In order to make multi-delete SP friendly we need to have all table
locks for the elements of main statement table list properly set
at the end of parsing.
Also performed small cleanup: We don't need relink_tables_for_multidelete()
any longer since the only case now when TABLE_LIST::correspondent_table
is non-zero are tables in auxilary table list of multi-delete and these
tables are handled specially in mysql_multi_delete_prepare().
mysql-test/r/sp-threads.result:
Added test case for bug #11158 "Can't perform multi-delete in stored
procedure".
mysql-test/t/sp-threads.test:
Added test case for bug #11158 "Can't perform multi-delete in stored
procedure".
sql/mysql_priv.h:
- Removed third argument from the declaration of multi_delete_precheck()
as nowdays we calculate number of tables in multi-delete from which
we are going to delete rows right at the end of statement parsing.
- Introduced definition of multi_delete_set_locks_and_link_aux_tables()
which is responsible for propagation of proper table locks from
multi-delete's auxilary table list to the main list and binding
corresponding tables in these two lists.
sql/sql_base.cc:
Removed relink_tables_for_multidelete() routine and its invocations.
We don't need them in 5.0 since the only case now when
TABLE_LIST::correspondent_table is non-zero are tables in auxilary table
list of multi-delete and these tables are handled specially in
mysql_multi_delete_prepare().
sql/sql_lex.h:
LEX::table_count
Added description of new role of this LEX member for multi-delete.
Now for this statement we store number of tables from which we should
delete records there.
sql/sql_parse.cc:
multi_delete_precheck():
Moved code which is responsible for iterating through auxilary table
list and binding its elements with corresponding elements of main
table list, and properly updating locks in it to separate function -
multi_delete_set_locks_and_link_aux_tables(). This is because in order
to make multi-delete SP friendly we need to have all locks set properly
at the end of statement parsing. So we are introducing new function
which will be called from parser.
We also calculate number of tables from which we are going to perform
deletions there and store this number for later usage in
LEX::table_count.
Also removed some no longer needed code.
sql/sql_prepare.cc:
mysql_test_multidelete():
Now multi_delete_precheck() takes only two arguments, so we don't
need to pass fake third parameter.
sql/sql_yacc.yy:
delete:
In order to make multi-delete SP friendly we need to have all table
locks for the elements of main statement table list properly set
at the end of parsing.
New more SP-locking friendly approach to handling locks in multi-update.
Now we mark all tables of multi-update as needing write lock at parsing
stage and if possible downgrade lock at execution stage (For its work
SP-locking mechanism needs to know all lock types right after parsing
stage).
mysql-test/r/sp-threads.result:
Added test for bug #9486 "Can't perform multi-update in stored procedure".
mysql-test/t/sp-threads.test:
Added test for bug #9486 "Can't perform multi-update in stored procedure".
sql/sp_head.cc:
SP_TABLE, sp_head::merge_table_list()/add_used_tables_to_table_list():
Since some queries during their execution (e.g. multi-update)
may change type of lock for some of their tables and thus change
lock_type member for some of elements of table list, we should
store type of lock in SP_TABLE struct explicitly instead of using
lock_type member of TABLE_LIST object pointed by SP_TABLE::table.
sql/sql_lex.h:
Removed no longer used LEX::multi_lock_option member.
sql/sql_prepare.cc:
mysql_test_update():
We don't need to bother about LEX::multi_lock_option if we convert
multi-update to update anymore. Since nowdays multi-update uses
TABLE_LIST::lock_type for specifying lock level of updated tables
instead of LEX::multi_lock_option.
sql/sql_update.cc:
mysql_update()/mysql_multi_update_prepare():
Now we mark all tables of multi-update as needing write lock at parsing
stage and if possible downgrade lock at execution stage. Old approach
(don't set lock type until execution stage) was not working well with
SP-locking (For its work SP-locking mechanism needs to know all lock
types right after parsing stage).
mysql_multi_update():
We should return FALSE if no error occurs.
sql/sql_yacc.yy:
update:
Now we mark all tables of multi-update as needing write lock at parsing
stage and if possible downgrade lock at execution stage. Old approach
(don't set lock type until execution stage) was not working well with
SP-locking (For its work SP-locking mechanism needs to know all lock
types right after parsing stage).
and removed a have_innodb.inc inclusion which was left by mistake
in an earlier change.
mysql-test/r/sp_trans.result:
Changed procedure name and delimiter setting to follow the style of the other SP test files.
mysql-test/t/sp-error.test:
Added comment with hint for bug test case style.
mysql-test/t/sp-threads.test:
Added comments, with hint for bug test case style.
mysql-test/t/sp.test:
Removed have_innodb.inc inclusion.
Added comments about different SP test files, table usage and
hint for bug test case style.
mysql-test/t/sp_trans.test:
Changed procedure name and delimiter setting to follow the style of the other SP test files.
...and added new test file, sp-threads, for multiple connection tests
(apart from the security tests that are in sp-security).
sql/sp.cc:
When removing an SP, invalidate the caches even if the
removing thread doesn't have one.
