Add metadata validation to ~20 more SQL commands. Make sure that
these commands actually work in ps-protocol, since until now they
were enabled, but not carefully tested.
Fixes the ml003 bug found by Matthias during internal testing of the
patch.
mysql-test/r/ps_ddl.result:
Update test results (WL#4165)
mysql-test/t/ps_ddl.test:
Cover with tests metadata validation of 26 SQL statements.
sql/mysql_priv.h:
Fix the name in the comment.
sql/sp_head.cc:
Changed the way the observer is removed in case of stored procedures
to support validation prepare stmt from "call p1(<expr>)": whereas
tables used in the expression must be validated, substatements
of p1 must not.
The previous scheme used to silence the observer only in stored
functions and triggers.
sql/sql_class.cc:
Now the observer is silenced in sp_head::execute(). Remove it from
Sub_statement_state.
sql/sql_class.h:
Now the observer is silenced in sp_head::execute(). Remove it from
Sub_statement_state.
sql/sql_parse.cc:
Add CF_REEXECUTION_FRAGILE to 20 more SQLCOMs that need it.
sql/sql_prepare.cc:
Add metadata validation to ~20 new SQLCOMs that need it.
Fix memory leaks with expressions used in SHOW DATABASES and CALL
(and prepared statements).
We need to fix all expressions at prepare, since if these expressions
use subqueries, there are one-time transformations of the parse
tree that must be done at prepare.
List of fixed commands includes: SHOW TABLES, SHOW DATABASES,
SHOW TRIGGERS, SHOW EVENTS, SHOW OPEN TABLES,SHOW KEYS, SHOW FIELDS,
SHOW COLLATIONS, SHOW CHARSETS, SHOW VARIABLES, SHOW TATUS, SHOW TABLE
STATUS, SHOW PROCEDURE STATUS, SHOW FUNCTION STATUS, CALL.
Add comment to set_parameters().
sql/table.h:
Update comments.
when --master-data is used
When using the --master-data option with mysqldump, mysqldump uses
a FLUSH TABLES command. However, this statement got replicated to
the slave(s), which caused the slave(s) to block unnecessarily while
the FLUSH tables command completed.
Now, if the master-data option is set to one of the two "on" modes,
then use the "LOCAL" qualifier to ensure that it's not replicated.
client/mysqldump.c:
If master_data is set to one of the two modes, then insert "LOCAL"
to the command to FLUSH TABLES so that the slaves aren't told to
flush also.
mysql-test/r/mysqldump.result:
Output of mysqldump changed.
into zippy.cornsilk.net:/home/cmiller/work/mysql/bug15776-encore/my51-bug15776-encore
sql/field.cc:
Auto merged
sql/item_create.cc:
Auto merged
sql/share/errmsg.txt:
Auto merged
sql/unireg.h:
Auto merged
mysql-test/r/type_blob.result:
manual merge.
mysql-test/t/type_blob.test:
manual merge.
sql/sql_yacc.yy:
merge by hand.
into zippy.cornsilk.net:/home/cmiller/work/mysql/bug15776/my51-bug15776
mysql-test/r/type_blob.result:
Auto merged
mysql-test/t/type_blob.test:
Auto merged
sql/field.cc:
Auto merged
sql/unireg.h:
Auto merged
into zippy.cornsilk.net:/home/cmiller/work/mysql/mysql-5.0-build
mysql-test/r/type_blob.result:
Auto merged
mysql-test/t/type_blob.test:
Auto merged
sql/field.cc:
Auto merged
sql/unireg.h:
Auto merged
Make maximum blob size to be 2**32-1, regardless of word size.
Fix failure of timestamp with size of 2**31-1. The method of
rounding up to the nearest even number would overflow.
mysql-test/r/type_blob.result:
2**32-1 is not a special case for timestamp.
Test 2**32-1 and 2**64 as the reliable test points for both 32-
and 64-bit machines. I'd like to test 2**32, but that would make
tests that vary between architectures.
I'd like to generalize the tests by pulling the max blob size from
the server, and then "eval"ing N-1, N, and N+1 instead of all these
literal numbers, but I have not found a way to get UINT_MAX.
mysql-test/t/type_blob.test:
2**32-1 is not a special case for timestamp.
Test 2**32-1 and 2**64 as the reliable test points for both 32-
and 64-bit machines. I'd like to test 2**32, but that would make
tests that vary between architectures.
I'd like to generalize the tests by pulling the max blob size from
the server, and then "eval"ing N-1, N, and N+1 instead of all these
literal numbers, but I have not found a way to get UINT_MAX.
sql/field.cc:
Fix a bug where the round-to-even code for TIMESTAMP fields
failed where the size would overflow the size to zero and then
fail.
Also, since we silently truncate the size of TIMESTAMP fields, set
the maximum size we report is allowable to be the largest parsable
number.
sql/unireg.h:
Make BLOB size the maximum that the packed value in
field_blob::get_length() allows.
