|
|
|
---------------------------------
|
|
|
|
mysql-test/t/federated.test
|
|
|
|
mysql-test/r/federated.result
|
|
|
|
mysql-test/r/have_federated_db.require
|
|
|
|
mysql-test/include/have_federated_db.inc
|
|
|
|
|
|
|
|
|
|
|
|
Other tidbits
|
|
|
|
-------------
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
These were the files that were modified or created for this
|
2007-03-24 01:18:19 -07:00
|
|
|
Federated handler to work, in 5.0:
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
./configure.in
|
|
|
|
./sql/Makefile.am
|
|
|
|
./config/ac_macros/ha_federated.m4
|
|
|
|
./sql/handler.cc
|
|
|
|
./sql/mysqld.cc
|
|
|
|
./sql/set_var.cc
|
|
|
|
./sql/field.h
|
|
|
|
./sql/sql_string.h
|
|
|
|
./mysql-test/mysql-test-run(.sh)
|
|
|
|
./mysql-test/t/federated.test
|
|
|
|
./mysql-test/r/federated.result
|
|
|
|
./mysql-test/r/have_federated_db.require
|
|
|
|
./mysql-test/include/have_federated_db.inc
|
|
|
|
./sql/ha_federated.cc
|
|
|
|
./sql/ha_federated.h
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
In 5.1
|
|
|
|
|
|
|
|
my:~/mysql-build/mysql-5.1-bkbits patg$ ls storage/federated/
|
|
|
|
CMakeLists.txt Makefile.in ha_federated.h plug.in
|
|
|
|
Makefile SCCS libfederated.a
|
|
|
|
Makefile.am ha_federated.cc libfederated_a-ha_federated.o
|
|
|
|
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
*/
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2006-08-09 17:41:35 -07:00
|
|
|
|
2006-08-18 21:19:19 -07:00
|
|
|
#define MYSQL_SERVER 1
|
2005-06-05 16:20:22 +02:00
|
|
|
#include "mysql_priv.h"
|
2006-08-18 21:19:19 -07:00
|
|
|
#include <mysql/plugin.h>
|
|
|
|
|
2005-06-05 16:20:22 +02:00
|
|
|
#ifdef USE_PRAGMA_IMPLEMENTATION
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
#pragma implementation // gcc: Class implementation
|
2004-12-11 12:03:51 -08:00
|
|
|
#endif
|
|
|
|
|
|
|
|
#include "ha_federated.h"
|
2005-07-15 17:33:47 -07:00
|
|
|
|
|
|
|
#include "m_string.h"
|
2006-04-13 13:49:29 -07:00
|
|
|
|
|
|
|
#include <mysql/plugin.h>
|
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
/* Variables for federated share methods */
|
2005-11-23 22:45:02 +02:00
|
|
|
static HASH federated_open_tables; // To track open tables
|
|
|
|
pthread_mutex_t federated_mutex; // To init the hash
|
2007-06-28 00:23:14 -07:00
|
|
|
static char ident_quote_char= '`'; // Character for quoting
|
|
|
|
// identifiers
|
|
|
|
static char value_quote_char= '\''; // Character for quoting
|
|
|
|
// literals
|
2007-06-28 16:03:01 -07:00
|
|
|
static const int bulk_padding= 64; // bytes "overhead" in packet
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2006-08-02 12:36:01 +02:00
|
|
|
/* Variables used when chopping off trailing characters */
|
|
|
|
static const uint sizeof_trailing_comma= sizeof(", ") - 1;
|
|
|
|
static const uint sizeof_trailing_closeparen= sizeof(") ") - 1;
|
|
|
|
static const uint sizeof_trailing_and= sizeof(" AND ") - 1;
|
|
|
|
static const uint sizeof_trailing_where= sizeof(" WHERE ") - 1;
|
|
|
|
|
2005-11-07 16:25:06 +01:00
|
|
|
/* Static declaration for handerton */
|
2006-09-29 17:19:02 -07:00
|
|
|
static handler *federated_create_handler(handlerton *hton,
|
|
|
|
TABLE_SHARE *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 18:52:22 +03:00
|
|
|
MEM_ROOT *mem_root);
|
2006-09-30 12:49:46 -07:00
|
|
|
static int federated_commit(handlerton *hton, THD *thd, bool all);
|
|
|
|
static int federated_rollback(handlerton *hton, THD *thd, bool all);
|
2005-11-07 16:25:06 +01:00
|
|
|
|
2006-09-30 12:49:46 -07:00
|
|
|
/* Federated storage engine handlerton */
|
2005-10-02 19:44:28 -07:00
|
|
|
|
2006-09-29 17:19:02 -07:00
|
|
|
static handler *federated_create_handler(handlerton *hton,
|
|
|
|
TABLE_SHARE *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 18:52:22 +03:00
|
|
|
MEM_ROOT *mem_root)
|
2005-11-07 16:25:06 +01:00
|
|
|
{
|
2006-09-29 17:19:02 -07:00
|
|
|
return new (mem_root) ha_federated(hton, table);
|
2005-11-07 16:25:06 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-11-23 22:45:02 +02:00
|
|
|
/* Function we use in the creation of our hash to get key */
|
2005-02-23 00:38:29 -08:00
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
static uchar *federated_get_key(FEDERATED_SHARE *share, size_t *length,
|
2007-06-30 13:17:49 +02:00
|
|
|
my_bool not_used __attribute__ ((unused)))
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2006-12-01 19:47:45 -05:00
|
|
|
*length= share->share_key_length;
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
return (uchar*) share->share_key;
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
2005-02-16 23:07:10 -08:00
|
|
|
/*
|
|
|
|
Initialize the federated handler.
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
federated_db_init()
|
2007-01-29 01:47:35 +02:00
|
|
|
p Handlerton
|
2005-02-16 23:07:10 -08:00
|
|
|
|
|
|
|
RETURN
|
|
|
|
FALSE OK
|
|
|
|
TRUE Error
|
|
|
|
*/
|
|
|
|
|
2006-09-15 10:28:00 -07:00
|
|
|
int federated_db_init(void *p)
|
2005-02-16 23:07:10 -08:00
|
|
|
{
|
2005-10-02 19:44:28 -07:00
|
|
|
DBUG_ENTER("federated_db_init");
|
2006-09-30 12:49:46 -07:00
|
|
|
handlerton *federated_hton= (handlerton *)p;
|
2006-09-15 10:28:00 -07:00
|
|
|
federated_hton->state= SHOW_OPTION_YES;
|
|
|
|
federated_hton->db_type= DB_TYPE_FEDERATED_DB;
|
|
|
|
federated_hton->commit= federated_commit;
|
|
|
|
federated_hton->rollback= federated_rollback;
|
|
|
|
federated_hton->create= federated_create_handler;
|
2006-12-19 16:37:50 +04:00
|
|
|
federated_hton->flags= HTON_ALTER_NOT_SUPPORTED | HTON_NO_PARTITION;
|
2006-05-28 14:51:01 +02:00
|
|
|
|
2007-07-20 11:35:19 -07:00
|
|
|
/*
|
|
|
|
Support for transactions disabled until WL#2952 fixes it.
|
|
|
|
We do it like this to avoid "defined but not used" compiler warnings.
|
|
|
|
*/
|
|
|
|
federated_hton->commit= 0;
|
|
|
|
federated_hton->rollback= 0;
|
|
|
|
|
2005-10-02 19:44:28 -07:00
|
|
|
if (pthread_mutex_init(&federated_mutex, MY_MUTEX_INIT_FAST))
|
|
|
|
goto error;
|
2006-07-07 11:17:50 -07:00
|
|
|
if (!hash_init(&federated_open_tables, &my_charset_bin, 32, 0, 0,
|
2005-10-02 19:44:28 -07:00
|
|
|
(hash_get_key) federated_get_key, 0, 0))
|
|
|
|
{
|
|
|
|
DBUG_RETURN(FALSE);
|
|
|
|
}
|
2005-11-23 22:45:02 +02:00
|
|
|
|
|
|
|
VOID(pthread_mutex_destroy(&federated_mutex));
|
2005-10-02 19:44:28 -07:00
|
|
|
error:
|
|
|
|
DBUG_RETURN(TRUE);
|
2005-02-16 23:07:10 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
Release the federated handler.
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
federated_db_end()
|
|
|
|
|
|
|
|
RETURN
|
|
|
|
FALSE OK
|
|
|
|
*/
|
|
|
|
|
2006-11-19 18:01:54 -08:00
|
|
|
int federated_done(void *p)
|
2005-02-16 23:07:10 -08:00
|
|
|
{
|
2006-11-19 18:01:54 -08:00
|
|
|
hash_free(&federated_open_tables);
|
|
|
|
VOID(pthread_mutex_destroy(&federated_mutex));
|
|
|
|
|
2005-11-07 16:25:06 +01:00
|
|
|
return 0;
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
2005-11-23 22:45:02 +02:00
|
|
|
|
2007-06-28 00:23:14 -07:00
|
|
|
/**
|
|
|
|
@brief Append identifiers to the string.
|
|
|
|
|
|
|
|
@param[in,out] string The target string.
|
|
|
|
@param[in] name Identifier name
|
|
|
|
@param[in] length Length of identifier name in bytes
|
|
|
|
@param[in] quote_char Quote char to use for quoting identifier.
|
|
|
|
|
|
|
|
@return Operation Status
|
|
|
|
@retval FALSE OK
|
|
|
|
@retval TRUE There was an error appending to the string.
|
|
|
|
|
|
|
|
@note This function is based upon the append_identifier() function
|
|
|
|
in sql_show.cc except that quoting always occurs.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static bool append_ident(String *string, const char *name, uint length,
|
|
|
|
const char quote_char)
|
|
|
|
{
|
|
|
|
bool result;
|
|
|
|
uint clen;
|
|
|
|
const char *name_end;
|
|
|
|
DBUG_ENTER("append_ident");
|
|
|
|
|
|
|
|
if (quote_char)
|
|
|
|
{
|
|
|
|
string->reserve(length * 2 + 2);
|
|
|
|
if ((result= string->append("e_char, 1, system_charset_info)))
|
|
|
|
goto err;
|
|
|
|
|
|
|
|
for (name_end= name+length; name < name_end; name+= clen)
|
|
|
|
{
|
|
|
|
uchar c= *(uchar *) name;
|
|
|
|
if (!(clen= my_mbcharlen(system_charset_info, c)))
|
|
|
|
clen= 1;
|
|
|
|
if (clen == 1 && c == (uchar) quote_char &&
|
|
|
|
(result= string->append("e_char, 1, system_charset_info)))
|
|
|
|
goto err;
|
|
|
|
if ((result= string->append(name, clen, string->charset())))
|
|
|
|
goto err;
|
|
|
|
}
|
|
|
|
result= string->append("e_char, 1, system_charset_info);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
result= string->append(name, length, system_charset_info);
|
|
|
|
|
|
|
|
err:
|
|
|
|
DBUG_RETURN(result);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-09-13 11:59:19 -07:00
|
|
|
static int parse_url_error(FEDERATED_SHARE *share, TABLE *table, int error_num)
|
|
|
|
{
|
2005-09-14 11:02:33 -07:00
|
|
|
char buf[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
int buf_len;
|
2005-09-13 11:59:19 -07:00
|
|
|
DBUG_ENTER("ha_federated parse_url_error");
|
2005-10-08 00:57:40 +03:00
|
|
|
|
|
|
|
buf_len= min(table->s->connect_string.length,
|
|
|
|
FEDERATED_QUERY_BUFFER_SIZE-1);
|
|
|
|
strmake(buf, table->s->connect_string.str, buf_len);
|
2005-09-13 11:59:19 -07:00
|
|
|
my_error(error_num, MYF(0), buf);
|
|
|
|
DBUG_RETURN(error_num);
|
|
|
|
}
|
2007-03-24 01:18:19 -07:00
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
/*
|
|
|
|
retrieve server object which contains server meta-data
|
|
|
|
from the system table given a server's name, set share
|
|
|
|
connection parameter members
|
|
|
|
*/
|
2007-03-24 01:18:19 -07:00
|
|
|
int get_connection(MEM_ROOT *mem_root, FEDERATED_SHARE *share)
|
2006-12-01 19:47:45 -05:00
|
|
|
{
|
|
|
|
int error_num= ER_FOREIGN_SERVER_DOESNT_EXIST;
|
|
|
|
char error_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
2007-03-24 01:18:19 -07:00
|
|
|
FOREIGN_SERVER *server, server_buffer;
|
2006-12-01 19:47:45 -05:00
|
|
|
DBUG_ENTER("ha_federated::get_connection");
|
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
/*
|
|
|
|
get_server_by_name() clones the server if exists and allocates
|
|
|
|
copies of strings in the supplied mem_root
|
|
|
|
*/
|
2006-12-01 19:47:45 -05:00
|
|
|
if (!(server=
|
2007-03-24 01:18:19 -07:00
|
|
|
get_server_by_name(mem_root, share->connection_string, &server_buffer)))
|
2006-12-01 19:47:45 -05:00
|
|
|
{
|
|
|
|
DBUG_PRINT("info", ("get_server_by_name returned > 0 error condition!"));
|
|
|
|
/* need to come up with error handling */
|
|
|
|
error_num=1;
|
|
|
|
goto error;
|
|
|
|
}
|
2007-03-24 01:18:19 -07:00
|
|
|
DBUG_PRINT("info", ("get_server_by_name returned server at %lx",
|
|
|
|
(long unsigned int) server));
|
2006-12-01 19:47:45 -05:00
|
|
|
|
|
|
|
/*
|
|
|
|
Most of these should never be empty strings, error handling will
|
|
|
|
need to be implemented. Also, is this the best way to set the share
|
|
|
|
members? Is there some allocation needed? In running this code, it works
|
|
|
|
except there are errors in the trace file of the share being overrun
|
|
|
|
at the address of the share.
|
|
|
|
*/
|
2007-03-24 01:18:19 -07:00
|
|
|
share->server_name_length= server->server_name_length;
|
|
|
|
share->server_name= server->server_name;
|
|
|
|
share->username= server->username;
|
|
|
|
share->password= server->password;
|
|
|
|
share->database= server->db;
|
|
|
|
#ifndef I_AM_PARANOID
|
|
|
|
share->port= server->port > 0 && server->port < 65536 ?
|
|
|
|
#else
|
|
|
|
share->port= server->port > 1023 && server->port < 65536 ?
|
|
|
|
#endif
|
|
|
|
(ushort) server->port : MYSQL_PORT;
|
|
|
|
share->hostname= server->host;
|
|
|
|
if (!(share->socket= server->socket) &&
|
|
|
|
!strcmp(share->hostname, my_localhost))
|
|
|
|
share->socket= (char *) MYSQL_UNIX_ADDR;
|
|
|
|
share->scheme= server->scheme;
|
2006-12-01 19:47:45 -05:00
|
|
|
|
|
|
|
DBUG_PRINT("info", ("share->username %s", share->username));
|
|
|
|
DBUG_PRINT("info", ("share->password %s", share->password));
|
|
|
|
DBUG_PRINT("info", ("share->hostname %s", share->hostname));
|
|
|
|
DBUG_PRINT("info", ("share->database %s", share->database));
|
|
|
|
DBUG_PRINT("info", ("share->port %d", share->port));
|
2006-12-08 22:19:51 -05:00
|
|
|
DBUG_PRINT("info", ("share->socket %s", share->socket));
|
2006-12-01 19:47:45 -05:00
|
|
|
DBUG_RETURN(0);
|
|
|
|
|
|
|
|
error:
|
|
|
|
my_sprintf(error_buffer,
|
|
|
|
(error_buffer, "server name: '%s' doesn't exist!",
|
|
|
|
share->connection_string));
|
|
|
|
my_error(error_num, MYF(0), error_buffer);
|
|
|
|
DBUG_RETURN(error_num);
|
|
|
|
}
|
2005-09-13 11:59:19 -07:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
/*
|
2005-09-12 18:02:17 -07:00
|
|
|
Parse connection info from table->s->connect_string
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
parse_url()
|
2007-03-24 01:18:19 -07:00
|
|
|
mem_root MEM_ROOT pointer for memory allocation
|
2005-11-23 22:45:02 +02:00
|
|
|
share pointer to FEDERATED share
|
|
|
|
table pointer to current TABLE class
|
|
|
|
table_create_flag determines what error to throw
|
2005-01-05 00:39:13 +02:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
DESCRIPTION
|
2005-11-23 22:45:02 +02:00
|
|
|
Populates the share with information about the connection
|
2005-02-23 00:38:29 -08:00
|
|
|
to the foreign database that will serve as the data source.
|
2006-12-01 19:47:45 -05:00
|
|
|
This string must be specified (currently) in the "CONNECTION" field,
|
2005-01-05 00:39:13 +02:00
|
|
|
listed in the CREATE TABLE statement.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
This string MUST be in the format of any of these:
|
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
CONNECTION="scheme://username:password@hostname:port/database/table"
|
|
|
|
CONNECTION="scheme://username@hostname/database/table"
|
|
|
|
CONNECTION="scheme://username@hostname:port/database/table"
|
|
|
|
CONNECTION="scheme://username:password@hostname/database/table"
|
|
|
|
|
|
|
|
_OR_
|
|
|
|
|
|
|
|
CONNECTION="connection name"
|
|
|
|
|
|
|
|
|
2004-12-16 19:08:32 -08:00
|
|
|
|
|
|
|
An Example:
|
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
CREATE TABLE t1 (id int(32))
|
|
|
|
ENGINE="FEDERATED"
|
|
|
|
CONNECTION="mysql://joe:joespass@192.168.1.111:9308/federated/testtable";
|
|
|
|
|
|
|
|
CREATE TABLE t2 (
|
|
|
|
id int(4) NOT NULL auto_increment,
|
|
|
|
name varchar(32) NOT NULL,
|
|
|
|
PRIMARY KEY(id)
|
|
|
|
) ENGINE="FEDERATED" CONNECTION="my_conn";
|
2004-12-16 19:08:32 -08:00
|
|
|
|
|
|
|
***IMPORTANT***
|
2006-12-01 19:47:45 -05:00
|
|
|
Currently, the Federated Storage Engine only supports connecting to another
|
|
|
|
MySQL Database ("scheme" of "mysql"). Connections using JDBC as well as
|
|
|
|
other connectors are in the planning stage.
|
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-11-23 22:45:02 +02:00
|
|
|
'password' and 'port' are both optional.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
RETURN VALUE
|
2005-07-15 17:33:47 -07:00
|
|
|
0 success
|
|
|
|
error_num particular error code
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
static int parse_url(MEM_ROOT *mem_root, FEDERATED_SHARE *share, TABLE *table,
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
uint table_create_flag)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
uint error_num= (table_create_flag ?
|
|
|
|
ER_FOREIGN_DATA_STRING_INVALID_CANT_CREATE :
|
|
|
|
ER_FOREIGN_DATA_STRING_INVALID);
|
2005-04-06 17:22:21 +03:00
|
|
|
DBUG_ENTER("ha_federated::parse_url");
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-04-06 17:22:21 +03:00
|
|
|
share->port= 0;
|
2005-09-22 03:23:07 +03:00
|
|
|
share->socket= 0;
|
2006-12-08 22:19:51 -05:00
|
|
|
DBUG_PRINT("info", ("share at %lx", (long unsigned int) share));
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
DBUG_PRINT("info", ("Length: %u", (uint) table->s->connect_string.length));
|
|
|
|
DBUG_PRINT("info", ("String: '%.*s'", (int) table->s->connect_string.length,
|
2005-09-12 18:02:17 -07:00
|
|
|
table->s->connect_string.str));
|
2007-03-24 01:18:19 -07:00
|
|
|
share->connection_string= strmake_root(mem_root, table->s->connect_string.str,
|
|
|
|
table->s->connect_string.length);
|
2006-12-01 19:47:45 -05:00
|
|
|
|
2006-12-08 22:19:51 -05:00
|
|
|
DBUG_PRINT("info",("parse_url alloced share->connection_string %lx",
|
|
|
|
(long unsigned int) share->connection_string));
|
2006-12-01 19:47:45 -05:00
|
|
|
|
|
|
|
DBUG_PRINT("info",("share->connection_string %s",share->connection_string));
|
2007-03-24 01:18:19 -07:00
|
|
|
/*
|
|
|
|
No :// or @ in connection string. Must be a straight connection name of
|
|
|
|
either "servername" or "servername/tablename"
|
|
|
|
*/
|
|
|
|
if ( (!strstr(share->connection_string, "://") &&
|
|
|
|
(!strchr(share->connection_string, '@'))))
|
2006-12-01 19:47:45 -05:00
|
|
|
{
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
DBUG_PRINT("info",
|
|
|
|
("share->connection_string %s internal format \
|
2006-12-08 22:19:51 -05:00
|
|
|
share->connection_string %lx",
|
|
|
|
share->connection_string,
|
|
|
|
(long unsigned int) share->connection_string));
|
2004-12-16 19:08:32 -08:00
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
/* ok, so we do a little parsing, but not completely! */
|
2006-12-01 19:47:45 -05:00
|
|
|
share->parsed= FALSE;
|
2007-03-24 01:18:19 -07:00
|
|
|
/*
|
|
|
|
If there is a single '/' in the connection string, this means the user is
|
|
|
|
specifying a table name
|
|
|
|
*/
|
|
|
|
|
|
|
|
if ((share->table_name= strchr(share->connection_string, '/')))
|
|
|
|
{
|
|
|
|
share->connection_string[share->table_name - share->connection_string]= '\0';
|
|
|
|
share->table_name++;
|
|
|
|
share->table_name_length= strlen(share->table_name);
|
2006-12-01 19:47:45 -05:00
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
DBUG_PRINT("info",
|
|
|
|
("internal format, parsed table_name share->connection_string \
|
|
|
|
%s share->table_name %s",
|
|
|
|
share->connection_string, share->table_name));
|
|
|
|
|
|
|
|
/*
|
|
|
|
there better not be any more '/'s !
