MariaDB 5.3.0 Changelog p4

Download | Release Notes | Changelog (page: 1 2 3 4 5 6 ) | Overview of 5.3

Release date: 26 July 2011

  • Revision #2945 [merge] Tue 2011-03-15 12:30:48 -0700
    • Merge.
    • Revision #2928.3.1 Thu 2011-03-03 18:24:41 -0800
      • Fixed Bug #702322: The bug was a result of the fix for Bug #668644 that turned out to be not quite correct. A problem appeared with HAVING conditions containing more than one predicate. If a query with an ORDER BY clause uses such HAVING condition and the required order can be obtained with a range/index scan then the HAVING condition has to be pushed into two different formulas (items). To be able to do it we have to create a copy of the ANDOR structure of the pushed condition.
  • Revision #2944 [merge] Sun 2011-03-13 16:57:05 +0000
    • Merge in the fix for Bug #730604, and a corrected fix for Bug #719198, after Monty's review.
    • Revision #2933.2.1 Sun 2011-03-13 15:03:26 +0000
      • Fix Bug #719198, Bug #730604
      • Analysis ( Bug #719198 ):
        • The assert failed because the execution code for partial matching is designed with the assumption that NULLs on the left side are detected as early as possible, and a NULL result is returned before any lookups are performed at all.
        • However, in the case of an Item_cache object on the left side, null was not detected properly, because detection was done via Item::is_null(), which is not implemented at all for Item_cache, and resolved to the default Item::is_null() which always returns FALSE.
      • Solution:
        • Imlpement Item::is_null().
      • Analysis ( Bug #730604 ):
        • The method Item_field::is_null() determines if an item is NULL from its Item_field::field object. However, for Item_fields that represent internal temporary tables, Item_field::field represents the field of the original table that was the source for the temporary table (in this case t1.f3). Both in the committed test case, and in the original bug report the current value of t1.f3 is not NULL. This results in an incorrect count of NULLs for this column. As a consequence, all related Ordered_key buffers are allocated with incorrect sizes. Depending on the exact query and data, these incorrect sizes result in various crashes or failed asserts.
      • Solution:
        • The correct value of the current field of the internal temp table is in Item_field::result_field. This value is determined by Item::is_null_result().
  • Revision #2943 [merge] Sun 2011-03-13 03:50:14 -0700
    • Merge
    • Revision #2933.1.1 Mon 2011-03-07 22:09:21 -0800
      • Fixed Bug #729039: If join condition is of the form <t2.key>=<t1.no_key> then the server performs no index look-ups when looking for matching rows of t2 for the rows from t1 with t1.no_key=NULL. It happens because the function add_not_null_conds() injects an additional condition of the form IS NOT NULL(<t1.no_key>) into the WHERE condition. However if the join condition was of the form <t.key>=<outer_ref> no additional null rejecting predicate was generated. This could lead to extra records in the result set if the value of <outer_ref> happened to be NULL.
      • The new code injects null rejecting predicates of the form IS NOT NULL(<outer_ref>) and evaluates them before the first row the subquery is constructed.
  • Revision #2942 Sat 2011-03-12 16:18:02 +0000
    • Fix wrong use of compiler flag causing the build to fail in handlersocket.
  • Revision #2941 [merge] Sat 2011-03-12 05:14:10 -0800
  • Revision #2940 [merge] Fri 2011-03-11 16:08:26 +0100
    • merge
  • Revision #2939 [merge] Fri 2011-03-11 15:47:15 +0100
    • merge
  • Revision #2938 [merge] Fri 2011-03-11 15:20:24 +0100
  • Revision #2937 Wed 2011-03-09 17:55:00 +0200
    • Added item.real_type() for easy access to the underlaying types for Item_ref and Item_cache_wrapper() This allows us to simplify and speed up some tests and also remove get_cached_item()
  • Revision #2936 [merge] Wed 2011-03-09 15:47:59 +0200
    • Merge with 5.2
  • Revision #2935 Wed 2011-03-09 00:51:22 +0200
    • Added define to get rid of compiler warnings on system without DLOPEN
  • Revision #2934 Tue 2011-03-08 23:23:44 +0200
    • Fix Bug #719198
      • Analysis:
        • The assert failed because the execution code for partial matching is designed with the assumption that NULLs on the left side are detected as early as possible, and a NULL result is returned before any lookups are performed at all.
        • However, in the case of an Item_cache object on the left side, null was not detected properly, because detection was done via Item::is_null(), which is not implemented at all for Item_cache, and resolved to the default Item::is_null() which always returns FALSE.