#12713 Error in a stored function called from a
SELECT doesn't cause ROLLBACK of statem
mysql-test/r/func_rollback.result:
Expected results
mysql-test/t/func_rollback.test:
New test
WL#4165 Prepared statements: validation
WL#4166 Prepared statements: automatic re-prepare
Fixes
Bug#27430 Crash in subquery code when in PS and table DDL changed after PREPARE
Bug#27690 Re-execution of prepared statement after table was replaced with a view crashes
Bug#27420 A combination of PS and view operations cause error + assertion on shutdown
The basic idea of the patch is to keep track of table metadata between
prepared statement prepare and execute. If some table used in the statement
has changed, the prepared statement is re-prepared before execution.
See WL#4165 and WL#4166 contents and comments in the code for details
of the implementation.
include/my_global.h:
Remove 'register' keyword to avoid warnings when swapping large structures
that don't fit into a register. Any modern compiler is capable of placing
a variable in a register when that would benefit performance.
mysql-test/r/ps_1general.result:
Update test results: since now we re-prepare automatically,
more correct results are produced in prepare-ddl-execute scenario.
mysql-test/r/query_cache_merge.result:
Ensure that the table definition cache is large enough for
the test to pass in --ps-protocol
mysql-test/r/trigger.result:
Update test results to reflect automatic statement reprepare.
mysql-test/t/disabled.def:
Enable ps_ddl.test, which now passes.
mysql-test/t/ps_1general.test:
Since now we re-execute prepared statements after DDL successfully,
change the test to produce repeatable results. Remove expectancy of
an error in one place where now we automatically reprepare the prepared
statement.
mysql-test/t/query_cache_merge.test:
Ensure the table definition cache is large enough for the test to pass
in --ps-protocol
mysql-test/t/trigger.test:
Sinc
sql/item.cc:
Implement Item_param "copy" functionality, used at re-prepare of
a prepared statement.
We copy the type of the original parameter, and move the assigned value,
if any. Sic, the value is "moved", since it can be quite big --
e.g. in case we deal with a LONG DATA parameter.
It's essential to move the value from the old parameter since
at the time of re-prepare the client packet with the necessary information
may be not available.
sql/item.h:
Declare a new method used for reprepare.
sql/my_decimal.h:
Implement "swap()" functionality of class my_decimal to be
able to easily swap two decimal values.
sql/mysql_priv.h:
Declare enum_metadata_type.
sql/mysqld.cc:
Implement a status variable for the number of reprepared statements.
sql/sql_base.cc:
Implement metadata version validation.
sql/share/errmsg.txt:
Add two new error messages: ER_NEED_REPREPARE and ER_PS_REBIND.
The first error (theoretically) never reaches the user.
It is issued by the metadata validation framework when a metadata version
has changed between prepare and execute. Later on it's intercepted
and the statement is automatically re-prepared. Only if the error
has occurred repeatedly MAX_REPREPARE_ATTEMTS (3) times do we
return it to the user.
The second error is issued when after re-prepare we discover
that the metadata we sent over to the client using the binary
protocol differs drammatically from the new result set metadata
that the reprepared statement produces (e.g. number of result
set columns is different).
sql/sql_class.cc:
Implement metadata version validation framework.
sql/sql_class.h:
Declarations for metadata version validation framework.
sql/sql_parse.cc:
Mark commands for which we must invalidate and reprepare a prepared
statement when metadata has changed.
sql/sql_prepare.cc:
Implement WL#4165 and WL#4166 (limited support of metadata validation
and re-prepare).
sql/table.h:
Implement metadata validation.
tests/mysql_client_test.c:
Add a test case for WL#4166
Problem: SELECTs prohibited for a transactional SE in autocommit mode
if read_only is set.
Fix: allow them.
mysql-test/r/read_only_innodb.result:
Fix for bug #35732: read-only blocks SELECT statements in InnoDB
- test result.
mysql-test/t/read_only_innodb.test:
Fix for bug #35732: read-only blocks SELECT statements in InnoDB
- test case.
sql/handler.cc:
Fix for bug #35732: read-only blocks SELECT statements in InnoDB
- in autocommit mode thd->transaction.all list is empty thus
is_real_trans set to TRUE for any SELECTs, so using it in the
"read_only" check is insufficient.
ha_check_and_coalesce_trx_read_only() changed to return number
of engines with read-write changes. This value is used in the
"read-only" check and checks for GLOBAL READ LOCK.
sql/lock.cc:
Fix for bug #35732: read-only blocks SELECT statements in InnoDB
- added assert(protect_against_global_read_lock) before decreasing,
in order to catch (uint) 0 - 1 situation due to wrong
wait_if_global_read_lock()/start_waiting_global_read_lock() call
sequence.
into zippy.cornsilk.net:/home/cmiller/work/mysql/mysql-5.0-build
sql/item_create.cc:
Auto merged
sql/share/errmsg.txt:
Auto merged
sql/sql_yacc.yy:
Auto merged
into zippy.cornsilk.net:/home/cmiller/work/mysql/mysql-5.0-build
sql/field.cc:
Auto merged
sql/share/errmsg.txt:
Auto merged
sql/sql_yacc.yy:
Auto merged
sql/unireg.h:
Auto merged
mysql-test/r/type_blob.result:
Manual merge.
mysql-test/t/type_blob.test:
Manual merge.