|
|
|
|
*/
|
|
|
|
if (strchr(share->table_name, '/'))
|
|
|
|
goto error;
|
|
|
|
|
|
|
|
}
|
2005-09-13 11:59:19 -07:00
|
|
|
/*
|
2007-03-24 01:18:19 -07:00
|
|
|
otherwise, straight server name, use tablename of federated table
|
|
|
|
as remote table name
|
2005-09-13 11:59:19 -07:00
|
|
|
*/
|
2007-03-24 01:18:19 -07:00
|
|
|
else
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
connection specifies everything but, resort to
|
|
|
|
expecting remote and foreign table names to match
|
|
|
|
*/
|
|
|
|
share->table_name= strmake_root(mem_root, table->s->table_name.str,
|
|
|
|
(share->table_name_length= table->s->table_name.length));
|
|
|
|
DBUG_PRINT("info",
|
|
|
|
("internal format, default table_name share->connection_string \
|
|
|
|
%s share->table_name %s",
|
|
|
|
share->connection_string, share->table_name));
|
|
|
|
}
|
|
|
|
|
|
|
|
if ((error_num= get_connection(mem_root, share)))
|
|
|
|
goto error;
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
else
|
2006-12-01 19:47:45 -05:00
|
|
|
{
|
|
|
|
share->parsed= TRUE;
|
|
|
|
// Add a null for later termination of table name
|
|
|
|
share->connection_string[table->s->connect_string.length]= 0;
|
|
|
|
share->scheme= share->connection_string;
|
2006-12-08 22:19:51 -05:00
|
|
|
DBUG_PRINT("info",("parse_url alloced share->scheme %lx",
|
|
|
|
(long unsigned int) share->scheme));
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
/*
|
|
|
|
remove addition of null terminator and store length
|
|
|
|
for each string in share
|
|
|
|
*/
|
|
|
|
if (!(share->username= strstr(share->scheme, "://")))
|
|
|
|
goto error;
|
|
|
|
share->scheme[share->username - share->scheme]= '\0';
|
2004-12-16 19:08:32 -08:00
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
if (strcmp(share->scheme, "mysql") != 0)
|
|
|
|
goto error;
|
2004-12-16 19:08:32 -08:00
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
share->username+= 3;
|
|
|
|
|
|
|
|
if (!(share->hostname= strchr(share->username, '@')))
|
|
|
|
goto error;
|
|
|
|
|
|
|
|
share->username[share->hostname - share->username]= '\0';
|
|
|
|
share->hostname++;
|
|
|
|
|
|
|
|
if ((share->password= strchr(share->username, ':')))
|
|
|
|
{
|
|
|
|
share->username[share->password - share->username]= '\0';
|
|
|
|
share->password++;
|
|
|
|
share->username= share->username;
|
|
|
|
/* make sure there isn't an extra / or @ */
|
|
|
|
if ((strchr(share->password, '/') || strchr(share->hostname, '@')))
|
|
|
|
goto error;
|
|
|
|
/*
|
|
|
|
Found that if the string is:
|
2008-03-26 10:51:16 +01:00
|
|
|
user:@hostname:port/db/table
|
|
|
|
Then password is a null string, so set to NULL
|
|
|
|
*/
|
2006-12-01 19:47:45 -05:00
|
|
|
if ((share->password[0] == '\0'))
|
|
|
|
share->password= NULL;
|
|
|
|
}
|
2005-09-13 11:59:19 -07:00
|
|
|
else
|
2006-12-01 19:47:45 -05:00
|
|
|
share->username= share->username;
|
2005-03-15 00:32:21 -08:00
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
/* make sure there isn't an extra / or @ */
|
|
|
|
if ((strchr(share->username, '/')) || (strchr(share->hostname, '@')))
|
|
|
|
goto error;
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
if (!(share->database= strchr(share->hostname, '/')))
|
|
|
|
goto error;
|
|
|
|
share->hostname[share->database - share->hostname]= '\0';
|
|
|
|
share->database++;
|
2004-12-16 19:08:32 -08:00
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
if ((share->sport= strchr(share->hostname, ':')))
|
|
|
|
{
|
|
|
|
share->hostname[share->sport - share->hostname]= '\0';
|
|
|
|
share->sport++;
|
|
|
|
if (share->sport[0] == '\0')
|
|
|
|
share->sport= NULL;
|
|
|
|
else
|
|
|
|
share->port= atoi(share->sport);
|
|
|
|
}
|
2004-12-16 19:08:32 -08:00
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
if (!(share->table_name= strchr(share->database, '/')))
|
|
|
|
goto error;
|
|
|
|
share->database[share->table_name - share->database]= '\0';
|
|
|
|
share->table_name++;
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
share->table_name_length= strlen(share->table_name);
|
|
|
|
|
|
|
|
/* make sure there's not an extra / */
|
|
|
|
if ((strchr(share->table_name, '/')))
|
|
|
|
goto error;
|
|
|
|
|
2008-03-26 10:51:16 +01:00
|
|
|
/*
|
|
|
|
If hostname is omitted, we set it to NULL. According to
|
|
|
|
mysql_real_connect() manual:
|
|
|
|
The value of host may be either a hostname or an IP address.
|
|
|
|
If host is NULL or the string "localhost", a connection to the
|
|
|
|
local host is assumed.
|
|
|
|
*/
|
2006-12-01 19:47:45 -05:00
|
|
|
if (share->hostname[0] == '\0')
|
|
|
|
share->hostname= NULL;
|
|
|
|
}
|
2008-03-26 16:22:10 +01:00
|
|
|
|
2005-09-13 11:59:19 -07:00
|
|
|
if (!share->port)
|
|
|
|
{
|
2008-03-20 19:07:17 +04:00
|
|
|
if (!share->hostname || strcmp(share->hostname, my_localhost) == 0)
|
2008-03-26 16:22:10 +01:00
|
|
|
share->socket= (char*) MYSQL_UNIX_ADDR;
|
2004-12-11 12:03:51 -08:00
|
|
|
else
|
2005-09-13 11:59:19 -07:00
|
|
|
share->port= MYSQL_PORT;
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
2005-09-13 11:59:19 -07:00
|
|
|
|
|
|
|
DBUG_PRINT("info",
|
2005-11-23 22:45:02 +02:00
|
|
|
("scheme: %s username: %s password: %s \
|
2006-12-01 19:47:45 -05:00
|
|
|
hostname: %s port: %d db: %s tablename: %s",
|
2005-09-13 11:59:19 -07:00
|
|
|
share->scheme, share->username, share->password,
|
|
|
|
share->hostname, share->port, share->database,
|
|
|
|
share->table_name));
|
2005-02-23 00:38:29 -08:00
|
|
|
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
|
|
|
|
error:
|
2005-09-13 11:59:19 -07:00
|
|
|
DBUG_RETURN(parse_url_error(share, table, error_num));
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
2005-07-19 22:21:12 +04:00
|
|
|
/*****************************************************************************
|
|
|
|
** FEDERATED tables
|
|
|
|
*****************************************************************************/
|
|
|
|
|
2006-09-29 17:19:02 -07:00
|
|
|
ha_federated::ha_federated(handlerton *hton,
|
|
|
|
TABLE_SHARE *table_arg)
|
|
|
|
:handler(hton, table_arg),
|
2006-06-27 23:49:48 -07:00
|
|
|
mysql(0), stored_result(0)
|
2005-11-09 05:53:34 -08:00
|
|
|
{
|
|
|
|
trx_next= 0;
|
2007-06-28 16:03:01 -07:00
|
|
|
bzero(&bulk_insert, sizeof(bulk_insert));
|
2005-11-09 05:53:34 -08:00
|
|
|
}
|
2005-07-19 22:21:12 +04:00
|
|
|
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
/*
|
2004-12-11 12:03:51 -08:00
|
|
|
Convert MySQL result set row to handler internal format
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
convert_row_to_internal_format()
|
2005-02-23 00:38:29 -08:00
|
|
|
record Byte pointer to record
|
2004-12-11 12:03:51 -08:00
|
|
|
row MySQL result set row from fetchrow()
|
2006-06-27 23:49:48 -07:00
|
|
|
result Result set to use
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
DESCRIPTION
|
2005-02-23 00:38:29 -08:00
|
|
|
This method simply iterates through a row returned via fetchrow with
|
2004-12-11 12:03:51 -08:00
|
|
|
values from a successful SELECT , and then stores each column's value
|
|
|
|
in the field object via the field object pointer (pointing to the table's
|
2005-02-23 00:38:29 -08:00
|
|
|
array of field object pointers). This is how the handler needs the data
|
2004-12-11 12:03:51 -08:00
|
|
|
to be stored to then return results back to the user
|
|
|
|
|
|
|
|
RETURN VALUE
|
2005-02-23 00:38:29 -08:00
|
|
|
0 After fields have had field values stored from record
|
2005-11-23 22:45:02 +02:00
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
uint ha_federated::convert_row_to_internal_format(uchar *record,
|
2006-06-27 23:49:48 -07:00
|
|
|
MYSQL_ROW row,
|
|
|
|
MYSQL_RES *result)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
ulong *lengths;
|
|
|
|
Field **field;
|
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 18:52:22 +03:00
|
|
|
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
|
2004-12-11 12:03:51 -08:00
|
|
|
DBUG_ENTER("ha_federated::convert_row_to_internal_format");
|
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
lengths= mysql_fetch_lengths(result);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-11-23 22:45:02 +02:00
|
|
|
for (field= table->field; *field; field++, row++, lengths++)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
/*
|
|
|
|
index variable to move us through the row at the
|
|
|
|
same iterative step as the field
|
|
|
|
*/
|
|
|
|
my_ptrdiff_t old_ptr;
|
|
|
|
old_ptr= (my_ptrdiff_t) (record - table->record[0]);
|
2005-11-23 22:45:02 +02:00
|
|
|
(*field)->move_field_offset(old_ptr);
|
|
|
|
if (!*row)
|
2008-08-15 11:40:05 +05:00
|
|
|
{
|
2004-12-11 12:03:51 -08:00
|
|
|
(*field)->set_null();
|
2008-08-15 11:40:05 +05:00
|
|
|
(*field)->reset();
|
|
|
|
}
|
2004-12-11 12:03:51 -08:00
|
|
|
else
|
2005-07-15 17:33:47 -07:00
|
|
|
{
|
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 18:52:22 +03:00
|
|
|
if (bitmap_is_set(table->read_set, (*field)->field_index))
|
|
|
|
{
|
|
|
|
(*field)->set_notnull();
|
|
|
|
(*field)->store(*row, *lengths, &my_charset_bin);
|
|
|
|
}
|
2005-07-15 17:33:47 -07:00
|
|
|
}
|
2005-11-23 22:45:02 +02:00
|
|
|
(*field)->move_field_offset(-old_ptr);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
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 18:52:22 +03:00
|
|
|
dbug_tmp_restore_column_map(table->write_set, old_map);
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool emit_key_part_name(String *to, KEY_PART_INFO *part)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("emit_key_part_name");
|
2007-06-28 00:23:14 -07:00
|
|
|
if (append_ident(to, part->field->field_name,
|
|
|
|
strlen(part->field->field_name), ident_quote_char))
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_RETURN(1); // Out of memory
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool emit_key_part_element(String *to, KEY_PART_INFO *part,
|
|
|
|
bool needs_quotes, bool is_like,
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
const uchar *ptr, uint len)
|
2005-07-15 17:33:47 -07:00
|
|
|
{
|
|
|
|
Field *field= part->field;
|
|
|
|
DBUG_ENTER("emit_key_part_element");
|
|
|
|
|
2006-08-02 12:36:01 +02:00
|
|
|
if (needs_quotes && to->append(STRING_WITH_LEN("'")))
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_RETURN(1);
|
|
|
|
|
|
|
|
if (part->type == HA_KEYTYPE_BIT)
|
|
|
|
{
|
|
|
|
char buff[STRING_BUFFER_USUAL_SIZE], *buf= buff;
|
|
|
|
|
|
|
|
*buf++= '0';
|
|
|
|
*buf++= 'x';
|
2005-10-12 00:58:22 +03:00
|
|
|
buf= octet2hex(buf, (char*) ptr, len);
|
|
|
|
if (to->append((char*) buff, (uint)(buf - buff)))
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_RETURN(1);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
2005-07-15 17:33:47 -07:00
|
|
|
else if (part->key_part_flag & HA_BLOB_PART)
|
|
|
|
{
|
|
|
|
String blob;
|
|
|
|
uint blob_length= uint2korr(ptr);
|
|
|
|
blob.set_quick((char*) ptr+HA_KEY_BLOB_LENGTH,
|
|
|
|
blob_length, &my_charset_bin);
|
|
|
|
if (append_escaped(to, &blob))
|
|
|
|
DBUG_RETURN(1);
|
|
|
|
}
|
|
|
|
else if (part->key_part_flag & HA_VAR_LENGTH_PART)
|
|
|
|
{
|
|
|
|
String varchar;
|
|
|
|
uint var_length= uint2korr(ptr);
|
|
|
|
varchar.set_quick((char*) ptr+HA_KEY_BLOB_LENGTH,
|
|
|
|
var_length, &my_charset_bin);
|
|
|
|
if (append_escaped(to, &varchar))
|
|
|
|
DBUG_RETURN(1);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
char strbuff[MAX_FIELD_WIDTH];
|
|
|
|
String str(strbuff, sizeof(strbuff), part->field->charset()), *res;
|
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
res= field->val_str(&str, ptr);
|
2005-07-15 17:33:47 -07:00
|
|
|
|
|
|
|
if (field->result_type() == STRING_RESULT)
|
|
|
|
{
|
|
|
|
if (append_escaped(to, res))
|
|
|
|
DBUG_RETURN(1);
|
|
|
|
}
|
|
|
|
else if (to->append(res->ptr(), res->length()))
|
|
|
|
DBUG_RETURN(1);
|
|
|
|
}
|
|
|
|
|
2006-08-02 12:36:01 +02:00
|
|
|
if (is_like && to->append(STRING_WITH_LEN("%")))
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_RETURN(1);
|
|
|
|
|
2006-08-02 12:36:01 +02:00
|
|
|
if (needs_quotes && to->append(STRING_WITH_LEN("'")))
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_RETURN(1);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
/*
|
|
|
|
Create a WHERE clause based off of values in keys
|
|
|
|
Note: This code was inspired by key_copy from key.cc
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
create_where_from_key ()
|
|
|
|
to String object to store WHERE clause
|
|
|
|
key_info KEY struct pointer
|
|
|
|
key byte pointer containing key
|
|
|
|
key_length length of key
|
2005-07-15 17:33:47 -07:00
|
|
|
range_type 0 - no range, 1 - min range, 2 - max range
|
|
|
|
(see enum range_operation)
|
2005-02-23 00:38:29 -08:00
|
|
|
|
|
|
|
DESCRIPTION
|
|
|
|
Using iteration through all the keys via a KEY_PART_INFO pointer,
|
|
|
|
This method 'extracts' the value of each key in the byte pointer
|
|
|
|
*key, and for each key found, constructs an appropriate WHERE clause
|
|
|
|
|
|
|
|
RETURN VALUE
|
|
|
|
0 After all keys have been accounted for to create the WHERE clause
|
|
|
|
1 No keys found
|
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
Range flags Table per Timour:
|
|
|
|
|
|
|
|
-----------------
|
|
|
|
- start_key:
|
|
|
|
* ">" -> HA_READ_AFTER_KEY
|
|
|
|
* ">=" -> HA_READ_KEY_OR_NEXT
|
|
|
|
* "=" -> HA_READ_KEY_EXACT
|
|
|
|
|
|
|
|
- end_key:
|
|
|
|
* "<" -> HA_READ_BEFORE_KEY
|
|
|
|
* "<=" -> HA_READ_AFTER_KEY
|
|
|
|
|
|
|
|
records_in_range:
|
|
|
|
-----------------
|
|
|
|
- start_key:
|
|
|
|
* ">" -> HA_READ_AFTER_KEY
|
|
|
|
* ">=" -> HA_READ_KEY_EXACT
|
|
|
|
* "=" -> HA_READ_KEY_EXACT
|
|
|
|
|
|
|
|
- end_key:
|
|
|
|
* "<" -> HA_READ_BEFORE_KEY
|
|
|
|
* "<=" -> HA_READ_AFTER_KEY
|
|
|
|
* "=" -> HA_READ_AFTER_KEY
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
0 HA_READ_KEY_EXACT, Find first record else error
|
|
|
|
1 HA_READ_KEY_OR_NEXT, Record or next record
|
|
|
|
2 HA_READ_KEY_OR_PREV, Record or previous
|
|
|
|
3 HA_READ_AFTER_KEY, Find next rec. after key-record
|
|
|
|
4 HA_READ_BEFORE_KEY, Find next rec. before key-record
|
|
|
|
5 HA_READ_PREFIX, Key which as same prefix
|
|
|
|
6 HA_READ_PREFIX_LAST, Last key with the same prefix
|
|
|
|
7 HA_READ_PREFIX_LAST_OR_PREV, Last or prev key with the same prefix
|
|
|
|
|
|
|
|
Flags that I've found:
|
|
|
|
|
|
|
|
id, primary key, varchar
|
|
|
|
|
|
|
|
id = 'ccccc'
|
|
|
|
records_in_range: start_key 0 end_key 3
|
|
|
|
read_range_first: start_key 0 end_key NULL
|
|
|
|
|
|
|
|
id > 'ccccc'
|
|
|
|
records_in_range: start_key 3 end_key NULL
|
|
|
|
read_range_first: start_key 3 end_key NULL
|
|
|
|
|
|
|
|
id < 'ccccc'
|
|
|
|
records_in_range: start_key NULL end_key 4
|
|
|
|
read_range_first: start_key NULL end_key 4
|
|
|
|
|
|
|
|
id <= 'ccccc'
|
|
|
|
records_in_range: start_key NULL end_key 3
|
|
|
|
read_range_first: start_key NULL end_key 3
|
|
|
|
|
|
|
|
id >= 'ccccc'
|
|
|
|
records_in_range: start_key 0 end_key NULL
|
|
|
|
read_range_first: start_key 1 end_key NULL
|
|
|
|
|
|
|
|
id like 'cc%cc'
|
|
|
|
records_in_range: start_key 0 end_key 3
|
|
|
|
read_range_first: start_key 1 end_key 3
|
|
|
|
|
|
|
|
id > 'aaaaa' and id < 'ccccc'
|
|
|
|
records_in_range: start_key 3 end_key 4
|
|
|
|
read_range_first: start_key 3 end_key 4
|
|
|
|
|
|
|
|
id >= 'aaaaa' and id < 'ccccc';
|
|
|
|
records_in_range: start_key 0 end_key 4
|
|
|
|
read_range_first: start_key 1 end_key 4
|
|
|
|
|
|
|
|
id >= 'aaaaa' and id <= 'ccccc';
|
|
|
|
records_in_range: start_key 0 end_key 3
|
|
|
|
read_range_first: start_key 1 end_key 3
|
|
|
|
|
|
|
|
id > 'aaaaa' and id <= 'ccccc';
|
|
|
|
records_in_range: start_key 3 end_key 3
|
|
|
|
read_range_first: start_key 3 end_key 3
|
|
|
|
|
|
|
|
numeric keys:
|
|
|
|
|
|
|
|
id = 4
|
|
|
|
index_read_idx: start_key 0 end_key NULL
|
|
|
|
|
|
|
|
id > 4
|
|
|
|
records_in_range: start_key 3 end_key NULL
|
|
|
|
read_range_first: start_key 3 end_key NULL
|
|
|
|
|
|
|
|
id >= 4
|
|
|
|
records_in_range: start_key 0 end_key NULL
|
|
|
|
read_range_first: start_key 1 end_key NULL
|
|
|
|
|
|
|
|
id < 4
|
|
|
|
records_in_range: start_key NULL end_key 4
|
|
|
|
read_range_first: start_key NULL end_key 4
|
|
|
|
|
|
|
|
id <= 4
|
|
|
|
records_in_range: start_key NULL end_key 3
|
|
|
|
read_range_first: start_key NULL end_key 3
|
|
|
|
|
|
|
|
id like 4
|
|
|
|
full table scan, select * from
|
|
|
|
|
|
|
|
id > 2 and id < 8
|
|
|
|
records_in_range: start_key 3 end_key 4
|
|
|
|
read_range_first: start_key 3 end_key 4
|
|
|
|
|
|
|
|
id >= 2 and id < 8
|
|
|
|
records_in_range: start_key 0 end_key 4
|
|
|
|
read_range_first: start_key 1 end_key 4
|
|
|
|
|
|
|
|
id >= 2 and id <= 8
|
|
|
|
records_in_range: start_key 0 end_key 3
|
|
|
|
read_range_first: start_key 1 end_key 3
|
|
|
|
|
|
|
|
id > 2 and id <= 8
|
|
|
|
records_in_range: start_key 3 end_key 3
|
|
|
|
read_range_first: start_key 3 end_key 3
|
|
|
|
|
|
|
|
multi keys (id int, name varchar, other varchar)
|
|
|
|
|
|
|
|
id = 1;
|
|
|
|
records_in_range: start_key 0 end_key 3
|
|
|
|
read_range_first: start_key 0 end_key NULL
|
|
|
|
|
|
|
|
id > 4;
|
|
|
|
id > 2 and name = '333'; remote: id > 2
|
|
|
|
id > 2 and name > '333'; remote: id > 2
|
|
|
|
id > 2 and name > '333' and other < 'ddd'; remote: id > 2 no results
|
|
|
|
id > 2 and name >= '333' and other < 'ddd'; remote: id > 2 1 result
|
|
|
|
id >= 4 and name = 'eric was here' and other > 'eeee';
|
|
|
|
records_in_range: start_key 3 end_key NULL
|
|
|
|
read_range_first: start_key 3 end_key NULL
|
|
|
|
|
|
|
|
id >= 4;
|
|
|
|
id >= 2 and name = '333' and other < 'ddd';
|
|
|
|
remote: `id` >= 2 AND `name` >= '333';
|
|
|
|
records_in_range: start_key 0 end_key NULL
|
|
|
|
read_range_first: start_key 1 end_key NULL
|
|
|
|
|
|
|
|
id < 4;
|
|
|
|
id < 3 and name = '222' and other <= 'ccc'; remote: id < 3
|
|
|
|
records_in_range: start_key NULL end_key 4
|
|
|
|
read_range_first: start_key NULL end_key 4
|
|
|
|
|
|
|
|
id <= 4;
|
|
|
|
records_in_range: start_key NULL end_key 3
|
|
|
|
read_range_first: start_key NULL end_key 3
|
|
|
|
|
|
|
|
id like 4;
|
|
|
|
full table scan
|
|
|
|
|
|
|
|
id > 2 and id < 4;
|
|
|
|
records_in_range: start_key 3 end_key 4
|
|
|
|
read_range_first: start_key 3 end_key 4
|
|
|
|
|
|
|
|
id >= 2 and id < 4;
|
|
|
|
records_in_range: start_key 0 end_key 4
|
|
|
|
read_range_first: start_key 1 end_key 4
|
|
|
|
|
|
|
|
id >= 2 and id <= 4;
|
|
|
|
records_in_range: start_key 0 end_key 3
|
|
|
|
read_range_first: start_key 1 end_key 3
|
|
|
|
|
|
|
|
id > 2 and id <= 4;
|
|
|
|
id = 6 and name = 'eric was here' and other > 'eeee';
|
|
|
|
remote: (`id` > 6 AND `name` > 'eric was here' AND `other` > 'eeee')
|
|
|
|
AND (`id` <= 6) AND ( AND `name` <= 'eric was here')
|
|
|
|
no results
|
|
|
|
records_in_range: start_key 3 end_key 3
|
|
|
|
read_range_first: start_key 3 end_key 3
|
|
|
|
|
|
|
|
Summary:
|
|
|
|
|
|
|
|
* If the start key flag is 0 the max key flag shouldn't even be set,
|
|
|
|
and if it is, the query produced would be invalid.