      • Solution:
        • Use the property Item::null_value instead of is_null(), which is properly updated for Item_cache objects as well.
  • Revision #2933 [merge] Fri 2011-03-04 18:54:30 +0300
    • Merge in MRR interface fixes.
    • Revision #2931.1.1 Fri 2011-03-04 12:06:03 +0300
      • MRR interface: change range_info's type from char* to range_id_t typedef. The goals are:
        • cleaner code
        • ability to change from using pointers to offsets at some point
  • Revision #2932 Fri 2011-03-04 12:14:46 +0300
    • Make testcase pass on systems with lower_case_table_names=2.
      • Generally, we should use only small letters for table names but here it's easier to fix with one --replace.
  • Revision #2931 [merge] Fri 2011-03-04 01:30:25 +0300
    • Merge fix for Bug #693747
    • Revision #2928.2.1 Wed 2011-03-02 23:08:13 +0300
      • Bug #693747: Assertion multi_range_read.cc:908: int DsMrr_impl::dsmrr_init
        • Make DsMrr_impl::dsmrr_init() handle the case of
          1. 1st MRR scan using DS-MRR strategy (i.e. doing key sorting and rowid sorting)
          2. 2nd MRR scan getting a buffer that's too small to fit one key element and one rowid element, and so falling back to default MRR implementation
        • In this case, dsmrr_init() is invoked with {primary_handler, secondary_handler} initialized for DS-MRR scan and have to reset them to be initialized for the default MRR scan.
        • (attempt 2, with simplified testcase)
  • Revision #2930 [merge] Fri 2011-03-04 01:28:02 +0300
    • Merge Bug #707925.
    • Revision #2928.1.1 Fri 2011-03-04 00:54:10 +0300
      • Bug #707925: Wrong result with join_cache_level=6 optimizer_use_mrr = force (incremental, BKA join)
        • The problem was that Mrr_ordered_index_reader's interrupt_read() and resume_read() would save and restore 1) index tuple 2) the rowid (as bytes returned by handler->position()). Clustered primary key columns were not saved/restored.
        • They are not explicitly present in the index tuple (i.e. table->key_info[secondary_key].key_parts doesn't list them), but they are actually there, in particular table->field[clustered_primary_key_member].part_of_key(secondary_key) == 1. Index condition pushdown code [correctly] uses the latter as inidication that pushed index condition can refer to clustered PK members.
        • The fix was to make interrupt_read()/resume_read() to save/restore clustered primary key members as well, so that we get correct values for them when evaluating pushed index condition.
      • [3rd attempt: remove the debugging aids, fix comments in testcase]
  • Revision #2929 Thu 2011-03-03 23:48:31 +0200
    • Fix Bug #718763
      • Analysis:
        • The reason for the crash was that the inner subquery was executed via a scan on a final temporary table applied after all other operations. This final operation is implemented by changing the contents of the JOIN object of the subquery to represent a table scan over the temp table. At the same time query optimization of the outer subquery required evaluation of the inner subquery, which happened before the actual EXPLAIN. The evaluation left the JOIN object of the inner subquery in the changed state, where it represented a table scan over a temp table, and EXPLAIN crashed because the temp table is not associated with any table reference (TABLE_LIST object). The reason the JOIN was not restored was because its saving/restoration was controlled by the join->select_lex->uncacheable flag, which was not set in the case of materialization.
      • Solution:
        • In the methods Item_in_subselect::[single | row]_value_transformer() set: select_lex->uncacheable|= UNCACHEABLE_EXPLAIN;
        • In addition, for symmetry, change: master_unit->uncacheable|= UNCACHEABLE_EXPLAIN;
        • instead of UNCACHEABLE_DEPENDENT because if a subquery was not dependent initially, the changed methods do not change this fact. The subquery may later become correlated if it is transformed to an EXISTS query, but it may stay uncorrelated if executed via materialization.
  • Revision #2928 [merge] Tue 2011-03-01 10:22:22 +0300
    • Merge fix for Bug #725275
    • Revision #2925.1.1 Tue 2011-03-01 00:29:59 +0300
      • Bug #724275: Crash in JOIN::optimize in maria-5.3
        • Make equality-substitution-for-ref-access code in JOIN::optimize() treat join_tab->ref.key_copy correctly (in the way create_ref_for_key() has filled it).
  • Revision #2927 Mon 2011-02-28 17:27:41 -0800
    • Moved the test case for Bug #725050 into a new test file.