Corrects build problems embedded on Windows
Makefile.am:
Install .sym or mysqld-debug if exists
query_cache_debug.test, query_cache_debug.result:
Set more resonable query cache size (bug#35749)
CMakeLists.txt:
Added missing stacktrace.c
mysql-test/r/query_cache_debug.result:
Set more resonable query cache size (bug#35749)
mysql-test/t/query_cache_debug.test:
Set more resonable query cache size (bug#35749)
libmysqld/CMakeLists.txt:
Added missing stacktrace.c
sql/Makefile.am:
Install .sym or mysqld-debug if exists
sql/mysqld.cc:
Corrects build problems embedded on Windows
Based on contributed patch from Martin Friebe, CLA from 2007-02-24.
The parser lacked support for field sizes after signed long,
when it should extend to 2**32-1.
Now, we correct that limitation, and also make the error handling
consistent for casts.
---
Fix minor complaints of Marc Alff, for patch against B-g#15776.
---
Merge zippy.cornsilk.net:/home/cmiller/work/mysql/bug15776/my50-bug15776
into zippy.cornsilk.net:/home/cmiller/work/mysql/bug15776/my51-bug15776
---
Merge zippy.cornsilk.net:/home/cmiller/work/mysql/bug15776/my51-bug15776
into zippy.cornsilk.net:/home/cmiller/work/mysql/mysql-5.1-build
---
testing
mysql-test/r/type_blob.result:
Verify that blobs may be created with the size that is already
documented.
Additionally, test the limits of several other types.
mysql-test/t/type_blob.test:
Verify that blobs may be created with the size that is already
documented.
Additionally, test the limits of several other types.
---
Drop table in case we start from a bad state.
sql/field.cc:
atoi() insufficient to gauge the length of some fields. Change
it to strtoul().
sql/item_create.cc:
atoi() insufficient to gauge the length of some fields. Change
it to strtoul().
If a casted length is too long, raise an error.
sql/share/errmsg.txt:
Change ER_TOO_BIG_FIELDLENGTH so that it can accept sizes larger
than 2**15 -- instead, 2**32.
---
Manual merge.
sql/sql_yacc.yy:
Make lengths take, in addition to NUM, LONG_NUM, ULONGLONG_NUM,
and DECIMAL_NUM.
---
yacc/bison is left-recursive, so FIXME statement is wrong.
---
Manual merge and reformatting.
sql/unireg.h:
Define new constant.
into zippy.cornsilk.net:/home/cmiller/work/mysql/bug15776/my51-bug15776
mysql-test/r/type_blob.result:
Auto merged
mysql-test/t/type_blob.test:
Auto merged
sql/field.cc:
Auto merged
sql/unireg.h:
Auto merged
sql/item_create.cc:
Manual merge.
sql/share/errmsg.txt:
Manual merge.
sql/sql_yacc.yy:
Manual merge.
mysql-test/r/rpl_packet.result:
Result change.
mysql-test/t/rpl_packet.test:
Setting net_buffer_length correctly for the test instead of relying on it
being set correctly. Waiting for slave to stop after issuing a SLAVE STOP
and waiting for slave to start when issuing a SLAVE START to prevent race
conditions causing test failure.
tables is not zero any more. For row-based logging, there is an extra commit
for sending rows changed by the statement to the binary log.
mysql-test/include/commit.inc:
For row-based logging, an extra commit is done for each statement to commit
non-transactional changes to the binary log.
mysql-test/r/commit_1innodb.result:
Result change.
into magare.gmz:/home/kgeorge/mysql/work/merge-build-5.1-bugteam
sql/log.cc:
Auto merged
sql/slave.cc:
Auto merged
sql/share/errmsg.txt:
merged 5.1-main to 5.1-bugteam
If a binlog file is manually replaced with a namesake directory the internal purging did
not handle the error of deleting the file so that eventually
a post-execution guards fires an assert.
Fixed with reusing a snippet of code for bug@18199 to tolerate lack of the file but no other error
at an attempt to delete it.
The same applied to the index file deletion.
The cset carries pieces of manual merging.
mysql-test/r/binlog_index.result:
new results
mysql-test/r/ctype_big5.result:
results changed
mysql-test/suite/binlog/r/binlog_auto_increment_bug33029.result:
new results
mysql-test/suite/binlog/r/binlog_stm_blackhole.result:
new results
mysql-test/suite/binlog/t/binlog_auto_increment_bug33029.test:
cleanup. still todo: to let the test run multiple times w/o restarting the server (just ./mtr test)
mysql-test/suite/rpl/r/rpl_stm_auto_increment_bug33029.result:
results changed
mysql-test/suite/rpl/t/rpl_stm_auto_increment_bug33029.test:
guarding the test with statement format condition as the logics of the test requires
sql/log.cc:
two changes. One for the bug, other manual merge.
The bug change needs MYF(0) to pass to my_delete because not all error out of the function
are critical. The finer check is done on the caller of my_delete similarly how it was implemented
for bug@18199 fixes. Non-existance of a file is not a critical error.
sql/sql_class.cc:
manual merge, removing extra automatically brought hunk.