|
|
|
|
* Multipart keys, even if containing some or all numeric columns,
|
|
|
|
are treated the same as non-numeric keys
|
|
|
|
|
|
|
|
If the query is " = " (quotes or not):
|
|
|
|
- records in range start key flag HA_READ_KEY_EXACT,
|
|
|
|
end key flag HA_READ_AFTER_KEY (incorrect)
|
|
|
|
- any other: start key flag HA_READ_KEY_OR_NEXT,
|
|
|
|
end key flag HA_READ_AFTER_KEY (correct)
|
|
|
|
|
|
|
|
* 'like' queries (of key)
|
|
|
|
- Numeric, full table scan
|
|
|
|
- Non-numeric
|
|
|
|
records_in_range: start_key 0 end_key 3
|
|
|
|
other : start_key 1 end_key 3
|
|
|
|
|
|
|
|
* If the key flag is HA_READ_AFTER_KEY:
|
|
|
|
if start_key, append >
|
|
|
|
if end_key, append <=
|
|
|
|
|
|
|
|
* If create_where_key was called by records_in_range:
|
|
|
|
|
|
|
|
- if the key is numeric:
|
|
|
|
start key flag is 0 when end key is NULL, end key flag is 3 or 4
|
|
|
|
- if create_where_key was called by any other function:
|
|
|
|
start key flag is 1 when end key is NULL, end key flag is 3 or 4
|
|
|
|
- if the key is non-numeric, or multipart
|
|
|
|
When the query is an exact match, the start key flag is 0,
|
|
|
|
end key flag is 3 for what should be a no-range condition where
|
|
|
|
you should have 0 and max key NULL, which it is if called by
|
|
|
|
read_range_first
|
|
|
|
|
|
|
|
Conclusion:
|
|
|
|
|
|
|
|
1. Need logic to determin if a key is min or max when the flag is
|
|
|
|
HA_READ_AFTER_KEY, and handle appending correct operator accordingly
|
|
|
|
|
|
|
|
2. Need a boolean flag to pass to create_where_from_key, used in the
|
|
|
|
switch statement. Add 1 to the flag if:
|
|
|
|
- start key flag is HA_READ_KEY_EXACT and the end key is NULL
|
|
|
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
bool ha_federated::create_where_from_key(String *to,
|
|
|
|
KEY *key_info,
|
|
|
|
const key_range *start_key,
|
|
|
|
const key_range *end_key,
|
2007-01-27 03:46:45 +02:00
|
|
|
bool from_records_in_range,
|
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 18:52:22 +03:00
|
|
|
bool eq_range)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
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 18:52:22 +03:00
|
|
|
bool both_not_null=
|
2005-07-17 04:10:19 +02:00
|
|
|
(start_key != NULL && end_key != NULL) ? TRUE : FALSE;
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
const uchar *ptr;
|
2005-07-15 17:33:47 -07:00
|
|
|
uint remainder, length;
|
|
|
|
char tmpbuff[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
String tmp(tmpbuff, sizeof(tmpbuff), system_charset_info);
|
|
|
|
const key_range *ranges[2]= { start_key, end_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 18:52:22 +03:00
|
|
|
my_bitmap_map *old_map;
|
2005-01-13 18:21:54 -08:00
|
|
|
DBUG_ENTER("ha_federated::create_where_from_key");
|
2005-10-08 00:57:40 +03:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
tmp.length(0);
|
|
|
|
if (start_key == NULL && end_key == NULL)
|
|
|
|
DBUG_RETURN(1);
|
2005-01-13 18:21:54 -08:00
|
|
|
|
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 18:52:22 +03:00
|
|
|
old_map= dbug_tmp_use_all_columns(table, table->write_set);
|
|
|
|
for (uint i= 0; i <= 1; i++)
|
2005-07-15 17:33:47 -07:00
|
|
|
{
|
|
|
|
bool needs_quotes;
|
|
|
|
KEY_PART_INFO *key_part;
|
|
|
|
if (ranges[i] == NULL)
|
|
|
|
continue;
|
2005-01-13 18:21:54 -08:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
if (both_not_null)
|
2005-01-13 18:21:54 -08:00
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
if (i > 0)
|
2006-08-02 12:36:01 +02:00
|
|
|
tmp.append(STRING_WITH_LEN(") AND ("));
|
2005-07-15 17:33:47 -07:00
|
|
|
else
|
2006-08-02 12:36:01 +02:00
|
|
|
tmp.append(STRING_WITH_LEN(" ("));
|
2005-01-13 18:21:54 -08:00
|
|
|
}
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
for (key_part= key_info->key_part,
|
|
|
|
remainder= key_info->key_parts,
|
|
|
|
length= ranges[i]->length,
|
|
|
|
ptr= ranges[i]->key; ;
|
|
|
|
remainder--,
|
|
|
|
key_part++)
|
2005-01-13 18:21:54 -08:00
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
Field *field= key_part->field;
|
|
|
|
uint store_length= key_part->store_length;
|
|
|
|
uint part_length= min(store_length, length);
|
2006-07-25 16:12:48 -07:00
|
|
|
needs_quotes= field->str_needs_quotes();
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
DBUG_DUMP("key, start of loop", ptr, length);
|
2005-02-02 22:39:21 +02:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
if (key_part->null_bit)
|
|
|
|
{
|
|
|
|
if (*ptr++)
|
|
|
|
{
|
2008-02-14 16:27:01 +04:00
|
|
|
/*
|
|
|
|
We got "IS [NOT] NULL" condition against nullable column. We
|
|
|
|
distinguish between "IS NOT NULL" and "IS NULL" by flag. For
|
|
|
|
"IS NULL", flag is set to HA_READ_KEY_EXACT.
|
|
|
|
*/
|
2005-07-15 17:33:47 -07:00
|
|
|
if (emit_key_part_name(&tmp, key_part) ||
|
2008-02-14 18:23:58 +04:00
|
|
|
(ranges[i]->flag == HA_READ_KEY_EXACT ?
|
|
|
|
tmp.append(STRING_WITH_LEN(" IS NULL ")) :
|
|
|
|
tmp.append(STRING_WITH_LEN(" IS NOT NULL "))))
|
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 18:52:22 +03:00
|
|
|
goto err;
|
2008-02-14 16:27:01 +04:00
|
|
|
/*
|
|
|
|
We need to adjust pointer and length to be prepared for next
|
|
|
|
key part. As well as check if this was last key part.
|
|
|
|
*/
|
|
|
|
goto prepare_for_next_key_part;
|
2005-07-15 17:33:47 -07:00
|
|
|
}
|
|
|
|
}
|
2005-01-20 18:36:40 -08:00
|
|
|
|
2006-08-02 12:36:01 +02:00
|
|
|
if (tmp.append(STRING_WITH_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 18:52:22 +03:00
|
|
|
goto err;
|
2004-12-11 12:03:51 -08:00
|
|
|
|
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 18:52:22 +03:00
|
|
|
switch (ranges[i]->flag) {
|
|
|
|
case HA_READ_KEY_EXACT:
|
2005-10-18 15:54:50 -07:00
|
|
|
DBUG_PRINT("info", ("federated HA_READ_KEY_EXACT %d", i));
|
2005-07-15 17:33:47 -07:00
|
|
|
if (store_length >= length ||
|
|
|
|
!needs_quotes ||
|
|
|
|
key_part->type == HA_KEYTYPE_BIT ||
|
|
|
|
field->result_type() != STRING_RESULT)
|
|
|
|
{
|
|
|
|
if (emit_key_part_name(&tmp, key_part))
|
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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-15 17:33:47 -07:00
|
|
|
|
2006-12-15 00:51:37 +02:00
|
|
|
if (from_records_in_range)
|
2005-07-15 17:33:47 -07:00
|
|
|
{
|
2006-08-02 12:36:01 +02:00
|
|
|
if (tmp.append(STRING_WITH_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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-15 17:33:47 -07:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2006-08-02 12:36:01 +02:00
|
|
|
if (tmp.append(STRING_WITH_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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-15 17:33:47 -07:00
|
|
|
}
|
2005-07-17 04:10:19 +02:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
if (emit_key_part_element(&tmp, key_part, needs_quotes, 0, ptr,
|
|
|
|
part_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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-15 17:33:47 -07:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2005-11-23 22:45:02 +02:00
|
|
|
/* LIKE */
|
2005-07-15 17:33:47 -07:00
|
|
|
if (emit_key_part_name(&tmp, key_part) ||
|
2006-08-02 12:36:01 +02:00
|
|
|
tmp.append(STRING_WITH_LEN(" LIKE ")) ||
|
2005-07-17 04:10:19 +02:00
|
|
|
emit_key_part_element(&tmp, key_part, needs_quotes, 1, ptr,
|
|
|
|
part_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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-15 17:33:47 -07:00
|
|
|
}
|
|
|
|
break;
|
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 18:52:22 +03:00
|
|
|
case HA_READ_AFTER_KEY:
|
|
|
|
if (eq_range)
|
|
|
|
{
|
|
|
|
if (tmp.append("1=1")) // Dummy
|
|
|
|
goto err;
|
|
|
|
break;
|
|
|
|
}
|
2005-10-18 15:54:50 -07:00
|
|
|
DBUG_PRINT("info", ("federated HA_READ_AFTER_KEY %d", i));
|
2005-07-15 17:33:47 -07:00
|
|
|
if (store_length >= length) /* end key */
|
|
|
|
{
|
|
|
|
if (emit_key_part_name(&tmp, key_part))
|
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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-15 17:33:47 -07:00
|
|
|
|
|
|
|
if (i > 0) /* end key */
|
2005-07-17 04:10:19 +02:00
|
|
|
{
|
2006-08-02 12:36:01 +02:00
|
|
|
if (tmp.append(STRING_WITH_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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-17 04:10:19 +02:00
|
|
|
}
|
2005-07-15 17:33:47 -07:00
|
|
|
else /* start key */
|
2005-07-17 04:10:19 +02:00
|
|
|
{
|
2006-08-02 12:36:01 +02:00
|
|
|
if (tmp.append(STRING_WITH_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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-17 04:10:19 +02:00
|
|
|
}
|
2005-07-15 17:33:47 -07:00
|
|
|
|
|
|
|
if (emit_key_part_element(&tmp, key_part, needs_quotes, 0, ptr,
|
2005-07-17 04:10:19 +02:00
|
|
|
part_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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-17 04:10:19 +02:00
|
|
|
}
|
2005-07-15 17:33:47 -07:00
|
|
|
break;
|
|
|
|
}
|
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 18:52:22 +03:00
|
|
|
case HA_READ_KEY_OR_NEXT:
|
2005-10-18 15:54:50 -07:00
|
|
|
DBUG_PRINT("info", ("federated HA_READ_KEY_OR_NEXT %d", i));
|
2005-07-17 04:10:19 +02:00
|
|
|
if (emit_key_part_name(&tmp, key_part) ||
|
2006-08-02 12:36:01 +02:00
|
|
|
tmp.append(STRING_WITH_LEN(" >= ")) ||
|
2005-07-15 17:33:47 -07:00
|
|
|
emit_key_part_element(&tmp, key_part, needs_quotes, 0, ptr,
|
2005-07-17 04:10:19 +02:00
|
|
|
part_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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-15 17:33:47 -07:00
|
|
|
break;
|
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 18:52:22 +03:00
|
|
|
case HA_READ_BEFORE_KEY:
|
2005-10-18 15:54:50 -07:00
|
|
|
DBUG_PRINT("info", ("federated HA_READ_BEFORE_KEY %d", i));
|
2005-07-15 17:33:47 -07:00
|
|
|
if (store_length >= length)
|
|
|
|
{
|
|
|
|
if (emit_key_part_name(&tmp, key_part) ||
|
2006-08-02 12:36:01 +02:00
|
|
|
tmp.append(STRING_WITH_LEN(" < ")) ||
|
2005-07-17 04:10:19 +02:00
|
|
|
emit_key_part_element(&tmp, key_part, needs_quotes, 0, ptr,
|
|
|
|
part_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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-15 17:33:47 -07:00
|
|
|
break;
|
|
|
|
}
|
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 18:52:22 +03:00
|
|
|
case HA_READ_KEY_OR_PREV:
|
2005-10-18 15:54:50 -07:00
|
|
|
DBUG_PRINT("info", ("federated HA_READ_KEY_OR_PREV %d", i));
|
2005-07-17 04:10:19 +02:00
|
|
|
if (emit_key_part_name(&tmp, key_part) ||
|
2006-08-02 12:36:01 +02:00
|
|
|
tmp.append(STRING_WITH_LEN(" <= ")) ||
|
2005-07-15 17:33:47 -07:00
|
|
|
emit_key_part_element(&tmp, key_part, needs_quotes, 0, ptr,
|
|
|
|
part_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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-15 17:33:47 -07:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
DBUG_PRINT("info",("cannot handle flag %d", ranges[i]->flag));
|
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
2006-06-04 18:52:22 +03:00
|
|
|
goto err;
|
2005-01-13 18:21:54 -08:00
|
|
|
}
|
2006-08-02 12:36:01 +02:00
|
|
|
if (tmp.append(STRING_WITH_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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-15 17:33:47 -07:00
|
|
|
|
2008-02-14 16:27:01 +04:00
|
|
|
prepare_for_next_key_part:
|
2005-07-15 17:33:47 -07:00
|
|
|
if (store_length >= length)
|
|
|
|
break;
|
|
|
|
DBUG_PRINT("info", ("remainder %d", remainder));
|
|
|
|
DBUG_ASSERT(remainder > 1);
|
|
|
|
length-= store_length;
|
2008-02-14 16:27:01 +04:00
|
|
|
/*
|
|
|
|
For nullable columns, null-byte is already skipped before, that is
|
|
|
|
ptr was incremented by 1. Since store_length still counts null-byte,
|
|
|
|
we need to subtract 1 from store_length.
|
|
|
|
*/
|
|
|
|
ptr+= store_length - test(key_part->null_bit);
|
2006-08-02 12:36:01 +02:00
|
|
|
if (tmp.append(STRING_WITH_LEN(" AND ")))
|
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 18:52:22 +03:00
|
|
|
goto err;
|
2005-07-15 17:33:47 -07:00
|
|
|
|
|
|
|
DBUG_PRINT("info",
|
|
|
|
("create_where_from_key WHERE clause: %s",
|
|
|
|
tmp.c_ptr_quick()));
|
2005-01-13 18:21:54 -08:00
|
|
|
}
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
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 18:52:22 +03:00
|
|
|
dbug_tmp_restore_column_map(table->write_set, old_map);
|
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
if (both_not_null)
|
2006-08-02 12:36:01 +02:00
|
|
|
if (tmp.append(STRING_WITH_LEN(") ")))
|
2005-07-17 04:10:19 +02:00
|
|
|
DBUG_RETURN(1);
|
2005-07-15 17:33:47 -07:00
|
|
|
|
2006-08-02 12:36:01 +02:00
|
|
|
if (to->append(STRING_WITH_LEN(" WHERE ")))
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_RETURN(1);
|
|
|
|
|
|
|
|
if (to->append(tmp))
|
|
|
|
DBUG_RETURN(1);
|
|
|
|
|
|
|
|
DBUG_RETURN(0);
|
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
2006-06-04 18:52:22 +03:00
|
|
|
|
|
|
|
err:
|
|
|
|
dbug_tmp_restore_column_map(table->write_set, old_map);
|
|
|
|
DBUG_RETURN(1);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
Example of simple lock controls. The "share" it creates is structure we will
|
|
|
|
pass to each federated handler. Do you have to have one of these? Well, you
|
|
|
|
have pieces that are used for locking, and they are needed to function.
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
static FEDERATED_SHARE *get_share(const char *table_name, TABLE *table)
|
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
char query_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
Field **field;
|
2004-12-11 12:03:51 -08:00
|
|
|
String query(query_buffer, sizeof(query_buffer), &my_charset_bin);
|
2006-06-27 23:49:48 -07:00
|
|
|
FEDERATED_SHARE *share= NULL, tmp_share;
|
2007-03-24 01:18:19 -07:00
|
|
|
MEM_ROOT mem_root;
|
|
|
|
DBUG_ENTER("ha_federated.cc::get_share");
|
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
/*
|
|
|
|
In order to use this string, we must first zero it's length,
|
|
|
|
or it will contain garbage
|
2005-11-23 22:45:02 +02:00
|
|
|
*/
|
2004-12-11 12:03:51 -08:00
|
|
|
query.length(0);
|
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
init_alloc_root(&mem_root, 256, 0);
|
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
pthread_mutex_lock(&federated_mutex);
|
|
|
|
|
2006-12-01 19:47:45 -05:00
|
|
|
tmp_share.share_key= table_name;
|
|
|
|
tmp_share.share_key_length= strlen(table_name);
|
2007-03-24 01:18:19 -07:00
|
|
|
if (parse_url(&mem_root, &tmp_share, table, 0))
|
2006-06-27 23:49:48 -07:00
|
|
|
goto error;
|
|
|
|
|
|
|
|
/* TODO: change tmp_share.scheme to LEX_STRING object */
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
if (!(share= (FEDERATED_SHARE *) hash_search(&federated_open_tables,
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
(uchar*) tmp_share.share_key,
|
2006-06-27 23:49:48 -07:00
|
|
|
tmp_share.
|
2006-12-01 19:47:45 -05:00
|
|
|
share_key_length)))
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
|
|
|
query.set_charset(system_charset_info);
|
2006-08-02 12:36:01 +02:00
|
|
|
query.append(STRING_WITH_LEN("SELECT "));
|
2005-07-15 17:33:47 -07:00
|
|
|
for (field= table->field; *field; field++)
|
|
|
|
{
|
2007-06-28 00:23:14 -07:00
|
|
|
append_ident(&query, (*field)->field_name,
|
|
|
|
strlen((*field)->field_name), ident_quote_char);
|
2007-06-30 13:17:49 +02:00
|
|
|
query.append(STRING_WITH_LEN(", "));
|
2005-07-15 17:33:47 -07:00
|
|
|
}
|
2006-08-02 12:36:01 +02:00
|
|
|
/* chops off trailing comma */
|
|
|
|
query.length(query.length() - sizeof_trailing_comma);
|
|
|
|
|
2007-06-30 18:17:20 +02:00
|
|
|
query.append(STRING_WITH_LEN(" FROM "));
|
2007-06-28 00:23:14 -07:00
|
|
|
|
|
|
|
append_ident(&query, tmp_share.table_name,
|
|
|
|
tmp_share.table_name_length, ident_quote_char);
|
2005-07-15 17:33:47 -07:00
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
if (!(share= (FEDERATED_SHARE *) memdup_root(&mem_root, (char*)&tmp_share, sizeof(*share))) ||
|
2007-06-30 13:17:49 +02:00
|
|
|
!(share->select_query= (char*) strmake_root(&mem_root, query.ptr(), query.length() + 1)))
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
goto error;
|
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
share->use_count= 0;
|
2007-03-24 01:18:19 -07:00
|
|
|
share->mem_root= mem_root;
|
|
|
|
|
2005-03-15 18:41:06 -08:00
|
|
|
DBUG_PRINT("info",
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
("share->select_query %s", share->select_query));
|
2005-07-15 17:33:47 -07:00
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
if (my_hash_insert(&federated_open_tables, (uchar*) share))
|
2004-12-11 12:03:51 -08:00
|
|
|
goto error;
|
|
|
|
thr_lock_init(&share->lock);
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
pthread_mutex_init(&share->mutex, MY_MUTEX_INIT_FAST);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
2007-03-24 01:18:19 -07:00
|
|
|
else
|
|
|
|
free_root(&mem_root, MYF(0)); /* prevents memory leak */
|
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
share->use_count++;
|
|
|
|
pthread_mutex_unlock(&federated_mutex);
|
|
|
|
|
2007-03-24 01:18:19 -07:00
|
|
|
DBUG_RETURN(share);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
error:
|
|
|
|
pthread_mutex_unlock(&federated_mutex);
|
2007-03-24 01:18:19 -07:00
|
|
|
free_root(&mem_root, MYF(0));
|
|
|
|
DBUG_RETURN(NULL);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
2005-02-23 00:38:29 -08:00
|
|
|
Free lock controls. We call this whenever we close a table.
|
2004-12-11 12:03:51 -08:00
|
|
|
If the table had the last reference to the share then we
|
|
|
|
free memory associated with it.
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
static int free_share(FEDERATED_SHARE *share)
|
|
|
|
{
|
2007-03-24 01:18:19 -07:00
|
|
|
MEM_ROOT mem_root= share->mem_root;
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_ENTER("free_share");
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
|
2005-10-08 00:57:40 +03:00
|
|
|
pthread_mutex_lock(&federated_mutex);
|
2004-12-11 12:03:51 -08:00
|
|
|
if (!--share->use_count)
|
|
|
|
{
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
hash_delete(&federated_open_tables, (uchar*) share);
|
2004-12-11 12:03:51 -08:00
|
|
|
thr_lock_delete(&share->lock);
|
2005-02-16 23:07:10 -08:00
|
|
|
VOID(pthread_mutex_destroy(&share->mutex));
|
2007-03-24 01:18:19 -07:00
|
|
|
free_root(&mem_root, MYF(0));
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
pthread_mutex_unlock(&federated_mutex);
|
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_RETURN(0);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
ha_rows ha_federated::records_in_range(uint inx, key_range *start_key,
|
2006-12-15 00:51:37 +02:00
|
|
|
key_range *end_key)
|
2005-07-15 17:33:47 -07:00
|
|
|
{
|
|
|
|
/*
|
|
|
|
|
2005-07-17 04:10:19 +02:00
|
|
|
We really want indexes to be used as often as possible, therefore
|
|
|
|
we just need to hard-code the return value to a very low number to
|
|
|
|
force the issue
|
2005-07-15 17:33:47 -07:00
|
|
|
|
|
|
|
*/
|
|
|
|
DBUG_ENTER("ha_federated::records_in_range");
|
|
|
|
DBUG_RETURN(FEDERATED_RECORDS_IN_RANGE);
|
|
|
|
}
|
2004-12-11 12:03:51 -08:00
|
|
|
/*
|
|
|
|
If frm_error() is called then we will use this to to find out
|
2005-02-23 00:38:29 -08:00
|
|
|
what file extentions exist for the storage engine. This is
|
|
|
|
also used by the default rename_table and delete_table method
|
2004-12-11 12:03:51 -08:00
|
|
|
in handler.cc.
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
const char **ha_federated::bas_ext() const
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
static const char *ext[]=
|
|
|
|
{
|
|
|
|
NullS
|
|
|
|
};
|
|
|
|
return ext;
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
}
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
Used for opening tables. The name will be the name of the file.
|
|
|
|
A table is opened when it needs to be opened. For instance
|
|
|
|
when a request comes in for a select on the table (tables are not
|
|
|
|
open and closed for each request, they are cached).
|
|
|
|
|
|
|
|
Called from handler.cc by handler::ha_open(). The server opens
|
|
|
|
all tables by calling ha_open() which then calls the handler
|
|
|
|
specific open().