  • Revision #2926 Sun 2011-02-27 22:37:46 -0800
    • Fixed Bug #725050: The bug in the function print_keyuse() caused crashes if hash join could be used. It happened because the function ignored the fact that KEYUSE structures could be created for hash joins as well.
  • Revision #2925 [merge] Sun 2011-02-27 10:14:11 -0800
  • Revision #2924 [merge] Sun 2011-02-27 00:21:45 -0800
    • Merge
    • Revision #2900.2.2 Thu 2011-02-24 10:36:32 -0800
      • Made a newly added EXPLAIN platform independent.
    • Revision #2900.2.1 Wed 2011-02-23 22:23:12 -0800
      • BNLH algorithm always used a full table scan over the joined table even in the cases when there existed range/index-merge scans that were cheaper than the full table scan.
      • This was a defect/bug of the implementation of mwl #128. Now hash join can work not only with full table scan of the joined table, but also with full index scan, range and index-merge scans.
      • Accordingly, in the cases when hash join is used the column 'type' in the EXPLAINs can contain now 'hash_ALL', 'hash_index', 'hash_range' and 'hash_index_merge'. If hash join is coupled with a range/index_merge scan then the columns 'key' and 'key_len' contain info not only on the used hash index, but also on the indexes used for the scan.
  • Revision #2923 Sat 2011-02-26 23:09:58 +0300
  • Revision #2922 [merge] Fri 2011-02-25 21:45:21 +0300
    • Merge of fix for Bug #723822
    • Revision #2920.1.1 Fri 2011-02-25 21:43:57 +0300
      • Bug #723822: Crash in get_constant_key_infix with EXISTS ( SELECT .. DISTINCT )
        • Make get_constant_key_infix() take into account that there may be SEL_TREEs with type=SEL_ARG::MAYBE_KEY, which it cannot process, because they are not real ranges but rather indications that we might have been able to construct a range if we had values for some other tables' fields. (check_quick_select() already has such check)
  • Revision #2921 [merge] Fri 2011-02-25 20:32:18 +0200
  • Revision #2920 Thu 2011-02-24 23:32:00 +0300
    • Bug #724537: innodb_mysql.test fails with --embedded-server
      • Don't access THD::killed directly from ha_innodb.cc. This is forbidden because THD has different membership (and so, different member offsets) in regular and embedded server. Access must be done through thd_killed() function.
      • if we're interrupted by the user while in XtraDB, return the proper code.
  • Revision #2919 Wed 2011-02-23 14:46:16 +0200
    • Fixed build issues
      • Linking now with g++ instead of gcc with 'compile-dist' to solve problems with handlersocket/client
      • Fixed bug in heap tables when doing handler read next-prev over last row
  • Revision #2918 Wed 2011-02-23 02:06:58 +0200
    • Fixed compiler warnings and some test failures found by buildbot
  • Revision #2917 [merge] Tue 2011-02-22 11:30:51 +0200
    • Merge with main
    • Revision #2914.1.1 [merge] Tue 2011-02-22 11:15:47 +0200
      • Merge in new handler and handlersocket code into 5.3 main
  • Revision #2916 Mon 2011-02-21 20:17:26 +0100
    • Bug #53240 :Fixed dependency to prevent occasional situations where bison runs in parallel with the same input and output files
  • Revision #2915 Mon 2011-02-21 14:23:44 +0100
    • Fix remaining 5.2 buildbot problems:
      • Cherrypick Bug #688404 (PBXT crashes in debug/x64)
      • Fix unixism (rm -rf) in the test suite_timeout
      • Avoid plugin tests on Windows/embedded, plugins do not and will not work here
  • Revision #2914 Sat 2011-02-19 15:21:50 +0100
    • add newline at the end of file
  • Revision #2913 Sat 2011-02-19 15:16:31 +0100
    • Fix remaining windows (32 bit) warnings.
  • Revision #2912 Sat 2011-02-19 13:43:01 +0100
    • Fixed high-impact Windows 64bit warnings (at least 4000 of them)
  • Revision #2911 Sat 2011-02-19 02:42:08 +0100
    • Fixed DBUG_PRINT formatting (compile error on Linux with -Werror)
  • Revision #2910 Fri 2011-02-18 23:31:01 +0100
    • Fix numerous warnings introduced in the last pushes on Windows
  • Revision #2909 Fri 2011-02-18 22:38:22 +0100
    • Linker error, missing extern "C" for mi_killed.
    • Fix is to move extern "C" at the start of header file.
  • Revision #2908 [merge] Fri 2011-02-18 21:45:32 +0200
    • Merge with bugfix
    • Revision #2900.1.1 Fri 2011-02-18 17:43:59 +0200
      • Fix for Bug #711565 "Index Condition Pushdown can make a thread hold MyISAM locks as well as be unKILLable for long time"
        • In Maria/MyISAM: Release/re-acquire locks to give queries that wait on them a chance to make progress
        • In Maria/MyISAM: Change from numeric constants to ICP_RES values.
        • In Maria: Do check index condition in maria_rprev() (was lost in the merge/backport?)
        • In Maria/MyISAM/XtraDB: Check if the query was killed, and return immediately if it was.
        • Added new storage engine error: HA_ERR_ABORTED_BY_USER, for handler to signal that it detected a kill of the query and aborted
        • Authors: Sergey Petrunia & Monty
  • Revision #2907 Sat 2011-02-12 18:43:22 +0100
    • Workaround CMake bug http://www.vtk.org/Bug/view.php?id=11240 Huge static libraries like libmysqld might not build if /MACHINE flag is missing for librarian with the correct processor architecture.
    • Fix is to add /MACHINE flag for x64 builds
  • Revision #2906 Sat 2011-02-12 17:44:01 +0100
    • Fix test suite - adjust result files
  • Revision #2905 Sat 2011-02-12 17:17:19 +0100
    • Fix MySQL Bug #60057 : sel_arg_range_seq_next loops in optimized compilation/VS2010
      • When mariadb 5.3 is compiler with VS2010, several tests would enter infinite loop in sel_arg_range_seq_next(). The reason is compiler backend bug. This bug is not present in either VS2008 or VS2010 SP1 RC.
      • Workaround is to compile this function without most aggresive optimization flag (-Og ) using #pragma optimize ("g", {on|off}) for this version of MSVC compiler.
  • Revision #2904 Sat 2011-02-12 15:43:24 +0100
    • Fix Aria engine build
      • add forgotten source file
  • Revision #2903 [merge] Fri 2011-02-11 13:27:35 +0300
    • Merge: Bug #716293: "Range checked for each record" is not used if condition refers to outside of subquery
    • Revision #2891.1.1 Thu 2011-02-10 11:36:43 +0300
      • Bug #716293: "Range checked for each record" is not used if condition refers to outside of subquery
        • Assume that outside subquery references are known when doing "Range-checked-for-each-record" check.
  • Revision #2902 [merge] Wed 2011-02-09 11:22:26 -0800
    • Merge
    • Revision #2899.1.1 Fri 2011-02-04 19:06:35 -0800
      • Introduced optimizer switch flag 'optimize_join_buffer_size'.
      • When this flag is 'off' the size of the used join buffer is taken directly from the system variable 'join_buffer_size'.
      • When this flag is 'on' then the size of the buffer depends on the estimated number of rows in the partial join whose records are to be stored in the buffer.
      • By default this flag is set 'on'.
  • Revision #2901 Tue 2011-02-08 19:17:12 -0800
  • Revision #2900 [merge] Mon 2011-02-07 15:19:03 -0800
    • Merge
    • Revision #2897.1.1 Sat 2011-02-05 20:57:03 -0800
      • Fixed Bug #702403 that caused a crash on the tree for MWL#128 with the test case added by this patch.
      • The bug cannot be reproduced with the same test case for the main 5.3 tree because the backported fix for MySQL Bug #59696 masks the problem that causes the crash in the mentioned test case. It's not clear weather this fix masks this problem in all possible cases.
      • Anyway the patch for Bug #698882 introduced some inconsistent data structures that could contain indirect references to deleted object.
      • It happened when two Item_equal objects were merged and the Item_field list of the second object was joined to such list of the first object.
      • This operation required adjustment of the backward pointers in Item fields from the joined list. However the adjustment was missing and this caused crashes in the tree for MWL#128.
      • Now the backward pointers are set only when Item_equal items are completely built and are not changed anymore.
  • Revision #2899 Tue 2011-02-01 14:19:58 +0100
    • Fix compile errors:
      • declaration in the middle of the block in C file.
      • round() is only available in C99.
  • Revision #2898 Mon 2011-01-31 19:33:32 -0800
    • Back-ported the patch for MySQL Bug #59696 from mysql-5.6 code line.