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
int ha_federated::open(const char *name, int mode, uint test_if_locked)
|
|
|
|
{
|
2005-02-23 00:38:29 -08:00
|
|
|
DBUG_ENTER("ha_federated::open");
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
if (!(share= get_share(name, table)))
|
|
|
|
DBUG_RETURN(1);
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
thr_lock_data_init(&share->lock, &lock, NULL);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2007-07-25 12:23:24 -07:00
|
|
|
DBUG_ASSERT(mysql == NULL);
|
2005-11-09 05:53:34 -08:00
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
ref_length= (table->s->primary_key != MAX_KEY ?
|
|
|
|
table->key_info[table->s->primary_key].key_length :
|
|
|
|
table->s->reclength);
|
|
|
|
DBUG_PRINT("info", ("ref_length: %u", ref_length));
|
|
|
|
|
2007-06-28 00:23:14 -07:00
|
|
|
reset();
|
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
Closes a table. We call the free_share() function to free any resources
|
|
|
|
that we have allocated in the "shared" structure.
|
|
|
|
|
|
|
|
Called from sql_base.cc, sql_select.cc, and table.cc.
|
|
|
|
In sql_select.cc it is only used to close up temporary tables or during
|
|
|
|
the process where a temporary table is converted over to being a
|
|
|
|
myisam table.
|
|
|
|
For sql_base.cc look at close_data_tables().
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
int ha_federated::close(void)
|
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
int retval;
|
2004-12-11 12:03:51 -08:00
|
|
|
DBUG_ENTER("ha_federated::close");
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
|
2005-02-17 16:17:21 -08:00
|
|
|
/* free the result set */
|
2005-07-17 04:10:19 +02:00
|
|
|
if (stored_result)
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
{
|
2005-07-17 04:10:19 +02:00
|
|
|
mysql_free_result(stored_result);
|
|
|
|
stored_result= 0;
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
}
|
2004-12-11 12:03:51 -08:00
|
|
|
/* Disconnect from mysql */
|
2007-07-25 12:23:24 -07:00
|
|
|
mysql_close(mysql);
|
|
|
|
mysql= NULL;
|
2005-07-15 17:33:47 -07:00
|
|
|
retval= free_share(share);
|
|
|
|
DBUG_RETURN(retval);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
Checks if a field in a record is SQL NULL.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
field_in_record_is_null()
|
|
|
|
table TABLE pointer, MySQL table object
|
|
|
|
field Field pointer, MySQL field object
|
|
|
|
record char pointer, contains record
|
|
|
|
|
|
|
|
DESCRIPTION
|
|
|
|
This method uses the record format information in table to track
|
2005-02-23 00:38:29 -08:00
|
|
|
the null bit in record.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
RETURN VALUE
|
|
|
|
1 if NULL
|
2005-02-23 00:38:29 -08:00
|
|
|
0 otherwise
|
2004-12-11 12:03:51 -08:00
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2006-11-10 17:22:43 +02:00
|
|
|
static inline uint field_in_record_is_null(TABLE *table,
|
2005-02-17 16:17:21 -08:00
|
|
|
Field *field,
|
|
|
|
char *record)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
|
|
|
int null_offset;
|
|
|
|
DBUG_ENTER("ha_federated::field_in_record_is_null");
|
|
|
|
|
|
|
|
if (!field->null_ptr)
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
|
2005-02-16 23:07:10 -08:00
|
|
|
null_offset= (uint) ((char*)field->null_ptr - (char*)table->record[0]);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
if (record[null_offset] & field->null_bit)
|
|
|
|
DBUG_RETURN(1);
|
|
|
|
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
2005-11-23 22:45:02 +02:00
|
|
|
|
2007-06-28 16:03:01 -07:00
|
|
|
/**
|
|
|
|
@brief Construct the INSERT statement.
|
|
|
|
|
|
|
|
@details This method will construct the INSERT statement and appends it to
|
|
|
|
the supplied query string buffer.
|
|
|
|
|
|
|
|
@return
|
|
|
|
@retval FALSE No error
|
|
|
|
@retval TRUE Failure
|
2004-12-11 12:03:51 -08:00
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2007-06-28 16:03:01 -07:00
|
|
|
bool ha_federated::append_stmt_insert(String *query)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2007-06-28 16:03:01 -07:00
|
|
|
char insert_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
Field **field;
|
|
|
|
uint tmp_length;
|
2007-06-30 18:17:20 +02:00
|
|
|
bool added_field= FALSE;
|
2006-08-08 09:01:32 -07:00
|
|
|
|
2007-06-28 16:03:01 -07:00
|
|
|
/* The main insert query string */
|
|
|
|
String insert_string(insert_buffer, sizeof(insert_buffer), &my_charset_bin);
|
|
|
|
DBUG_ENTER("ha_federated::append_stmt_insert");
|
2006-08-08 09:01:32 -07:00
|
|
|
|
2007-06-28 16:03:01 -07:00
|
|
|
insert_string.length(0);
|
2006-08-08 09:01:32 -07:00
|
|
|
|
2007-06-28 16:03:01 -07:00
|
|
|
if (replace_duplicates)
|
|
|
|
insert_string.append(STRING_WITH_LEN("REPLACE INTO "));
|
|
|
|
else if (ignore_duplicates && !insert_dup_update)
|
|
|
|
insert_string.append(STRING_WITH_LEN("INSERT IGNORE INTO "));
|
|
|
|
else
|
|
|
|
insert_string.append(STRING_WITH_LEN("INSERT INTO "));
|
|
|
|
append_ident(&insert_string, share->table_name, share->table_name_length,
|
|
|
|
ident_quote_char);
|
2007-06-30 18:17:20 +02:00
|
|
|
tmp_length= insert_string.length();
|
|
|
|
insert_string.append(STRING_WITH_LEN(" ("));
|
2006-08-08 09:01:32 -07:00
|
|
|
|
2007-06-28 16:03:01 -07:00
|
|
|
/*
|
|
|
|
loop through the field pointer array, add any fields to both the values
|
|
|
|
list and the fields list that match the current query id
|
|
|
|
*/
|
|
|
|
for (field= table->field; *field; field++)
|
|
|
|
{
|
2007-06-30 13:17:49 +02:00
|
|
|
if (bitmap_is_set(table->write_set, (*field)->field_index))
|
|
|
|
{
|
|
|
|
/* append the field name */
|
|
|
|
append_ident(&insert_string, (*field)->field_name,
|
|
|
|
strlen((*field)->field_name), ident_quote_char);
|
2006-08-08 09:01:32 -07:00
|
|
|
|
2007-06-30 13:17:49 +02:00
|
|
|
/* append commas between both fields and fieldnames */
|
|
|
|
/*
|
|
|
|
unfortunately, we can't use the logic if *(fields + 1) to
|
|
|
|
make the following appends conditional as we don't know if the
|
|
|
|
next field is in the write set
|
|
|
|
*/
|
|
|
|
insert_string.append(STRING_WITH_LEN(", "));
|
2007-06-30 18:17:20 +02:00
|
|
|
added_field= TRUE;
|
2007-06-30 13:17:49 +02:00
|
|
|
}
|
2007-06-28 16:03:01 -07:00
|
|
|
}
|
2006-08-08 09:01:32 -07:00
|
|
|
|
2007-06-30 18:17:20 +02:00
|
|
|
if (added_field)
|
|
|
|
{
|
|
|
|
/* Remove trailing comma. */
|
|
|
|
insert_string.length(insert_string.length() - sizeof_trailing_comma);
|
|
|
|
insert_string.append(STRING_WITH_LEN(") "));
|
|
|
|
}
|
|
|
|
else
|
2007-06-28 16:03:01 -07:00
|
|
|
{
|
2007-06-30 18:17:20 +02:00
|
|
|
/* If there were no fields, we don't want to add a closing paren. */
|
|
|
|
insert_string.length(tmp_length);
|
2007-06-28 16:03:01 -07:00
|
|
|
}
|
2006-08-08 09:01:32 -07:00
|
|
|
|
2007-06-30 13:17:49 +02:00
|
|
|
insert_string.append(STRING_WITH_LEN(" VALUES "));
|
2006-08-08 09:01:32 -07:00
|
|
|
|
2007-06-28 16:03:01 -07:00
|
|
|
DBUG_RETURN(query->append(insert_string));
|
|
|
|
}
|
2006-08-08 09:01:32 -07:00
|
|
|
|
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
/*
|
|
|
|
write_row() inserts a row. No extra() hint is given currently if a bulk load
|
|
|
|
is happeneding. buf() is a byte array of data. You can use the field
|
|
|
|
information to extract the data from the native byte array type.
|
|
|
|
Example of this would be:
|
|
|
|
for (Field **field=table->field ; *field ; field++)
|
|
|
|
{
|
|
|
|
...
|
|
|
|
}
|
2006-08-08 09:01:32 -07:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
Called from item_sum.cc, item_sum.cc, sql_acl.cc, sql_insert.cc,
|
|
|
|
sql_insert.cc, sql_select.cc, sql_table.cc, sql_udf.cc, and sql_update.cc.
|
|
|
|
*/
|
2006-08-08 09:01:32 -07:00
|
|
|
|
2007-06-30 13:17:49 +02:00
|
|
|
int ha_federated::write_row(uchar *buf)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
char values_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
char insert_field_value_buffer[STRING_BUFFER_USUAL_SIZE];
|
2005-01-20 23:23:15 +01:00
|
|
|
Field **field;
|
2007-06-28 16:03:01 -07:00
|
|
|
uint tmp_length;
|
|
|
|
int error= 0;
|
|
|
|
bool use_bulk_insert;
|
2007-06-29 15:14:08 -07:00
|
|
|
bool auto_increment_update_required= (table->next_number_field != NULL);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-02-17 16:17:21 -08:00
|
|
|
/* The string containing the values to be added to the insert */
|
2004-12-11 12:03:51 -08:00
|
|
|
String values_string(values_buffer, sizeof(values_buffer), &my_charset_bin);
|
2005-02-17 16:17:21 -08:00
|
|
|
/* The actual value of the field, to be added to the values_string */
|
2004-12-11 12:03:51 -08:00
|
|
|
String insert_field_value_string(insert_field_value_buffer,
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
sizeof(insert_field_value_buffer),
|
|
|
|
&my_charset_bin);
|
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 18:52:22 +03:00
|
|
|
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
|
2004-12-11 12:03:51 -08:00
|
|
|
DBUG_ENTER("ha_federated::write_row");
|
2005-02-17 16:17:21 -08:00
|
|
|
|
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 18:52:22 +03:00
|
|
|
values_string.length(0);
|
|
|
|
insert_field_value_string.length(0);
|
2007-03-02 08:43:45 -08:00
|
|
|
ha_statistic_increment(&SSV::ha_write_count);
|
2004-12-11 12:03:51 -08:00
|
|
|
if (table->timestamp_field_type & TIMESTAMP_AUTO_SET_ON_INSERT)
|
|
|
|
table->timestamp_field->set_time();
|
|
|
|
|
2005-07-17 04:10:19 +02:00
|
|
|
/*
|
|
|
|
start both our field and field values strings
|
2007-06-28 16:03:01 -07:00
|
|
|
We must disable multi-row insert for "INSERT...ON DUPLICATE KEY UPDATE"
|
|
|
|
Ignore duplicates is always true when insert_dup_update is true.
|
|
|
|
When replace_duplicates == TRUE, we can safely enable multi-row insert.
|
|
|
|
When performing multi-row insert, we only collect the columns values for
|
|
|
|
the row. The start of the statement is only created when the first
|
|
|
|
row is copied in to the bulk_insert string.
|
2005-07-17 04:10:19 +02:00
|
|
|
*/
|
2007-06-28 16:03:01 -07:00
|
|
|
if (!(use_bulk_insert= bulk_insert.str &&
|
|
|
|
(!insert_dup_update || replace_duplicates)))
|
|
|
|
append_stmt_insert(&values_string);
|
2005-07-17 04:10:19 +02:00
|
|
|
|
2006-08-08 09:01:32 -07:00
|
|
|
values_string.append(STRING_WITH_LEN(" ("));
|
2007-06-28 16:03:01 -07:00
|
|
|
tmp_length= values_string.length();
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
/*
|
2005-02-23 00:38:29 -08:00
|
|
|
loop through the field pointer array, add any fields to both the values
|
2005-04-28 14:45:27 +02:00
|
|
|
list and the fields list that is part of the write set
|
2004-12-11 12:03:51 -08:00
|
|
|
*/
|
2005-07-15 17:33:47 -07:00
|
|
|
for (field= table->field; *field; field++)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
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 18:52:22 +03:00
|
|
|
if (bitmap_is_set(table->write_set, (*field)->field_index))
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
|
|
|
if ((*field)->is_null())
|
2006-08-11 12:19:41 -07:00
|
|
|
values_string.append(STRING_WITH_LEN(" NULL "));
|
2004-12-11 12:03:51 -08:00
|
|
|
else
|
|
|
|
{
|
2006-08-11 12:19:41 -07:00
|
|
|
bool needs_quote= (*field)->str_needs_quotes();
|
2004-12-11 12:03:51 -08:00
|
|
|
(*field)->val_str(&insert_field_value_string);
|
2006-08-11 12:19:41 -07:00
|
|
|
if (needs_quote)
|
2007-06-30 13:17:49 +02:00
|
|
|
values_string.append(value_quote_char);
|
2006-08-08 09:01:32 -07:00
|
|
|
insert_field_value_string.print(&values_string);
|
2006-08-11 12:19:41 -07:00
|
|
|
if (needs_quote)
|
2007-06-30 13:17:49 +02:00
|
|
|
values_string.append(value_quote_char);
|
2006-07-18 18:41:36 -07:00
|
|
|
|
2006-08-08 09:01:32 -07:00
|
|
|
insert_field_value_string.length(0);
|
2005-05-02 15:45:33 +02:00
|
|
|
}
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-02-17 16:17:21 -08:00
|
|
|
/* append commas between both fields and fieldnames */
|
2005-07-15 17:33:47 -07:00
|
|
|
/*
|
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 18:52:22 +03:00
|
|
|
unfortunately, we can't use the logic if *(fields + 1) to
|
|
|
|
make the following appends conditional as we don't know if the
|
|
|
|
next field is in the write set
|
2005-07-15 17:33:47 -07:00
|
|
|
*/
|
2006-08-02 12:36:01 +02:00
|
|
|
values_string.append(STRING_WITH_LEN(", "));
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
}
|
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 18:52:22 +03:00
|
|
|
dbug_tmp_restore_column_map(table->read_set, old_map);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
/*
|
2005-02-23 00:38:29 -08:00
|
|
|
if there were no fields, we don't want to add a closing paren
|
|
|
|
AND, we don't want to chop off the last char '('
|
|
|
|
insert will be "INSERT INTO t1 VALUES ();"
|
2004-12-11 12:03:51 -08:00
|
|
|
*/
|
2007-06-28 16:03:01 -07:00
|
|
|
if (values_string.length() > tmp_length)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2007-06-30 13:17:49 +02:00
|
|
|
/* chops off trailing comma */
|
2006-08-02 12:36:01 +02:00
|
|
|
values_string.length(values_string.length() - sizeof_trailing_comma);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
2005-02-17 16:17:21 -08:00
|
|
|
/* we always want to append this, even if there aren't any fields */
|
2006-08-02 12:36:01 +02:00
|
|
|
values_string.append(STRING_WITH_LEN(") "));
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
|
2007-06-28 16:03:01 -07:00
|
|
|
if (use_bulk_insert)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
Send the current bulk insert out if appending the current row would
|
|
|
|
cause the statement to overflow the packet size, otherwise set
|
|
|
|
auto_increment_update_required to FALSE as no query was executed.
|
|
|
|
*/
|
|
|
|
if (bulk_insert.length + values_string.length() + bulk_padding >
|
|
|
|
mysql->net.max_packet_size && bulk_insert.length)
|
|
|
|
{
|
2007-07-25 12:23:24 -07:00
|
|
|
error= real_query(bulk_insert.str, bulk_insert.length);
|
2007-06-28 16:03:01 -07:00
|
|
|
bulk_insert.length= 0;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
auto_increment_update_required= FALSE;
|
|
|
|
|
|
|
|
if (bulk_insert.length == 0)
|
|
|
|
{
|
|
|
|
char insert_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
String insert_string(insert_buffer, sizeof(insert_buffer),
|
|
|
|
&my_charset_bin);
|
|
|
|
insert_string.length(0);
|
|
|
|
append_stmt_insert(&insert_string);
|
|
|
|
dynstr_append_mem(&bulk_insert, insert_string.ptr(),
|
|
|
|
insert_string.length());
|
|
|
|
}
|
|
|
|
else
|
|
|
|
dynstr_append_mem(&bulk_insert, ",", 1);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2007-06-28 16:03:01 -07:00
|
|
|
dynstr_append_mem(&bulk_insert, values_string.ptr(),
|
|
|
|
values_string.length());
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2007-07-25 12:23:24 -07:00
|
|
|
error= real_query(values_string.ptr(), values_string.length());
|
2007-06-28 16:03:01 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
if (error)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-09-13 23:31:17 -07:00
|
|
|
DBUG_RETURN(stash_remote_error());
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
2006-01-27 15:43:44 -08:00
|
|
|
/*
|
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 18:52:22 +03:00
|
|
|
If the table we've just written a record to contains an auto_increment
|
|
|
|
field, then store the last_insert_id() value from the foreign server
|
2006-01-27 15:43:44 -08:00
|
|
|
*/
|
2007-06-28 16:03:01 -07:00
|
|
|
if (auto_increment_update_required)
|
2007-07-23 23:35:43 -07:00
|
|
|
{
|
2006-01-27 15:43:44 -08:00
|
|
|
update_auto_increment();
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2007-07-23 23:35:43 -07:00
|
|
|
/* mysql_insert() uses this for protocol return value */
|
2007-07-24 07:23:59 -07:00
|
|
|
table->next_number_field->store(stats.auto_increment_value, 1);
|
2007-07-23 23:35:43 -07:00
|
|
|
}
|
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
2007-06-28 16:03:01 -07:00
|
|
|
|
|
|
|
/**
|
|
|
|
@brief Prepares the storage engine for bulk inserts.
|
|
|
|
|
|
|
|
@param[in] rows estimated number of rows in bulk insert
|
|
|
|
or 0 if unknown.
|
|
|
|
|
|
|
|
@details Initializes memory structures required for bulk insert.
|
|
|
|
*/
|
|
|
|
|
|
|
|
void ha_federated::start_bulk_insert(ha_rows rows)
|
|
|
|
{
|
|
|
|
uint page_size;
|
|
|
|
DBUG_ENTER("ha_federated::start_bulk_insert");
|
|
|
|
|
|
|
|
dynstr_free(&bulk_insert);
|
|
|
|
|
|
|
|
/**
|
|
|
|
We don't bother with bulk-insert semantics when the estimated rows == 1
|
|
|
|
The rows value will be 0 if the server does not know how many rows
|
|
|
|
would be inserted. This can occur when performing INSERT...SELECT
|
|
|
|
*/
|
|
|
|
|
|
|
|
if (rows == 1)
|
|
|
|
DBUG_VOID_RETURN;
|
|
|
|
|
2007-07-25 12:23:24 -07:00
|
|
|
/*
|
|
|
|
Make sure we have an open connection so that we know the
|
|
|
|
maximum packet size.
|
|
|
|
*/
|
|
|
|
if (!mysql && real_connect())
|
|
|
|
DBUG_VOID_RETURN;
|
|
|
|
|
2007-06-28 16:03:01 -07:00
|
|
|
page_size= (uint) my_getpagesize();
|
|
|
|
|
|
|
|
if (init_dynamic_string(&bulk_insert, NULL, page_size, page_size))
|
|
|
|
DBUG_VOID_RETURN;
|
|
|
|
|
|
|
|
bulk_insert.length= 0;
|
|
|
|
DBUG_VOID_RETURN;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
@brief End bulk insert.
|
|
|
|
|
|
|
|
@details This method will send any remaining rows to the remote server.
|
|
|
|
Finally, it will deinitialize the bulk insert data structure.
|
|
|
|
|
|
|
|
@return Operation status
|
|
|
|
@retval 0 No error
|
|
|
|
@retval != 0 Error occured at remote server. Also sets my_errno.
|
|
|
|
*/
|
|
|
|
|
|
|
|
int ha_federated::end_bulk_insert()
|
|
|
|
{
|
|
|
|
int error= 0;
|
|
|
|
DBUG_ENTER("ha_federated::end_bulk_insert");
|
|
|
|
|
|
|
|
if (bulk_insert.str && bulk_insert.length)
|
|
|
|
{
|
2007-07-25 12:23:24 -07:00
|
|
|
if (real_query(bulk_insert.str, bulk_insert.length))
|
2007-06-28 16:03:01 -07:00
|
|
|
error= stash_remote_error();
|
|
|
|
else
|
|
|
|
if (table->next_number_field)
|
|
|
|
update_auto_increment();
|
|
|
|
}
|
|
|
|
|
|
|
|
dynstr_free(&bulk_insert);
|
|
|
|
|
|
|
|
DBUG_RETURN(my_errno= error);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2006-01-27 15:43:44 -08:00
|
|
|
/*
|
|
|
|
ha_federated::update_auto_increment
|
|
|
|
|
|
|
|
This method ensures that last_insert_id() works properly. What it simply does
|
|
|
|
is calls last_insert_id() on the foreign database immediately after insert
|
|
|
|
(if the table has an auto_increment field) and sets the insert id via
|
WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
2006-07-09 17:52:19 +02:00
|
|
|
thd->insert_id(ID)).
|
2006-01-27 15:43:44 -08:00
|
|
|
*/
|
|
|
|
void ha_federated::update_auto_increment(void)
|
|
|
|
{
|
|
|
|
THD *thd= current_thd;
|
|
|
|
DBUG_ENTER("ha_federated::update_auto_increment");
|
|
|
|
|
2007-07-23 23:35:43 -07:00
|
|
|
ha_federated::info(HA_STATUS_AUTO);
|
WL#3146 "less locking in auto_increment":
this is a cleanup patch for our current auto_increment handling:
new names for auto_increment variables in THD, new methods to manipulate them
(see sql_class.h), some move into handler::, causing less backup/restore
work when executing substatements.
This makes the logic hopefully clearer, less work is is needed in
mysql_insert().
By cleaning up, using different variables for different purposes (instead
of one for 3 things...), we fix those bugs, which someone may want to fix
in 5.0 too:
BUG#20339 "stored procedure using LAST_INSERT_ID() does not replicate
statement-based"
BUG#20341 "stored function inserting into one auto_increment puts bad
data in slave"
BUG#19243 "wrong LAST_INSERT_ID() after ON DUPLICATE KEY UPDATE"
(now if a row is updated, LAST_INSERT_ID() will return its id)
and re-fixes:
BUG#6880 "LAST_INSERT_ID() value changes during multi-row INSERT"
(already fixed differently by Ramil in 4.1)
Test of documented behaviour of mysql_insert_id() (there was no test).
The behaviour changes introduced are:
- LAST_INSERT_ID() now returns "the first autogenerated auto_increment value
successfully inserted", instead of "the first autogenerated auto_increment
value if any row was successfully inserted", see auto_increment.test.
Same for mysql_insert_id(), see mysql_client_test.c.
- LAST_INSERT_ID() returns the id of the updated row if ON DUPLICATE KEY
UPDATE, see auto_increment.test. Same for mysql_insert_id(), see
mysql_client_test.c.
- LAST_INSERT_ID() does not change if no autogenerated value was successfully
inserted (it used to then be 0), see auto_increment.test.
- if in INSERT SELECT no autogenerated value was successfully inserted,
mysql_insert_id() now returns the id of the last inserted row (it already
did this for INSERT VALUES), see mysql_client_test.c.
- if INSERT SELECT uses LAST_INSERT_ID(X), mysql_insert_id() now returns X
(it already did this for INSERT VALUES), see mysql_client_test.c.
- NDB now behaves like other engines wrt SET INSERT_ID: with INSERT IGNORE,
the id passed in SET INSERT_ID is re-used until a row succeeds; SET INSERT_ID
influences not only the first row now.
Additionally, when unlocking a table we check that the thread is not keeping
a next_insert_id (as the table is unlocked that id is potentially out-of-date);
forgetting about this next_insert_id is done in a new
handler::ha_release_auto_increment().
Finally we prepare for engines capable of reserving finite-length intervals
of auto_increment values: we store such intervals in THD. The next step
(to be done by the replication team in 5.1) is to read those intervals from
THD and actually store them in the statement-based binary log. NDB
will be a good engine to test that.
2006-07-09 17:52:19 +02:00
|
|
|
thd->first_successful_insert_id_in_cur_stmt=
|
2007-07-24 00:03:01 -07:00
|
|
|
stats.auto_increment_value;
|
2006-11-22 14:11:36 +02:00
|
|
|
DBUG_PRINT("info",("last_insert_id: %ld", (long) stats.auto_increment_value));
|
2006-01-27 15:43:44 -08:00
|
|
|
|
|
|
|
DBUG_VOID_RETURN;
|
|
|
|
}
|
2005-03-15 18:41:06 -08:00
|
|
|
|
2005-07-19 03:04:51 +02:00
|
|
|
int ha_federated::optimize(THD* thd, HA_CHECK_OPT* check_opt)
|
|
|
|
{
|
|
|
|
char query_buffer[STRING_BUFFER_USUAL_SIZE];
|
|
|
|
String query(query_buffer, sizeof(query_buffer), &my_charset_bin);
|
|
|
|
DBUG_ENTER("ha_federated::optimize");
|
|
|
|
|
|
|
|
query.length(0);
|
|
|
|
|
|
|
|
query.set_charset(system_charset_info);
|
2007-06-30 13:17:49 +02:00
|
|
|
query.append(STRING_WITH_LEN("OPTIMIZE TABLE "));
|
2007-06-28 00:23:14 -07:00
|
|
|
append_ident(&query, share->table_name, share->table_name_length,
|
|
|
|
ident_quote_char);
|
2005-07-19 03:04:51 +02:00
|
|
|
|
2007-07-25 12:23:24 -07:00
|
|
|
if (real_query(query.ptr(), query.length()))
|
2005-07-19 03:04:51 +02:00
|
|
|
{
|
2005-09-13 23:31:17 -07:00
|
|
|
DBUG_RETURN(stash_remote_error());
|
2005-07-19 03:04:51 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int ha_federated::repair(THD* thd, HA_CHECK_OPT* check_opt)
|
|
|
|
{
|
|
|
|
char query_buffer[STRING_BUFFER_USUAL_SIZE];
|
|
|
|
String query(query_buffer, sizeof(query_buffer), &my_charset_bin);
|
|
|
|
DBUG_ENTER("ha_federated::repair");
|
|
|
|
|
|
|
|
query.length(0);
|
|
|
|
|
|
|
|
query.set_charset(system_charset_info);
|
2007-06-30 13:17:49 +02:00
|
|
|
query.append(STRING_WITH_LEN("REPAIR TABLE "));
|
2007-06-28 00:23:14 -07:00
|
|
|
append_ident(&query, share->table_name, share->table_name_length,
|
|
|
|
ident_quote_char);
|
2005-07-19 03:04:51 +02:00
|
|
|
if (check_opt->flags & T_QUICK)
|
2006-08-02 12:36:01 +02:00
|
|
|
query.append(STRING_WITH_LEN(" QUICK"));
|
2005-07-19 03:04:51 +02:00
|
|
|
if (check_opt->flags & T_EXTEND)
|
2006-08-02 12:36:01 +02:00
|
|
|
query.append(STRING_WITH_LEN(" EXTENDED"));
|
2005-07-19 03:04:51 +02:00
|
|
|
if (check_opt->sql_flags & TT_USEFRM)
|
2006-08-02 12:36:01 +02:00
|
|
|
query.append(STRING_WITH_LEN(" USE_FRM"));
|
2006-06-27 23:49:48 -07:00
|
|
|
|
2007-07-25 12:23:24 -07:00
|
|
|
if (real_query(query.ptr(), query.length()))
|
2005-07-19 03:04:51 +02:00
|
|
|
{
|
2005-09-13 23:31:17 -07:00
|
|
|
DBUG_RETURN(stash_remote_error());
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
2005-03-15 18:41:06 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
/*
|
|
|
|
Yes, update_row() does what you expect, it updates a row. old_data will have
|
|
|
|
the previous row record in it, while new_data will have the newest data in
|
|
|
|
it.
|
|
|
|
|
|
|
|
Keep in mind that the server can do updates based on ordering if an ORDER BY
|
2006-07-21 10:43:10 +02:00
|
|
|
clause was used. Consecutive ordering is not guaranteed.
|
2004-12-11 12:03:51 -08:00
|
|
|
Currently new_data will not have an updated auto_increament record, or
|
|
|
|
and updated timestamp field. You can do these for federated by doing these:
|
|
|
|
if (table->timestamp_on_update_now)
|
|
|
|
update_timestamp(new_row+table->timestamp_on_update_now-1);
|
|
|
|
if (table->next_number_field && record == table->record[0])
|
|
|
|
update_auto_increment();
|
|
|
|
|
|
|
|
Called from sql_select.cc, sql_acl.cc, sql_update.cc, and sql_insert.cc.
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
int ha_federated::update_row(const uchar *old_data, uchar *new_data)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
/*
|
2005-10-08 00:57:40 +03:00
|
|
|
This used to control how the query was built. If there was a
|
|
|
|
primary key, the query would be built such that there was a where
|
|
|
|
clause with only that column as the condition. This is flawed,
|
|
|
|
because if we have a multi-part primary key, it would only use the
|
|
|
|
first part! We don't need to do this anyway, because
|
|
|
|
read_range_first will retrieve the correct record, which is what
|
|
|
|
is used to build the WHERE clause. We can however use this to
|
|
|
|
append a LIMIT to the end if there is NOT a primary key. Why do
|
|
|
|
this? Because we only are updating one record, and LIMIT enforces
|
|
|
|
this.
|
2005-07-15 17:33:47 -07:00
|
|
|
*/
|
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 18:52:22 +03:00
|
|
|
bool has_a_primary_key= test(table->s->primary_key != MAX_KEY);
|
2006-08-08 09:01:32 -07:00
|
|
|
|
2006-07-21 10:43:10 +02:00
|
|
|
/*
|
2005-07-15 17:33:47 -07:00
|
|
|
buffers for following strings
|
|
|
|
*/
|
2006-06-28 12:11:52 -07:00
|
|
|
char field_value_buffer[STRING_BUFFER_USUAL_SIZE];
|
2005-07-15 17:33:47 -07:00
|
|
|
char update_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
char where_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2006-06-28 12:11:52 -07:00
|
|
|
/* Work area for field values */
|
|
|
|
String field_value(field_value_buffer, sizeof(field_value_buffer),
|
|
|
|
&my_charset_bin);
|
2005-02-17 16:17:21 -08:00
|
|
|
/* stores the update query */
|
2005-07-15 17:33:47 -07:00
|
|
|
String update_string(update_buffer,
|
|
|
|
sizeof(update_buffer),
|
|
|
|
&my_charset_bin);
|
2005-02-23 00:38:29 -08:00
|
|
|
/* stores the WHERE clause */
|
2005-07-15 17:33:47 -07:00
|
|
|
String where_string(where_buffer,
|
|
|
|
sizeof(where_buffer),
|
|
|
|
&my_charset_bin);
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
uchar *record= table->record[0];
|
2004-12-11 12:03:51 -08:00
|
|
|
DBUG_ENTER("ha_federated::update_row");
|
2006-07-21 10:43:10 +02:00
|
|
|
/*
|
2005-07-15 17:33:47 -07:00
|
|
|
set string lengths to 0 to avoid misc chars in string
|
|
|
|
*/
|
2006-06-28 12:11:52 -07:00
|
|
|
field_value.length(0);
|
2005-03-15 00:32:21 -08:00
|
|
|
update_string.length(0);
|
|
|
|
where_string.length(0);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2007-06-28 00:23:14 -07:00
|
|
|
if (ignore_duplicates)
|
|
|
|
update_string.append(STRING_WITH_LEN("UPDATE IGNORE "));
|
|
|
|
else
|
|
|
|
update_string.append(STRING_WITH_LEN("UPDATE "));
|
|
|
|
append_ident(&update_string, share->table_name,
|
|
|
|
share->table_name_length, ident_quote_char);
|
2007-06-30 13:17:49 +02:00
|
|
|
update_string.append(STRING_WITH_LEN(" SET "));
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-11-23 22:45:02 +02:00
|
|
|
/*
|
|
|
|
In this loop, we want to match column names to values being inserted
|
|
|
|
(while building INSERT statement).
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2006-07-20 21:28:55 +02:00
|
|
|
Iterate through table->field (new data) and share->old_field (old_data)
|
|
|
|
using the same index to create an SQL UPDATE statement. New data is
|
2005-11-23 22:45:02 +02:00
|
|
|
used to create SET field=value and old data is used to create WHERE
|
|
|
|
field=oldvalue
|
|
|
|
*/
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
for (Field **field= table->field; *field; field++)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
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 18:52:22 +03:00
|
|
|
if (bitmap_is_set(table->write_set, (*field)->field_index))
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2007-06-30 13:17:49 +02:00
|
|
|
uint field_name_length= strlen((*field)->field_name);
|
|
|
|
append_ident(&update_string, (*field)->field_name, field_name_length,
|
|
|
|
ident_quote_char);
|
2006-08-08 09:01:32 -07:00
|
|
|
update_string.append(STRING_WITH_LEN(" = "));
|
2006-07-21 10:43:10 +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 18:52:22 +03:00
|
|
|
if ((*field)->is_null())
|
2006-08-03 19:10:18 +02:00
|
|
|
update_string.append(STRING_WITH_LEN(" NULL "));
|
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 18:52:22 +03:00
|
|
|
else
|
|
|
|
{
|
|
|
|
/* otherwise = */
|
2006-08-11 12:19:41 -07:00
|
|
|
my_bitmap_map *old_map= tmp_use_all_columns(table, table->read_set);
|
|
|
|
bool needs_quote= (*field)->str_needs_quotes();
|
2006-08-08 09:01:32 -07:00
|
|
|
(*field)->val_str(&field_value);
|
2006-08-11 12:19:41 -07:00
|
|
|
if (needs_quote)
|
2007-06-30 13:17:49 +02:00
|
|
|
update_string.append(value_quote_char);
|
2006-08-08 09:01:32 -07:00
|
|
|
field_value.print(&update_string);
|
2006-08-11 12:19:41 -07:00
|
|
|
if (needs_quote)
|
2007-06-30 13:17:49 +02:00
|
|
|
update_string.append(value_quote_char);
|
2006-07-20 21:28:55 +02:00
|
|
|
field_value.length(0);
|
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
2006-06-04 18:52:22 +03:00
|
|
|
tmp_restore_column_map(table->read_set, old_map);
|
|
|
|
}
|
2006-08-02 12:36:01 +02:00
|
|
|
update_string.append(STRING_WITH_LEN(", "));
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
2005-07-15 17:33:47 -07:00
|
|
|
|
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 18:52:22 +03:00
|
|
|
if (bitmap_is_set(table->read_set, (*field)->field_index))
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2007-06-30 13:17:49 +02:00
|
|
|
uint field_name_length= strlen((*field)->field_name);
|
|
|
|
append_ident(&where_string, (*field)->field_name, field_name_length,
|
|
|
|
ident_quote_char);
|
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 18:52:22 +03:00
|
|
|
if (field_in_record_is_null(table, *field, (char*) old_data))
|
2006-08-02 12:36:01 +02:00
|
|
|
where_string.append(STRING_WITH_LEN(" IS NULL "));
|
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 18:52:22 +03:00
|
|
|
else
|
|
|
|
{
|
2006-08-11 12:19:41 -07:00
|
|
|
bool needs_quote= (*field)->str_needs_quotes();
|
2006-08-02 12:36:01 +02:00
|
|
|
where_string.append(STRING_WITH_LEN(" = "));
|
2006-07-21 10:43:10 +02:00
|
|
|
(*field)->val_str(&field_value,
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
(old_data + (*field)->offset(record)));
|
2006-08-11 12:19:41 -07:00
|
|
|
if (needs_quote)
|
2007-06-30 13:17:49 +02:00
|
|
|
where_string.append(value_quote_char);
|
2006-08-08 09:01:32 -07:00
|
|
|
field_value.print(&where_string);
|
2006-08-11 12:19:41 -07:00
|
|
|
if (needs_quote)
|
2007-06-30 13:17:49 +02:00
|
|
|
where_string.append(value_quote_char);
|
2006-07-21 10:43:10 +02:00
|
|
|
field_value.length(0);
|
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
2006-06-04 18:52:22 +03:00
|
|
|
}
|
2006-08-02 12:36:01 +02:00
|
|
|
where_string.append(STRING_WITH_LEN(" AND "));
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
}
|
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 18:52:22 +03:00
|
|
|
|
|
|
|
/* Remove last ', '. This works as there must be at least on updated field */
|
2006-08-02 12:36:01 +02:00
|
|
|
update_string.length(update_string.length() - sizeof_trailing_comma);
|
2006-08-08 09:01:32 -07:00
|
|
|
|
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 18:52:22 +03:00
|
|
|
if (where_string.length())
|
|
|
|
{
|
2006-08-02 12:36:01 +02:00
|
|
|
/* chop off trailing AND */
|
|
|
|
where_string.length(where_string.length() - sizeof_trailing_and);
|
|
|
|
update_string.append(STRING_WITH_LEN(" WHERE "));
|
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 18:52:22 +03:00
|
|
|
update_string.append(where_string);
|
|
|
|
}
|
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
/*
|
|
|
|
If this table has not a primary key, then we could possibly
|
|
|
|
update multiple rows. We want to make sure to only update one!
|
|
|
|
*/
|
2005-07-17 04:10:19 +02:00
|
|
|
if (!has_a_primary_key)
|
2006-08-02 12:36:01 +02:00
|
|
|
update_string.append(STRING_WITH_LEN(" LIMIT 1"));
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2007-07-25 12:23:24 -07:00
|
|
|
if (real_query(update_string.ptr(), update_string.length()))
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-09-13 23:31:17 -07:00
|
|
|
DBUG_RETURN(stash_remote_error());
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2005-07-15 17:33:47 -07:00
|
|
|
This will delete a row. 'buf' will contain a copy of the row to be =deleted.
|
2004-12-11 12:03:51 -08:00
|
|
|
The server will call this right after the current row has been called (from
|
2006-06-27 23:49:48 -07:00
|
|
|
either a previous rnd_next() or index call).
|
2004-12-11 12:03:51 -08:00
|
|
|
If you keep a pointer to the last row or can access a primary key it will
|
|
|
|
make doing the deletion quite a bit easier.
|
|
|
|
Keep in mind that the server does no guarentee consecutive deletions.
|
|
|
|
ORDER BY clauses can be used.
|
|
|
|
|
|
|
|
Called in sql_acl.cc and sql_udf.cc to manage internal table information.
|
2005-02-23 00:38:29 -08:00
|
|
|
Called in sql_delete.cc, sql_insert.cc, and sql_select.cc. In sql_select
|
2004-12-11 12:03:51 -08:00
|
|
|
it is used for removing duplicates while in insert it is used for REPLACE
|
|
|
|
calls.
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
int ha_federated::delete_row(const uchar *buf)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
char delete_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
char data_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
2004-12-11 12:03:51 -08:00
|
|
|
String delete_string(delete_buffer, sizeof(delete_buffer), &my_charset_bin);
|
|
|
|
String data_string(data_buffer, sizeof(data_buffer), &my_charset_bin);
|
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 18:52:22 +03:00
|
|
|
uint found= 0;
|
2004-12-11 12:03:51 -08:00
|
|
|
DBUG_ENTER("ha_federated::delete_row");
|
|
|
|
|
2005-09-12 18:02:17 -07:00
|
|
|
delete_string.length(0);
|
2007-06-30 13:17:49 +02:00
|
|
|
delete_string.append(STRING_WITH_LEN("DELETE FROM "));
|
2007-06-28 00:23:14 -07:00
|
|
|
append_ident(&delete_string, share->table_name,
|
|
|
|
share->table_name_length, ident_quote_char);
|
2007-06-30 13:17:49 +02:00
|
|
|
delete_string.append(STRING_WITH_LEN(" WHERE "));
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-06-29 12:44:40 +03:00
|
|
|
for (Field **field= table->field; *field; field++)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-09-12 18:02:17 -07:00
|
|
|
Field *cur_field= *field;
|
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 18:52:22 +03:00
|
|
|
found++;
|
|
|
|
if (bitmap_is_set(table->read_set, cur_field->field_index))
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2007-06-30 13:17:49 +02:00
|
|
|
append_ident(&delete_string, (*field)->field_name,
|
|
|
|
strlen((*field)->field_name), ident_quote_char);
|
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 18:52:22 +03:00
|
|
|
data_string.length(0);
|
|
|
|
if (cur_field->is_null())
|
|
|
|
{
|
2006-08-02 12:36:01 +02:00
|
|
|
delete_string.append(STRING_WITH_LEN(" IS NULL "));
|
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 18:52:22 +03:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
2006-08-11 12:19:41 -07:00
|
|
|
bool needs_quote= cur_field->str_needs_quotes();
|
|
|
|
delete_string.append(STRING_WITH_LEN(" = "));
|
|
|
|
cur_field->val_str(&data_string);
|
|
|
|
if (needs_quote)
|
2007-06-30 13:17:49 +02:00
|
|
|
delete_string.append(value_quote_char);
|
2006-08-11 12:19:41 -07:00
|
|
|
data_string.print(&delete_string);
|
|
|
|
if (needs_quote)
|
2007-06-30 13:17:49 +02:00
|
|
|
delete_string.append(value_quote_char);
|
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 18:52:22 +03:00
|
|
|
}
|
2006-08-02 12:36:01 +02:00
|
|
|
delete_string.append(STRING_WITH_LEN(" AND "));
|
2007-06-30 18:17:20 +02:00
|
|
|
}
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
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 18:52:22 +03:00
|
|
|
|
2006-08-02 12:36:01 +02:00
|
|
|
// Remove trailing AND
|
|
|
|
delete_string.length(delete_string.length() - sizeof_trailing_and);
|
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 18:52:22 +03:00
|
|
|
if (!found)
|
2006-08-02 12:36:01 +02:00
|
|
|
delete_string.length(delete_string.length() - sizeof_trailing_where);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2006-08-02 12:36:01 +02:00
|
|
|
delete_string.append(STRING_WITH_LEN(" LIMIT 1"));
|
2005-03-15 18:41:06 -08:00
|
|
|
DBUG_PRINT("info",
|
2004-12-11 12:03:51 -08:00
|
|
|
("Delete sql: %s", delete_string.c_ptr_quick()));
|
2007-07-25 12:23:24 -07:00
|
|
|
if (real_query(delete_string.ptr(), delete_string.length()))
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-09-13 23:31:17 -07:00
|
|
|
DBUG_RETURN(stash_remote_error());
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
2006-11-30 23:23:23 +02:00
|
|
|
stats.deleted+= (ha_rows) mysql->affected_rows;
|
|
|
|
stats.records-= (ha_rows) mysql->affected_rows;
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_PRINT("info",
|
2006-11-20 22:42:06 +02:00
|
|
|
("rows deleted %ld rows deleted for all time %ld",
|
2006-11-22 14:11:36 +02:00
|
|
|
(long) mysql->affected_rows, (long) stats.deleted));
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
Positions an index cursor to the index specified in the handle. Fetches the
|
|
|
|
row if available. If the key value is null, begin at the first key of the
|
|
|
|
index. This method, which is called in the case of an SQL statement having
|
|
|
|
a WHERE clause on a non-primary key index, simply calls index_read_idx.