    • The patch fixed the following optimizer defect: when performing substitution for best equal fields into where conditions to be able to do their evaluations as soon as possible the optimizer skipped conditions over views. That could lead to suboptimal execution of queries that used views.
    • Slightly changed the test case to demonstrate the performance improvements if this fix.
  • Revision #2897 [merge] Fri 2011-01-28 18:54:30 -0800
    • Merge
    • Revision #2893.1.2 Mon 2011-01-24 14:54:50 -0800
      • Post-second-review fixes for the patch that added the code allowing to use hash join over equi-join conditions without supporting indexes.
    • Revision #2893.1.1 [merge] Sun 2011-01-23 10:39:53 -0800
    • Merge
  • Revision #2896 Thu 2011-01-27 21:23:02 -0800
    • Fixed Bug #707827.
      • This bug could manifest itself when hash join over a varchar column with NULL values in some rows was used. It happened because the function key_buf_cmp erroneously returned FALSE when one of the joined key fields was null while the second was not.
      • Also fixed two other bugs in the functions key_hashnr and key_buf_cmp that could possibly lead to wrong results for some queries that used hash join over several columns with nulls.
      • Also reverted the latest addition of the test case for MySQL Bug #45092. It had been already backported earlier.
  • Revision #2895 Wed 2011-01-26 20:45:23 -0800
    • Fixed Bug #707848.
      • This was another bug in the patch for Bug #698882. The new code from this patch did not ensured that substitutions of fields for best equal fields were performed on all AND-OR levels. As a result substitutions for best fields in some predicates that had been used by the range optimizer were not actually performed while range plans could employ these substitutions. This could lead to inconsistent data structures and ultimately to a crash.
  • Revision #2894 Wed 2011-01-26 11:30:29 -0800
    • Fixed Bug #707555.
      • The bug was in the code of the patch fixing Bug #698882.
      • With improper casting the method store_key_field::change_source_field was called for the elements of the array TABLE_REF::key_copy that were either of a different type or not allocated at all. This caused crashes in some queries.
  • Revision #2893 Sat 2011-01-22 23:45:52 -0800
    • Fixed typo that caused printing 'range' instead of 'index_merge' as the type of sort_intersect scans.
  • Revision #2892 Fri 2011-01-21 09:56:55 +0200
    • Fix of reverting changes in depend_on list.
  • Revision #2891 Fri 2011-01-14 21:51:55 +0100
    • issue an informative error message for a common Aria problem when opening a table (incorrect block size)
  • Revision #2890 Fri 2011-01-14 21:02:51 +0100
    • compilation failures caused by adding new row format to Aria
  • Revision #2889 Fri 2011-01-14 12:05:46 +0100
    • Optimize use of SEARCH_SAVE_BUFF in Aria (less not-needed copies of key pages)
  • Revision #2888 Fri 2011-01-14 12:03:41 +0100
    • use bulk insert and repair by sort for unique keys in Aria and MyISAM in create_internal_tmp_table_from_heap()
      • (safe, as duplicates are impossible).
      • This gives a HUGE speed boost!
  • Revision #2887 Fri 2011-01-14 11:58:45 +0100
    • Added ha_write_tmp_row() for slightly faster write_row for internal temp tables.
    • This will also enable us in the future to collect statistics for writes to internal tmp tables.
  • Revision #2886 Fri 2011-01-14 11:54:39 +0100
    • Added support for NO_RECORD record format (don't store any row data) for Aria.
    • This makes the keys smaller (no row pointer) and gives us proper errors if we use the table wrongly.
  • Revision #2885 Fri 2011-01-14 11:43:42 +0100
    • use normal unique (HA_NOSAME) keys for expression cache temptables, not "uniques", that are hash-based keys.
  • Revision #2884 Fri 2011-01-14 11:37:23 +0100
    • Added to Aria better hash for packed numeric data for unique handling.
    • This was needed as the old code caused us to have LOTS of duplicate hash values when used by optimizer.
  • Revision #2883 Fri 2011-01-14 11:34:41 +0100
    • compare "Copying to tmp table" proc_info as a pointer, not as a string
  • Revision #2882 Fri 2011-01-14 11:31:09 +0100
    • Removed some old comments.
  • Revision #2881 [merge] Tue 2011-01-18 00:26:04 +0300

MariaDB 5.3.0 Changelog page: 1 2 3 4 5 6

Comments

Comments loading...
Content reproduced on this site is the property of its respective owners, and this content is not reviewed in advance by MariaDB. The views, information and opinions expressed by this content do not necessarily represent those of MariaDB or any other party.