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
int ha_federated::index_read(uchar *buf, const uchar *key,
|
2006-06-27 23:49:48 -07:00
|
|
|
uint key_len, ha_rkey_function find_flag)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::index_read");
|
2006-06-27 23:49:48 -07:00
|
|
|
|
|
|
|
if (stored_result)
|
|
|
|
mysql_free_result(stored_result);
|
|
|
|
DBUG_RETURN(index_read_idx_with_result_set(buf, active_index, key,
|
|
|
|
key_len, find_flag,
|
|
|
|
&stored_result));
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
Positions an index cursor to the index specified in key. Fetches the
|
|
|
|
row if any. This is only used to read whole keys.
|
|
|
|
|
|
|
|
This method is called via index_read in the case of a WHERE clause using
|
2006-06-27 23:49:48 -07:00
|
|
|
a primary key index OR is called DIRECTLY when the WHERE clause
|
2004-12-11 12:03:51 -08:00
|
|
|
uses a PRIMARY KEY index.
|
2006-06-27 23:49:48 -07:00
|
|
|
|
|
|
|
NOTES
|
|
|
|
This uses an internal result set that is deleted before function
|
|
|
|
returns. We need to be able to be calable from ha_rnd_pos()
|
2004-12-11 12:03:51 -08:00
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
int ha_federated::index_read_idx(uchar *buf, uint index, const uchar *key,
|
2005-07-15 17:33:47 -07:00
|
|
|
uint key_len, enum ha_rkey_function find_flag)
|
2006-06-27 23:49:48 -07:00
|
|
|
{
|
|
|
|
int retval;
|
|
|
|
MYSQL_RES *mysql_result;
|
|
|
|
DBUG_ENTER("ha_federated::index_read_idx");
|
|
|
|
|
|
|
|
if ((retval= index_read_idx_with_result_set(buf, index, key,
|
|
|
|
key_len, find_flag,
|
|
|
|
&mysql_result)))
|
|
|
|
DBUG_RETURN(retval);
|
|
|
|
mysql_free_result(mysql_result);
|
|
|
|
DBUG_RETURN(retval);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
Create result set for rows matching query and return first row
|
|
|
|
|
|
|
|
RESULT
|
|
|
|
0 ok In this case *result will contain the result set
|
|
|
|
table->status == 0
|
|
|
|
# error In this case *result will contain 0
|
|
|
|
table->status == STATUS_NOT_FOUND
|
|
|
|
*/
|
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
int ha_federated::index_read_idx_with_result_set(uchar *buf, uint index,
|
|
|
|
const uchar *key,
|
2006-06-27 23:49:48 -07:00
|
|
|
uint key_len,
|
|
|
|
ha_rkey_function find_flag,
|
|
|
|
MYSQL_RES **result)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
int retval;
|
|
|
|
char error_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
char index_value[STRING_BUFFER_USUAL_SIZE];
|
|
|
|
char sql_query_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
2006-06-27 23:49:48 -07:00
|
|
|
String index_string(index_value,
|
2005-07-15 17:33:47 -07:00
|
|
|
sizeof(index_value),
|
|
|
|
&my_charset_bin);
|
|
|
|
String sql_query(sql_query_buffer,
|
|
|
|
sizeof(sql_query_buffer),
|
|
|
|
&my_charset_bin);
|
|
|
|
key_range range;
|
2006-06-27 23:49:48 -07:00
|
|
|
DBUG_ENTER("ha_federated::index_read_idx_with_result_set");
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
*result= 0; // In case of errors
|
2005-07-15 17:33:47 -07:00
|
|
|
index_string.length(0);
|
2004-12-11 12:03:51 -08:00
|
|
|
sql_query.length(0);
|
2007-03-02 08:43:45 -08:00
|
|
|
ha_statistic_increment(&SSV::ha_read_key_count);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
sql_query.append(share->select_query);
|
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
range.key= key;
|
|
|
|
range.length= key_len;
|
|
|
|
range.flag= find_flag;
|
|
|
|
create_where_from_key(&index_string,
|
|
|
|
&table->key_info[index],
|
|
|
|
&range,
|
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 18:52:22 +03:00
|
|
|
NULL, 0, 0);
|
2004-12-11 12:03:51 -08:00
|
|
|
sql_query.append(index_string);
|
|
|
|
|
2007-07-25 12:23:24 -07:00
|
|
|
if (real_query(sql_query.ptr(), sql_query.length()))
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-11-24 02:36:28 +02:00
|
|
|
my_sprintf(error_buffer, (error_buffer, "error: %d '%s'",
|
2005-07-17 04:10:19 +02:00
|
|
|
mysql_errno(mysql), mysql_error(mysql)));
|
|
|
|
retval= ER_QUERY_ON_FOREIGN_DATA_SOURCE;
|
2005-07-15 17:33:47 -07:00
|
|
|
goto error;
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
2006-06-27 23:49:48 -07:00
|
|
|
if (!(*result= mysql_store_result(mysql)))
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
retval= HA_ERR_END_OF_FILE;
|
|
|
|
goto error;
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
2006-06-27 23:49:48 -07:00
|
|
|
if (!(retval= read_next(buf, *result)))
|
|
|
|
DBUG_RETURN(retval);
|
2005-07-15 17:33:47 -07:00
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
mysql_free_result(*result);
|
|
|
|
*result= 0;
|
|
|
|
table->status= STATUS_NOT_FOUND;
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_RETURN(retval);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
error:
|
|
|
|
table->status= STATUS_NOT_FOUND;
|
|
|
|
my_error(retval, MYF(0), error_buffer);
|
|
|
|
DBUG_RETURN(retval);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
|
2007-11-30 17:08:00 +04:00
|
|
|
/*
|
|
|
|
This method is used exlusevely by filesort() to check if we
|
|
|
|
can create sorting buffers of necessary size.
|
|
|
|
If the handler returns more records that it declares
|
|
|
|
here server can just crash on filesort().
|
|
|
|
We cannot guarantee that's not going to happen with
|
|
|
|
the FEDERATED engine, as we have records==0 always if the
|
|
|
|
client is a VIEW, and for the table the number of
|
|
|
|
records can inpredictably change during execution.
|
|
|
|
So we return maximum possible value here.
|
|
|
|
*/
|
|
|
|
|
|
|
|
ha_rows ha_federated::estimate_rows_upper_bound()
|
|
|
|
{
|
|
|
|
return HA_POS_ERROR;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
/* Initialized at each key walk (called multiple times unlike rnd_init()) */
|
2006-06-27 23:49:48 -07:00
|
|
|
|
2005-07-18 13:31:02 +02:00
|
|
|
int ha_federated::index_init(uint keynr, bool sorted)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::index_init");
|
2006-11-27 01:47:38 +02:00
|
|
|
DBUG_PRINT("info", ("table: '%s' key: %u", table->s->table_name.str, keynr));
|
2004-12-11 12:03:51 -08:00
|
|
|
active_index= keynr;
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
/*
|
|
|
|
Read first range
|
2005-07-15 17:33:47 -07:00
|
|
|
*/
|
2006-06-27 23:49:48 -07:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
int ha_federated::read_range_first(const key_range *start_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 18:52:22 +03:00
|
|
|
const key_range *end_key,
|
2006-12-15 00:51:37 +02:00
|
|
|
bool eq_range_arg, bool sorted)
|
2005-07-15 17:33:47 -07:00
|
|
|
{
|
|
|
|
char sql_query_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
int retval;
|
|
|
|
String sql_query(sql_query_buffer,
|
|
|
|
sizeof(sql_query_buffer),
|
|
|
|
&my_charset_bin);
|
|
|
|
DBUG_ENTER("ha_federated::read_range_first");
|
2005-10-08 00:57:40 +03:00
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
DBUG_ASSERT(!(start_key == NULL && end_key == NULL));
|
2005-07-15 17:33:47 -07:00
|
|
|
|
|
|
|
sql_query.length(0);
|
|
|
|
sql_query.append(share->select_query);
|
|
|
|
create_where_from_key(&sql_query,
|
|
|
|
&table->key_info[active_index],
|
2007-01-29 01:47:35 +02:00
|
|
|
start_key, end_key, 0, eq_range_arg);
|
2005-07-15 17:33:47 -07:00
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
if (stored_result)
|
|
|
|
{
|
|
|
|
mysql_free_result(stored_result);
|
|
|
|
stored_result= 0;
|
|
|
|
}
|
2007-07-25 12:23:24 -07:00
|
|
|
if (real_query(sql_query.ptr(), sql_query.length()))
|
2005-07-15 17:33:47 -07:00
|
|
|
{
|
2005-07-17 04:10:19 +02:00
|
|
|
retval= ER_QUERY_ON_FOREIGN_DATA_SOURCE;
|
2005-07-15 17:33:47 -07:00
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
sql_query.length(0);
|
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
if (!(stored_result= mysql_store_result(mysql)))
|
2005-07-15 17:33:47 -07:00
|
|
|
{
|
|
|
|
retval= HA_ERR_END_OF_FILE;
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
retval= read_next(table->record[0], stored_result);
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_RETURN(retval);
|
|
|
|
|
|
|
|
error:
|
2006-06-27 23:49:48 -07:00
|
|
|
table->status= STATUS_NOT_FOUND;
|
|
|
|
DBUG_RETURN(retval);
|
2005-07-15 17:33:47 -07:00
|
|
|
}
|
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
int ha_federated::read_range_next()
|
|
|
|
{
|
|
|
|
int retval;
|
|
|
|
DBUG_ENTER("ha_federated::read_range_next");
|
|
|
|
retval= rnd_next(table->record[0]);
|
|
|
|
DBUG_RETURN(retval);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
/* Used to read forward through the index. */
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
int ha_federated::index_next(uchar *buf)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::index_next");
|
2007-03-02 08:43:45 -08:00
|
|
|
ha_statistic_increment(&SSV::ha_read_next_count);
|
2006-06-27 23:49:48 -07:00
|
|
|
DBUG_RETURN(read_next(buf, stored_result));
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
2006-06-27 23:49:48 -07:00
|
|
|
|
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
/*
|
|
|
|
rnd_init() is called when the system wants the storage engine to do a table
|
|
|
|
scan.
|
|
|
|
|
|
|
|
This is the method that gets data for the SELECT calls.
|
|
|
|
|
|
|
|
See the federated in the introduction at the top of this file to see when
|
|
|
|
rnd_init() is called.
|
|
|
|
|
|
|
|
Called from filesort.cc, records.cc, sql_handler.cc, sql_select.cc,
|
|
|
|
sql_table.cc, and sql_update.cc.
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
int ha_federated::rnd_init(bool scan)
|
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_ENTER("ha_federated::rnd_init");
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
/*
|
2005-07-15 17:33:47 -07:00
|
|
|
The use of the 'scan' flag is incredibly important for this handler
|
|
|
|
to work properly, especially with updates containing WHERE clauses
|
|
|
|
using indexed columns.
|
2005-02-23 00:38:29 -08:00
|
|
|
|
|
|
|
When the initial query contains a WHERE clause of the query using an
|
|
|
|
indexed column, it's index_read_idx that selects the exact record from
|
|
|
|
the foreign database.
|
|
|
|
|
|
|
|
When there is NO index in the query, either due to not having a WHERE
|
|
|
|
clause, or the WHERE clause is using columns that are not indexed, a
|
|
|
|
'full table scan' done by rnd_init, which in this situation simply means
|
|
|
|
a 'select * from ...' on the foreign table.
|
|
|
|
|
|
|
|
In other words, this 'scan' flag gives us the means to ensure that if
|
|
|
|
there is an index involved in the query, we want index_read_idx to
|
|
|
|
retrieve the exact record (scan flag is 0), and do not want rnd_init
|
|
|
|
to do a 'full table scan' and wipe out that result set.
|
|
|
|
|
|
|
|
Prior to using this flag, the problem was most apparent with updates.
|
|
|
|
|
|
|
|
An initial query like 'UPDATE tablename SET anything = whatever WHERE
|
|
|
|
indexedcol = someval', index_read_idx would get called, using a query
|
|
|
|
constructed with a WHERE clause built from the values of index ('indexcol'
|
|
|
|
in this case, having a value of 'someval'). mysql_store_result would
|
|
|
|
then get called (this would be the result set we want to use).
|
|
|
|
|
|
|
|
After this rnd_init (from sql_update.cc) would be called, it would then
|
|
|
|
unecessarily call "select * from table" on the foreign table, then call
|
|
|
|
mysql_store_result, which would wipe out the correct previous result set
|
|
|
|
from the previous call of index_read_idx's that had the result set
|
|
|
|
containing the correct record, hence update the wrong row!
|
|
|
|
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
*/
|
2006-02-28 11:17:40 +01:00
|
|
|
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
if (scan)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-07-17 04:10:19 +02:00
|
|
|
if (stored_result)
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
{
|
2005-07-17 04:10:19 +02:00
|
|
|
mysql_free_result(stored_result);
|
|
|
|
stored_result= 0;
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
}
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2007-07-25 12:23:24 -07:00
|
|
|
if (real_query(share->select_query, strlen(share->select_query)))
|
2005-07-15 17:33:47 -07:00
|
|
|
goto error;
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
|
2005-07-17 04:10:19 +02:00
|
|
|
stored_result= mysql_store_result(mysql);
|
|
|
|
if (!stored_result)
|
2005-07-15 17:33:47 -07:00
|
|
|
goto error;
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
}
|
2004-12-11 12:03:51 -08:00
|
|
|
DBUG_RETURN(0);
|
2005-07-15 17:33:47 -07:00
|
|
|
|
|
|
|
error:
|
2006-06-27 23:49:48 -07:00
|
|
|
DBUG_RETURN(stash_remote_error());
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
int ha_federated::rnd_end()
|
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::rnd_end");
|
2006-06-27 23:49:48 -07:00
|
|
|
DBUG_RETURN(index_end());
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
2005-11-23 22:45:02 +02:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
int ha_federated::index_end(void)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::index_end");
|
2005-07-17 04:10:19 +02:00
|
|
|
if (stored_result)
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
{
|
2005-07-17 04:10:19 +02:00
|
|
|
mysql_free_result(stored_result);
|
|
|
|
stored_result= 0;
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
}
|
2004-12-11 12:03:51 -08:00
|
|
|
active_index= MAX_KEY;
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
2005-11-23 22:45:02 +02:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
/*
|
|
|
|
This is called for each row of the table scan. When you run out of records
|
|
|
|
you should return HA_ERR_END_OF_FILE. Fill buff up with the row information.
|
|
|
|
The Field structure for the table is the key to getting data into buf
|
|
|
|
in a manner that will allow the server to understand it.
|
|
|
|
|
|
|
|
Called from filesort.cc, records.cc, sql_handler.cc, sql_select.cc,
|
|
|
|
sql_table.cc, and sql_update.cc.
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
int ha_federated::rnd_next(uchar *buf)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::rnd_next");
|
|
|
|
|
2005-07-17 04:10:19 +02:00
|
|
|
if (stored_result == 0)
|
2005-03-15 00:32:21 -08:00
|
|
|
{
|
|
|
|
/*
|
|
|
|
Return value of rnd_init is not always checked (see records.cc),
|
|
|
|
so we can get here _even_ if there is _no_ pre-fetched result-set!
|
2006-06-27 23:49:48 -07:00
|
|
|
TODO: fix it. We can delete this in 5.1 when rnd_init() is checked.
|
|
|
|
*/
|
2005-03-15 00:32:21 -08:00
|
|
|
DBUG_RETURN(1);
|
|
|
|
}
|
2006-06-27 23:49:48 -07:00
|
|
|
DBUG_RETURN(read_next(buf, stored_result));
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
ha_federated::read_next
|
|
|
|
|
|
|
|
reads from a result set and converts to mysql internal
|
|
|
|
format
|
|
|
|
|
|
|
|
SYNOPSIS
|
|
|
|
field_in_record_is_null()
|
|
|
|
buf byte pointer to record
|
|
|
|
result mysql result set
|
|
|
|
|
|
|
|
DESCRIPTION
|
|
|
|
This method is a wrapper method that reads one record from a result
|
|
|
|
set and converts it to the internal table format
|
|
|
|
|
|
|
|
RETURN VALUE
|
|
|
|
1 error
|
|
|
|
0 no error
|
|
|
|
*/
|
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
int ha_federated::read_next(uchar *buf, MYSQL_RES *result)
|
2006-06-27 23:49:48 -07:00
|
|
|
{
|
|
|
|
int retval;
|
|
|
|
MYSQL_ROW row;
|
|
|
|
DBUG_ENTER("ha_federated::read_next");
|
|
|
|
|
|
|
|
table->status= STATUS_NOT_FOUND; // For easier return
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
/* Fetch a row, insert it back in a row format. */
|
2006-06-27 23:49:48 -07:00
|
|
|
if (!(row= mysql_fetch_row(result)))
|
2004-12-11 12:03:51 -08:00
|
|
|
DBUG_RETURN(HA_ERR_END_OF_FILE);
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
if (!(retval= convert_row_to_internal_format(buf, row, result)))
|
|
|
|
table->status= 0;
|
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_RETURN(retval);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
2006-06-27 23:49:48 -07:00
|
|
|
store reference to current row so that we can later find it for
|
|
|
|
a re-read, update or delete.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
In case of federated, a reference is either a primary key or
|
|
|
|
the whole record.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
Called from filesort.cc, sql_select.cc, sql_delete.cc and sql_update.cc.
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
void ha_federated::position(const uchar *record)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::position");
|
2006-06-27 23:49:48 -07:00
|
|
|
if (table->s->primary_key != MAX_KEY)
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
key_copy(ref, (uchar *)record, table->key_info + table->s->primary_key,
|
2006-06-27 23:49:48 -07:00
|
|
|
ref_length);
|
|
|
|
else
|
|
|
|
memcpy(ref, record, ref_length);
|
2004-12-11 12:03:51 -08:00
|
|
|
DBUG_VOID_RETURN;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
This is like rnd_next, but you are given a position to use to determine the
|
2006-06-27 23:49:48 -07:00
|
|
|
row. The position will be of the type that you stored in ref.
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
This method is required for an ORDER BY
|
2004-12-11 12:03:51 -08:00
|
|
|
|
|
|
|
Called from filesort.cc records.cc sql_insert.cc sql_select.cc sql_update.cc.
|
|
|
|
*/
|
2006-06-27 23:49:48 -07:00
|
|
|
|
WL#3817: Simplify string / memory area types and make things more consistent (first part)
The following type conversions was done:
- Changed byte to uchar
- Changed gptr to uchar*
- Change my_string to char *
- Change my_size_t to size_t
- Change size_s to size_t
Removed declaration of byte, gptr, my_string, my_size_t and size_s.
Following function parameter changes was done:
- All string functions in mysys/strings was changed to use size_t
instead of uint for string lengths.
- All read()/write() functions changed to use size_t (including vio).
- All protocoll functions changed to use size_t instead of uint
- Functions that used a pointer to a string length was changed to use size_t*
- Changed malloc(), free() and related functions from using gptr to use void *
as this requires fewer casts in the code and is more in line with how the
standard functions work.
- Added extra length argument to dirname_part() to return the length of the
created string.
- Changed (at least) following functions to take uchar* as argument:
- db_dump()
- my_net_write()
- net_write_command()
- net_store_data()
- DBUG_DUMP()
- decimal2bin() & bin2decimal()
- Changed my_compress() and my_uncompress() to use size_t. Changed one
argument to my_uncompress() from a pointer to a value as we only return
one value (makes function easier to use).
- Changed type of 'pack_data' argument to packfrm() to avoid casts.
- Changed in readfrm() and writefrom(), ha_discover and handler::discover()
the type for argument 'frmdata' to uchar** to avoid casts.
- Changed most Field functions to use uchar* instead of char* (reduced a lot of
casts).
- Changed field->val_xxx(xxx, new_ptr) to take const pointers.
Other changes:
- Removed a lot of not needed casts
- Added a few new cast required by other changes
- Added some cast to my_multi_malloc() arguments for safety (as string lengths
needs to be uint, not size_t).
- Fixed all calls to hash-get-key functions to use size_t*. (Needed to be done
explicitely as this conflict was often hided by casting the function to
hash_get_key).
- Changed some buffers to memory regions to uchar* to avoid casts.
- Changed some string lengths from uint to size_t.
- Changed field->ptr to be uchar* instead of char*. This allowed us to
get rid of a lot of casts.
- Some changes from true -> TRUE, false -> FALSE, unsigned char -> uchar
- Include zlib.h in some files as we needed declaration of crc32()
- Changed MY_FILE_ERROR to be (size_t) -1.
- Changed many variables to hold the result of my_read() / my_write() to be
size_t. This was needed to properly detect errors (which are
returned as (size_t) -1).
- Removed some very old VMS code
- Changed packfrm()/unpackfrm() to not be depending on uint size
(portability fix)
- Removed windows specific code to restore cursor position as this
causes slowdown on windows and we should not mix read() and pread()
calls anyway as this is not thread safe. Updated function comment to
reflect this. Changed function that depended on original behavior of
my_pwrite() to itself restore the cursor position (one such case).
- Added some missing checking of return value of malloc().
- Changed definition of MOD_PAD_CHAR_TO_FULL_LENGTH to avoid 'long' overflow.
- Changed type of table_def::m_size from my_size_t to ulong to reflect that
m_size is the number of elements in the array, not a string/memory
length.
- Moved THD::max_row_length() to table.cc (as it's not depending on THD).
Inlined max_row_length_blob() into this function.
- More function comments
- Fixed some compiler warnings when compiled without partitions.
- Removed setting of LEX_STRING() arguments in declaration (portability fix).
- Some trivial indentation/variable name changes.
- Some trivial code simplifications:
- Replaced some calls to alloc_root + memcpy to use
strmake_root()/strdup_root().
- Changed some calls from memdup() to strmake() (Safety fix)
- Simpler loops in client-simple.c
2007-05-10 12:59:39 +03:00
|
|
|
int ha_federated::rnd_pos(uchar *buf, uchar *pos)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2006-06-27 23:49:48 -07:00
|
|
|
int result;
|
2004-12-11 12:03:51 -08:00
|
|
|
DBUG_ENTER("ha_federated::rnd_pos");
|
2007-03-02 08:43:45 -08:00
|
|
|
ha_statistic_increment(&SSV::ha_read_rnd_count);
|
2006-06-27 23:49:48 -07:00
|
|
|
if (table->s->primary_key != MAX_KEY)
|
|
|
|
{
|
|
|
|
/* We have a primary key, so use index_read_idx to find row */
|
|
|
|
result= index_read_idx(buf, table->s->primary_key, pos,
|
|
|
|
ref_length, HA_READ_KEY_EXACT);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
/* otherwise, get the old record ref as obtained in ::position */
|
|
|
|
memcpy(buf, pos, ref_length);
|
|
|
|
result= 0;
|
|
|
|
}
|
|
|
|
table->status= result ? STATUS_NOT_FOUND : 0;
|
|
|
|
DBUG_RETURN(result);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
::info() is used to return information to the optimizer.
|
|
|
|
Currently this table handler doesn't implement most of the fields
|
|
|
|
really needed. SHOW also makes use of this data
|
|
|
|
Another note, you will probably want to have the following in your
|
|
|
|
code:
|
|
|
|
if (records < 2)
|
|
|
|
records = 2;
|
|
|
|
The reason is that the server will optimize for cases of only a single
|
|
|
|
record. If in a table scan you don't know the number of records
|
|
|
|
it will probably be better to set records to two so you can return
|
|
|
|
as many records as you need.
|
|
|
|
Along with records a few more variables you may wish to set are:
|
|
|
|
records
|
|
|
|
deleted
|
|
|
|
data_file_length
|
|
|
|
index_file_length
|
|
|
|
delete_length
|
|
|
|
check_time
|
|
|
|
Take a look at the public variables in handler.h for more information.
|
|
|
|
|
|
|
|
Called in:
|
|
|
|
filesort.cc
|
|
|
|
ha_heap.cc
|
|
|
|
item_sum.cc
|
|
|
|
opt_sum.cc
|
|
|
|
sql_delete.cc
|
|
|
|
sql_delete.cc
|
|
|
|
sql_derived.cc
|
|
|
|
sql_select.cc
|
|
|
|
sql_select.cc
|
|
|
|
sql_select.cc
|
|
|
|
sql_select.cc
|
|
|
|
sql_select.cc
|
|
|
|
sql_show.cc
|
|
|
|
sql_show.cc
|
|
|
|
sql_show.cc
|
|
|
|
sql_show.cc
|
|
|
|
sql_table.cc
|
|
|
|
sql_union.cc
|
|
|
|
sql_update.cc
|
|
|
|
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2006-10-16 17:41:29 +10:00
|
|
|
int ha_federated::info(uint flag)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
char error_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
char status_buf[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
int error;
|
|
|
|
uint error_code;
|
2005-07-17 04:10:19 +02:00
|
|
|
MYSQL_RES *result= 0;
|
2005-07-15 17:33:47 -07:00
|
|
|
MYSQL_ROW row;
|
|
|
|
String status_query_string(status_buf, sizeof(status_buf), &my_charset_bin);
|
2004-12-11 12:03:51 -08:00
|
|
|
DBUG_ENTER("ha_federated::info");
|
2005-07-15 17:33:47 -07:00
|
|
|
|
2005-07-17 04:10:19 +02:00
|
|
|
error_code= ER_QUERY_ON_FOREIGN_DATA_SOURCE;
|
2005-07-15 17:33:47 -07:00
|
|
|
/* we want not to show table status if not needed to do so */
|
|
|
|
if (flag & (HA_STATUS_VARIABLE | HA_STATUS_CONST))
|
|
|
|
{
|
|
|
|
status_query_string.length(0);
|
2007-06-30 13:17:49 +02:00
|
|
|
status_query_string.append(STRING_WITH_LEN("SHOW TABLE STATUS LIKE "));
|
2007-06-28 00:23:14 -07:00
|
|
|
append_ident(&status_query_string, share->table_name,
|
|
|
|
share->table_name_length, value_quote_char);
|
2005-07-15 17:33:47 -07:00
|
|
|
|
2007-07-25 12:23:24 -07:00
|
|
|
if (real_query(status_query_string.ptr(), status_query_string.length()))
|
2005-07-15 17:33:47 -07:00
|
|
|
goto error;
|
|
|
|
|
|
|
|
status_query_string.length(0);
|
|
|
|
|
|
|
|
result= mysql_store_result(mysql);
|
2007-10-15 10:11:52 +05:00
|
|
|
|
|
|
|
/*
|
|
|
|
We're going to use fields num. 4, 12 and 13 of the resultset,
|
|
|
|
so make sure we have these fields.
|
|
|
|
*/
|
|
|
|
if (!result || (mysql_num_fields(result) < 14))
|
2005-07-15 17:33:47 -07:00
|
|
|
goto error;
|
|
|
|
|
2005-07-17 04:10:19 +02:00
|
|
|
if (!mysql_num_rows(result))
|
2005-07-15 17:33:47 -07:00
|
|
|
goto error;
|
|
|
|
|
|
|
|
if (!(row= mysql_fetch_row(result)))
|
|
|
|
goto error;
|
|
|
|
|
2008-12-16 10:12:22 -02:00
|
|
|
if (flag & (HA_STATUS_VARIABLE | HA_STATUS_CONST))
|
2005-07-15 17:33:47 -07:00
|
|
|
{
|
2006-07-06 16:07:24 -07:00
|
|
|
/*
|
2005-07-15 17:33:47 -07:00
|
|
|
deleted is set in ha_federated::info
|
|
|
|
*/
|
|
|
|
/*
|
|
|
|
need to figure out what this means as far as federated is concerned,
|
|
|
|
since we don't have a "file"
|
|
|
|
|
|
|
|
data_file_length = ?
|
|
|
|
index_file_length = ?
|
|
|
|
delete_length = ?
|
|
|
|
*/
|
|
|
|
if (row[4] != NULL)
|
2006-07-06 16:07:24 -07:00
|
|
|
stats.records= (ha_rows) my_strtoll10(row[4], (char**) 0,
|
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
2006-06-04 18:52:22 +03:00
|
|
|
&error);
|
2005-07-15 17:33:47 -07:00
|
|
|
if (row[5] != NULL)
|
2007-09-21 10:15:16 +02:00
|
|
|
stats.mean_rec_length= (ulong) my_strtoll10(row[5], (char**) 0, &error);
|
2006-06-27 23:49:48 -07:00
|
|
|
|
2006-07-06 16:07:24 -07:00
|
|
|
stats.data_file_length= stats.records * stats.mean_rec_length;
|
2006-06-27 23:49:48 -07:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
if (row[12] != NULL)
|
2007-09-21 10:15:16 +02:00
|
|
|
stats.update_time= (time_t) my_strtoll10(row[12], (char**) 0,
|
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
2006-06-04 18:52:22 +03:00
|
|
|
&error);
|
2005-07-15 17:33:47 -07:00
|
|
|
if (row[13] != NULL)
|
2007-09-21 10:15:16 +02:00
|
|
|
stats.check_time= (time_t) my_strtoll10(row[13], (char**) 0,
|
This changeset is largely a handler cleanup changeset (WL#3281), but includes fixes and cleanups that was found necessary while testing the handler changes
Changes that requires code changes in other code of other storage engines.
(Note that all changes are very straightforward and one should find all issues
by compiling a --debug build and fixing all compiler errors and all
asserts in field.cc while running the test suite),
- New optional handler function introduced: reset()
This is called after every DML statement to make it easy for a handler to
statement specific cleanups.
(The only case it's not called is if force the file to be closed)
- handler::extra(HA_EXTRA_RESET) is removed. Code that was there before
should be moved to handler::reset()
- table->read_set contains a bitmap over all columns that are needed
in the query. read_row() and similar functions only needs to read these
columns
- table->write_set contains a bitmap over all columns that will be updated
in the query. write_row() and update_row() only needs to update these
columns.
The above bitmaps should now be up to date in all context
(including ALTER TABLE, filesort()).
The handler is informed of any changes to the bitmap after
fix_fields() by calling the virtual function
handler::column_bitmaps_signal(). If the handler does caching of
these bitmaps (instead of using table->read_set, table->write_set),
it should redo the caching in this code. as the signal() may be sent
several times, it's probably best to set a variable in the signal
and redo the caching on read_row() / write_row() if the variable was
set.
- Removed the read_set and write_set bitmap objects from the handler class
- Removed all column bit handling functions from the handler class.
(Now one instead uses the normal bitmap functions in my_bitmap.c instead
of handler dedicated bitmap functions)
- field->query_id is removed. One should instead instead check
table->read_set and table->write_set if a field is used in the query.
- handler::extra(HA_EXTRA_RETRIVE_ALL_COLS) and
handler::extra(HA_EXTRA_RETRIEVE_PRIMARY_KEY) are removed. One should now
instead use table->read_set to check for which columns to retrieve.
- If a handler needs to call Field->val() or Field->store() on columns
that are not used in the query, one should install a temporary
all-columns-used map while doing so. For this, we provide the following
functions:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->read_set);
field->val();
dbug_tmp_restore_column_map(table->read_set, old_map);
and similar for the write map:
my_bitmap_map *old_map= dbug_tmp_use_all_columns(table, table->write_set);
field->val();
dbug_tmp_restore_column_map(table->write_set, old_map);
If this is not done, you will sooner or later hit a DBUG_ASSERT
in the field store() / val() functions.
(For not DBUG binaries, the dbug_tmp_restore_column_map() and
dbug_tmp_restore_column_map() are inline dummy functions and should
be optimized away be the compiler).
- If one needs to temporary set the column map for all binaries (and not
just to avoid the DBUG_ASSERT() in the Field::store() / Field::val()
methods) one should use the functions tmp_use_all_columns() and
tmp_restore_column_map() instead of the above dbug_ variants.
- All 'status' fields in the handler base class (like records,
data_file_length etc) are now stored in a 'stats' struct. This makes
it easier to know what status variables are provided by the base
handler. This requires some trivial variable names in the extra()
function.
- New virtual function handler::records(). This is called to optimize
COUNT(*) if (handler::table_flags() & HA_HAS_RECORDS()) is true.
(stats.records is not supposed to be an exact value. It's only has to
be 'reasonable enough' for the optimizer to be able to choose a good
optimization path).
- Non virtual handler::init() function added for caching of virtual
constants from engine.
- Removed has_transactions() virtual method. Now one should instead return
HA_NO_TRANSACTIONS in table_flags() if the table handler DOES NOT support
transactions.
- The 'xxxx_create_handler()' function now has a MEM_ROOT_root argument
that is to be used with 'new handler_name()' to allocate the handler
in the right area. The xxxx_create_handler() function is also
responsible for any initialization of the object before returning.
For example, one should change:
static handler *myisam_create_handler(TABLE_SHARE *table)
{
return new ha_myisam(table);
}
->
static handler *myisam_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root)
{
return new (mem_root) ha_myisam(table);
}
- New optional virtual function: use_hidden_primary_key().
This is called in case of an update/delete when
(table_flags() and HA_PRIMARY_KEY_REQUIRED_FOR_DELETE) is defined
but we don't have a primary key. This allows the handler to take precisions
in remembering any hidden primary key to able to update/delete any
found row. The default handler marks all columns to be read.
- handler::table_flags() now returns a ulonglong (to allow for more flags).
- New/changed table_flags()
- HA_HAS_RECORDS Set if ::records() is supported
- HA_NO_TRANSACTIONS Set if engine doesn't support transactions
- HA_PRIMARY_KEY_REQUIRED_FOR_DELETE
Set if we should mark all primary key columns for
read when reading rows as part of a DELETE
statement. If there is no primary key,
all columns are marked for read.
- HA_PARTIAL_COLUMN_READ Set if engine will not read all columns in some
cases (based on table->read_set)
- HA_PRIMARY_KEY_ALLOW_RANDOM_ACCESS
Renamed to HA_PRIMARY_KEY_REQUIRED_FOR_POSITION.
- HA_DUPP_POS Renamed to HA_DUPLICATE_POS
- HA_REQUIRES_KEY_COLUMNS_FOR_DELETE
Set this if we should mark ALL key columns for
read when when reading rows as part of a DELETE
statement. In case of an update we will mark
all keys for read for which key part changed
value.
- HA_STATS_RECORDS_IS_EXACT
Set this if stats.records is exact.
(This saves us some extra records() calls
when optimizing COUNT(*))
- Removed table_flags()
- HA_NOT_EXACT_COUNT Now one should instead use HA_HAS_RECORDS if
handler::records() gives an exact count() and
HA_STATS_RECORDS_IS_EXACT if stats.records is exact.
- HA_READ_RND_SAME Removed (no one supported this one)
- Removed not needed functions ha_retrieve_all_cols() and ha_retrieve_all_pk()
- Renamed handler::dupp_pos to handler::dup_pos
- Removed not used variable handler::sortkey
Upper level handler changes:
- ha_reset() now does some overall checks and calls ::reset()
- ha_table_flags() added. This is a cached version of table_flags(). The
cache is updated on engine creation time and updated on open.
MySQL level changes (not obvious from the above):
- DBUG_ASSERT() added to check that column usage matches what is set
in the column usage bit maps. (This found a LOT of bugs in current
column marking code).
- In 5.1 before, all used columns was marked in read_set and only updated
columns was marked in write_set. Now we only mark columns for which we
need a value in read_set.
- Column bitmaps are created in open_binary_frm() and open_table_from_share().
(Before this was in table.cc)
- handler::table_flags() calls are replaced with handler::ha_table_flags()
- For calling field->val() you must have the corresponding bit set in
table->read_set. For calling field->store() you must have the
corresponding bit set in table->write_set. (There are asserts in
all store()/val() functions to catch wrong usage)
- thd->set_query_id is renamed to thd->mark_used_columns and instead
of setting this to an integer value, this has now the values:
MARK_COLUMNS_NONE, MARK_COLUMNS_READ, MARK_COLUMNS_WRITE
Changed also all variables named 'set_query_id' to mark_used_columns.
- In filesort() we now inform the handler of exactly which columns are needed
doing the sort and choosing the rows.
- The TABLE_SHARE object has a 'all_set' column bitmap one can use
when one needs a column bitmap with all columns set.
(This is used for table->use_all_columns() and other places)
- The TABLE object has 3 column bitmaps:
- def_read_set Default bitmap for columns to be read
- def_write_set Default bitmap for columns to be written
- tmp_set Can be used as a temporary bitmap when needed.
The table object has also two pointer to bitmaps read_set and write_set
that the handler should use to find out which columns are used in which way.
- count() optimization now calls handler::records() instead of using
handler->stats.records (if (table_flags() & HA_HAS_RECORDS) is true).
- Added extra argument to Item::walk() to indicate if we should also
traverse sub queries.
- Added TABLE parameter to cp_buffer_from_ref()
- Don't close tables created with CREATE ... SELECT but keep them in
the table cache. (Faster usage of newly created tables).
New interfaces:
- table->clear_column_bitmaps() to initialize the bitmaps for tables
at start of new statements.
- table->column_bitmaps_set() to set up new column bitmaps and signal
the handler about this.
- table->column_bitmaps_set_no_signal() for some few cases where we need
to setup new column bitmaps but don't signal the handler (as the handler
has already been signaled about these before). Used for the momement
only in opt_range.cc when doing ROR scans.
- table->use_all_columns() to install a bitmap where all columns are marked
as use in the read and the write set.
- table->default_column_bitmaps() to install the normal read and write
column bitmaps, but not signaling the handler about this.
This is mainly used when creating TABLE instances.
- table->mark_columns_needed_for_delete(),
table->mark_columns_needed_for_delete() and
table->mark_columns_needed_for_insert() to allow us to put additional
columns in column usage maps if handler so requires.
(The handler indicates what it neads in handler->table_flags())
- table->prepare_for_position() to allow us to tell handler that it
needs to read primary key parts to be able to store them in
future table->position() calls.
(This replaces the table->file->ha_retrieve_all_pk function)
- table->mark_auto_increment_column() to tell handler are going to update
columns part of any auto_increment key.
- table->mark_columns_used_by_index() to mark all columns that is part of
an index. It will also send extra(HA_EXTRA_KEYREAD) to handler to allow
it to quickly know that it only needs to read colums that are part
of the key. (The handler can also use the column map for detecting this,
but simpler/faster handler can just monitor the extra() call).
- table->mark_columns_used_by_index_no_reset() to in addition to other columns,
also mark all columns that is used by the given key.
- table->restore_column_maps_after_mark_index() to restore to default
column maps after a call to table->mark_columns_used_by_index().
- New item function register_field_in_read_map(), for marking used columns
in table->read_map. Used by filesort() to mark all used columns
- Maintain in TABLE->merge_keys set of all keys that are used in query.
(Simplices some optimization loops)
- Maintain Field->part_of_key_not_clustered which is like Field->part_of_key
but the field in the clustered key is not assumed to be part of all index.
(used in opt_range.cc for faster loops)
- dbug_tmp_use_all_columns(), dbug_tmp_restore_column_map()
tmp_use_all_columns() and tmp_restore_column_map() functions to temporally
mark all columns as usable. The 'dbug_' version is primarily intended
inside a handler when it wants to just call Field:store() & Field::val()
functions, but don't need the column maps set for any other usage.
(ie:: bitmap_is_set() is never called)
- We can't use compare_records() to skip updates for handlers that returns
a partial column set and the read_set doesn't cover all columns in the
write set. The reason for this is that if we have a column marked only for
write we can't in the MySQL level know if the value changed or not.
The reason this worked before was that MySQL marked all to be written
columns as also to be read. The new 'optimal' bitmaps exposed this 'hidden
bug'.
- open_table_from_share() does not anymore setup temporary MEM_ROOT
object as a thread specific variable for the handler. Instead we
send the to-be-used MEMROOT to get_new_handler().
(Simpler, faster code)
Bugs fixed:
- Column marking was not done correctly in a lot of cases.
(ALTER TABLE, when using triggers, auto_increment fields etc)
(Could potentially result in wrong values inserted in table handlers
relying on that the old column maps or field->set_query_id was correct)
Especially when it comes to triggers, there may be cases where the
old code would cause lost/wrong values for NDB and/or InnoDB tables.
- Split thd->options flag OPTION_STATUS_NO_TRANS_UPDATE to two flags:
OPTION_STATUS_NO_TRANS_UPDATE and OPTION_KEEP_LOG.
This allowed me to remove some wrong warnings about:
"Some non-transactional changed tables couldn't be rolled back"
- Fixed handling of INSERT .. SELECT and CREATE ... SELECT that wrongly reset
(thd->options & OPTION_STATUS_NO_TRANS_UPDATE) which caused us to loose
some warnings about
"Some non-transactional changed tables couldn't be rolled back")
- Fixed use of uninitialized memory in ha_ndbcluster.cc::delete_table()
which could cause delete_table to report random failures.
- Fixed core dumps for some tests when running with --debug
- Added missing FN_LIBCHAR in mysql_rm_tmp_tables()
(This has probably caused us to not properly remove temporary files after
crash)
- slow_logs was not properly initialized, which could maybe cause
extra/lost entries in slow log.
- If we get an duplicate row on insert, change column map to read and
write all columns while retrying the operation. This is required by
the definition of REPLACE and also ensures that fields that are only
part of UPDATE are properly handled. This fixed a bug in NDB and
REPLACE where REPLACE wrongly copied some column values from the replaced
row.
- For table handler that doesn't support NULL in keys, we would give an error
when creating a primary key with NULL fields, even after the fields has been
automaticly converted to NOT NULL.
- Creating a primary key on a SPATIAL key, would fail if field was not
declared as NOT NULL.
Cleanups:
- Removed not used condition argument to setup_tables
- Removed not needed item function reset_query_id_processor().
- Field->add_index is removed. Now this is instead maintained in
(field->flags & FIELD_IN_ADD_INDEX)
- Field->fieldnr is removed (use field->field_index instead)
- New argument to filesort() to indicate that it should return a set of
row pointers (not used columns). This allowed me to remove some references
to sql_command in filesort and should also enable us to return column
results in some cases where we couldn't before.
- Changed column bitmap handling in opt_range.cc to be aligned with TABLE
bitmap, which allowed me to use bitmap functions instead of looping over
all fields to create some needed bitmaps. (Faster and smaller code)
- Broke up found too long lines
- Moved some variable declaration at start of function for better code
readability.
- Removed some not used arguments from functions.
(setup_fields(), mysql_prepare_insert_check_table())
- setup_fields() now takes an enum instead of an int for marking columns
usage.
- For internal temporary tables, use handler::write_row(),
handler::delete_row() and handler::update_row() instead of
handler::ha_xxxx() for faster execution.
- Changed some constants to enum's and define's.
- Using separate column read and write sets allows for easier checking
of timestamp field was set by statement.
- Remove calls to free_io_cache() as this is now done automaticly in ha_reset()
- Don't build table->normalized_path as this is now identical to table->path
(after bar's fixes to convert filenames)
- Fixed some missed DBUG_PRINT(.."%lx") to use "0x%lx" to make it easier to
do comparision with the 'convert-dbug-for-diff' tool.
Things left to do in 5.1:
- We wrongly log failed CREATE TABLE ... SELECT in some cases when using
row based logging (as shown by testcase binlog_row_mix_innodb_myisam.result)
Mats has promised to look into this.
- Test that my fix for CREATE TABLE ... SELECT is indeed correct.
(I added several test cases for this, but in this case it's better that
someone else also tests this throughly).
Lars has promosed to do this.
2006-06-04 18:52:22 +03:00
|
|
|
&error);
|
2005-07-15 17:33:47 -07:00
|
|
|
}
|
2006-06-27 23:49:48 -07:00
|
|
|
/*
|
|
|
|
size of IO operations (This is based on a good guess, no high science
|
|
|
|
involved)
|
|
|
|
*/
|
2005-07-15 17:33:47 -07:00
|
|
|
if (flag & HA_STATUS_CONST)
|
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 18:52:22 +03:00
|
|
|
stats.block_size= 4096;
|
2006-07-07 11:17:50 -07:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
}
|
|
|
|
|
2007-07-23 23:35:43 -07:00
|
|
|
if (flag & HA_STATUS_AUTO)
|
2007-07-24 07:23:59 -07:00
|
|
|
stats.auto_increment_value= mysql->last_used_con->insert_id;
|
2007-07-23 23:35:43 -07:00
|
|
|
|
|
|
|
mysql_free_result(result);
|
2005-07-17 04:10:19 +02:00
|
|
|
|
2006-10-16 17:41:29 +10:00
|
|
|
DBUG_RETURN(0);
|
2005-07-15 17:33:47 -07:00
|
|
|
|
|
|
|
error:
|
2007-07-25 12:23:24 -07:00
|
|
|
mysql_free_result(result);
|
|
|
|
if (mysql)
|
|
|
|
{
|
|
|
|
my_sprintf(error_buffer, (error_buffer, ": %d : %s",
|
|
|
|
mysql_errno(mysql), mysql_error(mysql)));
|
|
|
|
my_error(error_code, MYF(0), error_buffer);
|
|
|
|
}
|
|
|
|
else
|
2007-07-25 17:22:52 -07:00
|
|
|
if (remote_error_number != -1 /* error already reported */)
|
2007-07-25 12:23:24 -07:00
|
|
|
{
|
|
|
|
error_code= remote_error_number;
|
|
|
|
my_error(error_code, MYF(0), ER(error_code));
|
|
|
|
}
|
2006-10-16 17:41:29 +10:00
|
|
|
DBUG_RETURN(error_code);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2007-06-28 00:23:14 -07:00
|
|
|
/**
|
|
|
|
@brief Handles extra signals from MySQL server
|
|
|
|
|
|
|
|
@param[in] operation Hint for storage engine
|
|
|
|
|
|
|
|
@return Operation Status
|
|
|
|
@retval 0 OK
|
|
|
|
*/
|
|
|
|
int ha_federated::extra(ha_extra_function operation)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::extra");
|
|
|
|
switch (operation) {
|
|
|
|
case HA_EXTRA_IGNORE_DUP_KEY:
|
|
|
|
ignore_duplicates= TRUE;
|
|
|
|
break;
|
|
|
|
case HA_EXTRA_NO_IGNORE_DUP_KEY:
|
2007-06-28 13:36:26 -07:00
|
|
|
insert_dup_update= FALSE;
|
2007-06-28 00:23:14 -07:00
|
|
|
ignore_duplicates= FALSE;
|
|
|
|
break;
|
|
|
|
case HA_EXTRA_WRITE_CAN_REPLACE:
|
|
|
|
replace_duplicates= TRUE;
|
|
|
|
break;
|
|
|
|
case HA_EXTRA_WRITE_CANNOT_REPLACE:
|
2007-06-29 13:55:16 -07:00
|
|
|
/*
|
|
|
|
We use this flag to ensure that we do not create an "INSERT IGNORE"
|
|
|
|
statement when inserting new rows into the remote table.
|
|
|
|
*/
|
2007-06-28 00:23:14 -07:00
|
|
|
replace_duplicates= FALSE;
|
|
|
|
break;
|
2007-06-28 13:36:26 -07:00
|
|
|
case HA_EXTRA_INSERT_WITH_UPDATE:
|
|
|
|
insert_dup_update= TRUE;
|
|
|
|
break;
|
2007-06-28 00:23:14 -07:00
|
|
|
default:
|
|
|
|
/* do nothing */
|
|
|
|
DBUG_PRINT("info",("unhandled operation: %d", (uint) operation));
|
|
|
|
}
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2007-06-30 18:17:20 +02:00
|
|
|
/**
|
|
|
|
@brief Reset state of file to after 'open'.
|
|
|
|
|
|
|
|
@detail This function is called after every statement for all tables
|
|
|
|
used by that statement.
|
|
|
|
|
|
|
|
@return Operation status
|
|
|
|
@retval 0 OK
|
|
|
|
*/
|
|
|
|
|
|
|
|
int ha_federated::reset(void)
|
|
|
|
{
|
|
|
|
insert_dup_update= FALSE;
|
|
|
|
ignore_duplicates= FALSE;
|
|
|
|
replace_duplicates= FALSE;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
/*
|
|
|
|
Used to delete all rows in a table. Both for cases of truncate and
|
|
|
|
for cases where the optimizer realizes that all rows will be
|
|
|
|
removed as a result of a SQL statement.
|
|
|
|
|
|
|
|
Called from item_sum.cc by Item_func_group_concat::clear(),
|
|
|
|
Item_sum_count_distinct::clear(), and Item_func_group_concat::clear().
|
|
|
|
Called from sql_delete.cc by mysql_delete().
|
|
|
|
Called from sql_select.cc by JOIN::reinit().
|
|
|
|
Called from sql_union.cc by st_select_lex_unit::exec().
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
int ha_federated::delete_all_rows()
|
|
|
|
{
|
2005-07-15 17:33:47 -07:00
|
|
|
char query_buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
2004-12-11 12:03:51 -08:00
|
|
|
String query(query_buffer, sizeof(query_buffer), &my_charset_bin);
|
2005-10-08 00:57:40 +03:00
|
|
|
DBUG_ENTER("ha_federated::delete_all_rows");
|
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
query.length(0);
|
|
|
|
|
|
|
|
query.set_charset(system_charset_info);
|
2007-06-30 13:17:49 +02:00
|
|
|
query.append(STRING_WITH_LEN("TRUNCATE "));
|
2007-06-28 00:23:14 -07:00
|
|
|
append_ident(&query, share->table_name, share->table_name_length,
|
|
|
|
ident_quote_char);
|
2004-12-11 12:03:51 -08:00
|
|
|
|
2005-07-15 17:33:47 -07:00
|
|
|
/*
|
|
|
|
TRUNCATE won't return anything in mysql_affected_rows
|
|
|
|
*/
|
2007-07-25 12:23:24 -07:00
|
|
|
if (real_query(query.ptr(), query.length()))
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
{
|
2005-09-13 23:31:17 -07:00
|
|
|
DBUG_RETURN(stash_remote_error());
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
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 18:52:22 +03:00
|
|
|
stats.deleted+= stats.records;
|
|
|
|
stats.records= 0;
|
2005-07-15 17:33:47 -07:00
|
|
|
DBUG_RETURN(0);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
The idea with handler::store_lock() is the following:
|
|
|
|
|
|
|
|
The statement decided which locks we should need for the table
|
|
|
|
for updates/deletes/inserts we get WRITE locks, for SELECT... we get
|
|
|
|
read locks.
|
|
|
|
|
|
|
|
Before adding the lock into the table lock handler (see thr_lock.c)
|
|
|
|
mysqld calls store lock with the requested locks. Store lock can now
|
|
|
|
modify a write lock to a read lock (or some other lock), ignore the
|
|
|
|
lock (if we don't want to use MySQL table locks at all) or add locks
|
|
|
|
for many tables (like we do when we are using a MERGE handler).
|
|
|
|
|
|
|
|
Berkeley DB for federated changes all WRITE locks to TL_WRITE_ALLOW_WRITE
|
|
|
|
(which signals that we are doing WRITES, but we are still allowing other
|
|
|
|
reader's and writer's.
|
|
|
|
|
|
|
|
When releasing locks, store_lock() are also called. In this case one
|
|
|
|
usually doesn't have to do anything.
|
|
|
|
|
|
|
|
In some exceptional cases MySQL may send a request for a TL_IGNORE;
|
|
|
|
This means that we are requesting the same lock as last time and this
|
|
|
|
should also be ignored. (This may happen when someone does a flush
|
|
|
|
table when we have opened a part of the tables, in which case mysqld
|
|
|
|
closes and reopens the tables and tries to get the same locks at last
|
|
|
|
time). In the future we will probably try to remove this.
|
|
|
|
|
|
|
|
Called from lock.cc by get_lock_data().
|
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
THR_LOCK_DATA **ha_federated::store_lock(THD *thd,
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
THR_LOCK_DATA **to,
|
|
|
|
enum thr_lock_type lock_type)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2006-06-27 23:49:48 -07:00
|
|
|
DBUG_ENTER("ha_federated::store_lock");
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
if (lock_type != TL_IGNORE && lock.type == TL_UNLOCK)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-02-23 00:38:29 -08:00
|
|
|
/*
|
|
|
|
Here is where we get into the guts of a row level lock.
|
|
|
|
If TL_UNLOCK is set
|
|
|
|
If we are not doing a LOCK TABLE or DISCARD/IMPORT
|
|
|
|
TABLESPACE, then allow multiple writers
|
2004-12-11 12:03:51 -08:00
|
|
|
*/
|
|
|
|
|
|
|
|
if ((lock_type >= TL_WRITE_CONCURRENT_INSERT &&
|
2005-07-15 17:33:47 -07:00
|
|
|
lock_type <= TL_WRITE) && !thd->in_lock_tables)
|
2004-12-11 12:03:51 -08:00
|
|
|
lock_type= TL_WRITE_ALLOW_WRITE;
|
|
|
|
|
2005-02-23 00:38:29 -08:00
|
|
|
/*
|
|
|
|
In queries of type INSERT INTO t1 SELECT ... FROM t2 ...
|
|
|
|
MySQL would use the lock TL_READ_NO_INSERT on t2, and that
|
|
|
|
would conflict with TL_WRITE_ALLOW_WRITE, blocking all inserts
|
|
|
|
to t2. Convert the lock to a normal read lock to allow
|
|
|
|
concurrent inserts to t2.
|
2004-12-11 12:03:51 -08:00
|
|
|
*/
|
|
|
|
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
if (lock_type == TL_READ_NO_INSERT && !thd->in_lock_tables)
|
2004-12-11 12:03:51 -08:00
|
|
|
lock_type= TL_READ;
|
|
|
|
|
|
|
|
lock.type= lock_type;
|
|
|
|
}
|
|
|
|
|
|
|
|
*to++= &lock;
|
|
|
|
|
2006-06-27 23:49:48 -07:00
|
|
|
DBUG_RETURN(to);
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
create() does nothing, since we have no local setup of our own.
|
2005-02-23 00:38:29 -08:00
|
|
|
FUTURE: We should potentially connect to the foreign database and
|
2004-12-11 12:03:51 -08:00
|
|
|
*/
|
2005-02-23 00:38:29 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
int ha_federated::create(const char *name, TABLE *table_arg,
|
WL# 2094
This patch contains all that my previous patch (1.1814) contained, with the addition of using cli_fetch_lengths for
handling binary data (Bar noted this on the review of 1.1814, Guilhem suggested using cli_fetch_lenghts by
making available via removal of static in method definition and declaration in mysql.h, but
Konstantin had some reservations, but he said to commit the patch using this anyway,
and I suppose this can be discussed. I abandoned 1.1814 because Monty made a couple
fixes to my code as well as formatting changes, and I thought it would just be easier
to hand-edit my changes into a fresh clone and then make a patch.
The reason for using cli_fetch_lengths is so that I can correctly get the length of
the field I am setting into the field. I was previously using 'strlen' but Bar pointed out this
won't correctly get the length of binary data and is also less effecient. Upon testing,
it was in fact verified that binary data in a blob table was being inserted correctly,
but not being retrieved correctly, all due to not having the correct value for the
field:
(*field)->store(row[x], strlen(row[x]), &my_charset_bin);
was changed to:
(*field)->store(row[x], lengths[x], &my_charset_bin);
lengths being a unsigned long pointer to the values of the field lengths from a
MYSQL_ROW.
Since the server doesn't have the function "mysql_fetch_lengths" available, I tried
to use "result->lengths", but this isn't set, so I finally successfully used
cli_fetch_lenghts, which does give the correct lengths, and now the binary data gets
retrieved correctly.
I've also run the code through indent-ex and am using Brian's vimrc to ensure correct formatting!
This code passes the entire test suite, without any errors or warning on both my
workstation and build.mysql.com
2005-02-06 09:40:07 -08:00
|
|
|
HA_CREATE_INFO *create_info)
|
2004-12-11 12:03:51 -08:00
|
|
|
{
|
2005-10-08 00:57:40 +03:00
|
|
|
int retval;
|
2007-03-24 01:18:19 -07:00
|
|
|
THD *thd= current_thd;
|
2005-10-08 00:57:40 +03:00
|
|
|
FEDERATED_SHARE tmp_share; // Only a temporary share, to test the url
|
2005-01-05 00:39:13 +02:00
|
|
|
DBUG_ENTER("ha_federated::create");
|
2005-03-15 00:32:21 -08:00
|
|
|
|
2007-07-25 12:29:44 -07:00
|
|
|
retval= parse_url(thd->mem_root, &tmp_share, table_arg, 1);
|
2005-07-15 17:33:47 -07:00
|
|
|
|
|
|
|
DBUG_RETURN(retval);
|
2005-03-15 00:32:21 -08:00
|
|
|
|
2004-12-11 12:03:51 -08:00
|
|
|
}
|
2005-09-13 23:31:17 -07:00
|
|
|
|
|
|
|
|
2007-07-25 12:23:24 -07:00
|
|
|
int ha_federated::real_connect()
|
|
|
|
{
|
|
|
|
char buffer[FEDERATED_QUERY_BUFFER_SIZE];
|
|
|
|
String sql_query(buffer, sizeof(buffer), &my_charset_bin);
|
|
|
|
DBUG_ENTER("ha_federated::real_connect");
|
|
|
|
|
|
|
|
/*
|
|
|
|
Bug#25679
|
|
|
|
Ensure that we do not hold the LOCK_open mutex while attempting
|
|
|
|
to establish Federated connection to guard against a trivial
|
|
|
|
Denial of Service scenerio.
|
|
|
|
*/
|
|
|
|
safe_mutex_assert_not_owner(&LOCK_open);
|
|
|
|
|
|
|
|
DBUG_ASSERT(mysql == NULL);
|
|
|
|
|
|
|
|
if (!(mysql= mysql_init(NULL)))
|
|
|
|
{
|
|
|
|
remote_error_number= HA_ERR_OUT_OF_MEM;
|
|
|
|
DBUG_RETURN(-1);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
BUG# 17044 Federated Storage Engine is not UTF8 clean
|
|
|
|
Add set names to whatever charset the table is at open
|
|
|
|
of table
|
|
|
|
*/
|
|
|
|
/* this sets the csname like 'set names utf8' */
|
|
|
|
mysql_options(mysql,MYSQL_SET_CHARSET_NAME,
|
|
|
|
this->table->s->table_charset->csname);
|
|
|
|
|
|
|
|
sql_query.length(0);
|
|
|
|
|
|
|
|
if (!mysql_real_connect(mysql,
|
|
|
|
share->hostname,
|
|
|
|
share->username,
|
|
|
|
share->password,
|
|
|
|
share->database,
|
|
|
|
share->port,
|
|
|
|
share->socket, 0))
|
|
|
|
{
|
|
|
|
stash_remote_error();
|
|
|
|
mysql_close(mysql);
|
|
|
|
mysql= NULL;
|
|
|
|
my_error(ER_CONNECT_TO_FOREIGN_DATA_SOURCE, MYF(0), remote_error_buf);
|
|
|
|
remote_error_number= -1;
|
|
|
|
DBUG_RETURN(-1);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
We have established a connection, lets try a simple dummy query just
|
|
|
|
to check that the table and expected columns are present.
|
|
|
|
*/
|
|
|
|
sql_query.append(share->select_query);
|
2007-07-25 12:29:44 -07:00
|
|
|
sql_query.append(STRING_WITH_LEN(" WHERE 1=0"));
|
2007-07-25 12:23:24 -07:00
|
|
|
if (mysql_real_query(mysql, sql_query.ptr(), sql_query.length()))
|
|
|
|
{
|
|
|
|
sql_query.length(0);
|
|
|
|
sql_query.append("error: ");
|
|
|
|
sql_query.qs_append(mysql_errno(mysql));
|
|
|
|
sql_query.append(" '");
|
|
|
|
sql_query.append(mysql_error(mysql));
|
|
|
|
sql_query.append("'");
|
|
|
|
mysql_close(mysql);
|
|
|
|
mysql= NULL;
|
|
|
|
my_error(ER_FOREIGN_DATA_SOURCE_DOESNT_EXIST, MYF(0), sql_query.ptr());
|
|
|
|
remote_error_number= -1;
|
|
|
|
DBUG_RETURN(-1);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Just throw away the result, no rows anyways but need to keep in sync */
|
|
|
|
mysql_free_result(mysql_store_result(mysql));
|
|
|
|
|
|
|
|
/*
|
|
|
|
Since we do not support transactions at this version, we can let the client
|
|
|
|
API silently reconnect. For future versions, we will need more logic to
|
|
|
|
deal with transactions
|
|
|
|
*/
|
|
|
|
|
|
|
|
mysql->reconnect= 1;
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int ha_federated::real_query(const char *query, uint length)
|
|
|
|
{
|
|
|
|
int rc= 0;
|
|
|
|
DBUG_ENTER("ha_federated::real_query");
|
|
|
|
|
|
|
|
if (!mysql && (rc= real_connect()))
|
|
|
|
goto end;
|
|
|
|
|
|
|
|
if (!query || !length)
|
|
|
|
goto end;
|
|
|
|
|
|
|
|
rc= mysql_real_query(mysql, query, length);
|
|
|
|
|
|
|
|
end:
|
|
|
|
DBUG_RETURN(rc);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2005-09-13 23:31:17 -07:00
|
|
|
int ha_federated::stash_remote_error()
|
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::stash_remote_error()");
|
2007-07-25 12:23:24 -07:00
|
|
|
if (!mysql)
|
|
|
|
DBUG_RETURN(remote_error_number);
|
2005-09-13 23:31:17 -07:00
|
|
|
remote_error_number= mysql_errno(mysql);
|
2006-01-06 00:47:49 +02:00
|
|
|
strmake(remote_error_buf, mysql_error(mysql), sizeof(remote_error_buf)-1);
|
2007-06-28 13:36:26 -07:00
|
|
|
if (remote_error_number == ER_DUP_ENTRY ||
|
|
|
|
remote_error_number == ER_DUP_KEY)
|
|
|
|
DBUG_RETURN(HA_ERR_FOUND_DUPP_KEY);
|
2005-09-13 23:31:17 -07:00
|
|
|
DBUG_RETURN(HA_FEDERATED_ERROR_WITH_REMOTE_SYSTEM);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
bool ha_federated::get_error_message(int error, String* buf)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::get_error_message");
|
|
|
|
DBUG_PRINT("enter", ("error: %d", error));
|
|
|
|
if (error == HA_FEDERATED_ERROR_WITH_REMOTE_SYSTEM)
|
|
|
|
{
|
2005-11-20 20:47:07 +02:00
|
|
|
buf->append(STRING_WITH_LEN("Error on remote system: "));
|
2005-09-13 23:31:17 -07:00
|
|
|
buf->qs_append(remote_error_number);
|
2005-11-20 20:47:07 +02:00
|
|
|
buf->append(STRING_WITH_LEN(": "));
|
2005-11-17 06:05:09 -08:00
|
|
|
buf->append(remote_error_buf);
|
2005-09-13 23:31:17 -07:00
|
|
|
|
|
|
|
remote_error_number= 0;
|
|
|
|
remote_error_buf[0]= '\0';
|
|
|
|
}
|
|
|
|
DBUG_PRINT("exit", ("message: %s", buf->ptr()));
|
|
|
|
DBUG_RETURN(FALSE);
|
|
|
|
}
|
|
|
|
|
2005-11-09 05:53:34 -08:00
|
|
|
int ha_federated::external_lock(THD *thd, int lock_type)
|
|
|
|
{
|
|
|
|
int error= 0;
|
|
|
|
DBUG_ENTER("ha_federated::external_lock");
|
|
|
|
|
2007-07-20 11:35:19 -07:00
|
|
|
/*
|
|
|
|
Support for transactions disabled until WL#2952 fixes it.
|
|
|
|
*/
|
|
|
|
#ifdef XXX_SUPERCEDED_BY_WL2952
|
2005-11-09 05:53:34 -08:00
|
|
|
if (lock_type != F_UNLCK)
|
|
|
|
{
|
2007-08-31 10:19:52 +04:00
|
|
|
ha_federated *trx= (ha_federated *)thd_get_ha_data(thd, ht);
|
2007-07-20 14:01:37 -07:00
|
|
|
|
2005-11-09 05:53:34 -08:00
|
|
|
DBUG_PRINT("info",("federated not lock F_UNLCK"));
|
|
|
|
if (!(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)))
|
|
|
|
{
|
|
|
|
DBUG_PRINT("info",("federated autocommit"));
|
|
|
|
/*
|
|
|
|
This means we are doing an autocommit
|
|
|
|
*/
|
|
|
|
error= connection_autocommit(TRUE);
|
|
|
|
if (error)
|
|
|
|
{
|
|
|
|
DBUG_PRINT("info", ("error setting autocommit TRUE: %d", error));
|
|
|
|
DBUG_RETURN(error);
|
|
|
|
}
|
2006-09-30 12:49:46 -07:00
|
|
|
trans_register_ha(thd, FALSE, ht);
|
2005-11-09 05:53:34 -08:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
DBUG_PRINT("info",("not autocommit"));
|
|
|
|
if (!trx)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
This is where a transaction gets its start
|
|
|
|
*/
|
|
|
|
error= connection_autocommit(FALSE);
|
|
|
|
if (error)
|
|
|
|
{
|
|
|
|
DBUG_PRINT("info", ("error setting autocommit FALSE: %d", error));
|
|
|
|
DBUG_RETURN(error);
|
|
|
|
}
|
2007-08-31 10:19:52 +04:00
|
|
|
thd_set_ha_data(thd, ht, this);
|
2006-09-30 12:49:46 -07:00
|
|
|
trans_register_ha(thd, TRUE, ht);
|
2005-11-09 05:53:34 -08:00
|
|
|
/*
|
|
|
|
Send a lock table to the remote end.
|
|
|
|
We do not support this at the moment
|
|
|
|
*/
|
|
|
|
if (thd->options & (OPTION_TABLE_LOCK))
|
|
|
|
{
|
|
|
|
DBUG_PRINT("info", ("We do not support lock table yet"));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
ha_federated *ptr;
|
|
|
|
for (ptr= trx; ptr; ptr= ptr->trx_next)
|
|
|
|
if (ptr == this)
|
|
|
|
break;
|
|
|
|
else if (!ptr->trx_next)
|
|
|
|
ptr->trx_next= this;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2007-07-20 11:35:19 -07:00
|
|
|
#endif /* XXX_SUPERCEDED_BY_WL2952 */
|
2007-07-20 14:01:37 -07:00
|
|
|
DBUG_RETURN(error);
|
2005-11-09 05:53:34 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2006-09-30 12:49:46 -07:00
|
|
|
static int federated_commit(handlerton *hton, THD *thd, bool all)
|
2005-11-09 05:53:34 -08:00
|
|
|
{
|
|
|
|
int return_val= 0;
|
2007-08-31 10:19:52 +04:00
|
|
|
ha_federated *trx= (ha_federated *) thd_get_ha_data(thd, hton);
|
2005-11-09 05:53:34 -08:00
|
|
|
DBUG_ENTER("federated_commit");
|
|
|
|
|
|
|
|
if (all)
|
|
|
|
{
|
|
|
|
int error= 0;
|
|
|
|
ha_federated *ptr, *old= NULL;
|
|
|
|
for (ptr= trx; ptr; old= ptr, ptr= ptr->trx_next)
|
|
|
|
{
|
|
|
|
if (old)
|
|
|
|
old->trx_next= NULL;
|
|
|
|
error= ptr->connection_commit();
|
2006-12-15 10:45:25 +01:00
|
|
|
if (error && !return_val)
|
2005-11-09 05:53:34 -08:00
|
|
|
return_val= error;
|
|
|
|
}
|
2007-08-31 10:19:52 +04:00
|
|
|
thd_set_ha_data(thd, hton, NULL);
|
2005-11-09 05:53:34 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
DBUG_PRINT("info", ("error val: %d", return_val));
|
|
|
|
DBUG_RETURN(return_val);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2006-09-30 12:49:46 -07:00
|
|
|
static int federated_rollback(handlerton *hton, THD *thd, bool all)
|
2005-11-09 05:53:34 -08:00
|
|
|
{
|
|
|
|
int return_val= 0;
|
2007-08-31 10:19:52 +04:00
|
|
|
ha_federated *trx= (ha_federated *)thd_get_ha_data(thd, hton);
|
2005-11-09 05:53:34 -08:00
|
|
|
DBUG_ENTER("federated_rollback");
|
|
|
|
|
|
|
|
if (all)
|
|
|
|
{
|
|
|
|
int error= 0;
|
|
|
|
ha_federated *ptr, *old= NULL;
|
|
|
|
for (ptr= trx; ptr; old= ptr, ptr= ptr->trx_next)
|
|
|
|
{
|
|
|
|
if (old)
|
|
|
|
old->trx_next= NULL;
|
|
|
|
error= ptr->connection_rollback();
|
|
|
|
if (error && !return_val)
|
|
|
|
return_val= error;
|
|
|
|
}
|
2007-08-31 10:19:52 +04:00
|
|
|
thd_set_ha_data(thd, hton, NULL);
|
2005-11-09 05:53:34 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
DBUG_PRINT("info", ("error val: %d", return_val));
|
|
|
|
DBUG_RETURN(return_val);
|
|
|
|
}
|
|
|
|
|
|
|
|
int ha_federated::connection_commit()
|
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::connection_commit");
|
|
|
|
DBUG_RETURN(execute_simple_query("COMMIT", 6));
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int ha_federated::connection_rollback()
|
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::connection_rollback");
|
|
|
|
DBUG_RETURN(execute_simple_query("ROLLBACK", 8));
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int ha_federated::connection_autocommit(bool state)
|
|
|
|
{
|
|
|
|
const char *text;
|
|
|
|
DBUG_ENTER("ha_federated::connection_autocommit");
|
2006-09-07 16:17:16 -07:00
|
|
|
text= (state == TRUE) ? "SET AUTOCOMMIT=1" : "SET AUTOCOMMIT=0";
|
2005-11-09 05:53:34 -08:00
|
|
|
DBUG_RETURN(execute_simple_query(text, 16));
|
2006-05-28 14:51:01 +02:00
|
|
|
}
|
2005-11-09 05:53:34 -08:00
|
|
|
|
|
|
|
|
|
|
|
int ha_federated::execute_simple_query(const char *query, int len)
|
|
|
|
{
|
|
|
|
DBUG_ENTER("ha_federated::execute_simple_query");
|
|
|
|
|
|
|
|
if (mysql_real_query(mysql, query, len))
|
|
|
|
{
|
|
|
|
DBUG_RETURN(stash_remote_error());
|
|
|
|
}
|
|
|
|
DBUG_RETURN(0);
|
|
|
|
}
|
|
|
|
|
2006-05-28 14:51:01 +02:00
|
|
|
struct st_mysql_storage_engine federated_storage_engine=
|
2006-09-26 21:26:04 -07:00
|
|
|
{ MYSQL_HANDLERTON_INTERFACE_VERSION };
|
2006-04-13 13:49:29 -07:00
|
|
|
|
|
|
|
mysql_declare_plugin(federated)
|
|
|
|
{
|
|
|
|
MYSQL_STORAGE_ENGINE_PLUGIN,
|
2006-05-28 14:51:01 +02:00
|
|
|
&federated_storage_engine,
|
|
|
|
"FEDERATED",
|
2006-04-13 13:49:29 -07:00
|
|
|
"Patrick Galbraith and Brian Aker, MySQL AB",
|
2006-05-28 14:51:01 +02:00
|
|
|
"Federated MySQL storage engine",
|
2006-10-05 00:41:29 -07:00
|
|
|
PLUGIN_LICENSE_GPL,
|
2006-05-28 14:51:01 +02:00
|
|
|
federated_db_init, /* Plugin Init */
|
2006-11-19 18:01:54 -08:00
|
|
|
federated_done, /* Plugin Deinit */
|
2006-04-13 13:49:29 -07:00
|
|
|
0x0100 /* 1.0 */,
|
2006-08-30 14:27:29 -07:00
|
|
|
NULL, /* status variables */
|
|
|
|
NULL, /* system variables */
|
|
|
|
NULL /* config options */
|
2006-04-13 13:49:29 -07:00
|
|
|
}
|
|
|
|
mysql_declare_plugin_end;
|