database/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-08-08 11:22:35 +03:00
Code Activity

Fix for LP#612894 Some aggregate functions (such as MIN MAX) work incorrectly in subqueries after getting NULL value

Browse Source 
mysql-test/r/group_by.result:
  Added test that showed problems that no_rows_in_results() didn't work for expressions
mysql-test/r/subselect4.result:
  Test case for LP#612894
mysql-test/t/group_by.test:
  Added test that showed problems that no_rows_in_results() didn't work for expressions
mysql-test/t/subselect4.test:
  Test case for LP#612894
sql/item.h:
  Added restore_to_before_no_rows_in_result()
  Added function processor for no_rows_in_results() and restore_to_before_no_rows_in_results() to ensure it works with functions
  Fix that above functions are handled by Item_ref()
sql/item_func.h:
  Ensure that no_rows_in_results() and restore_to_before_no_rows_in_result() are called for all function arguments
sql/item_sum.cc:
  Added restore_to_before_no_rows_in_result() to restore settings after Item_sum_hybrid::no_rows_in_result() was called.
  This is needed to handle the case where we have made 'make_const()' on the item in opt_sum(), but the item will be reused again in a sub query.
  Ignore multiple calls to no_rows_in_result() as Item_ref is calling it twice.
sql/item_sum.h:
  Added restore_to_before_no_rows_in_result();
sql/sql_select.cc:
  Added reset of no_rows_in_result() for JOIN::reinit()
sql/sql_select.h:
  Added marker if no_rows_in_result() is called.
This commit is contained in:
Michael Widenius
2010-08-23 12:46:25 +03:00
parent 096666fcbb
commit b6fe4713fe
10 changed files with 170 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
sql/sql_select.h
View File

@@ -365,24 +365,26 @@ public:
the number of rows in it may vary from one subquery execution to another.
*/
bool no_const_tables;
bool no_rows_in_result_called;
/**
Copy of this JOIN to be used with temporary tables.
tmp_join is used when the JOIN needs to be "reusable" (e.g. in a subquery
that gets re-executed several times) and we know will use temporary tables
for materialization. The materialization to a temporary table overwrites the
JOIN structure to point to the temporary table after the materialization is
done. This is where tmp_join is used : it's a copy of the JOIN before the
materialization and is used in restoring before re-execution by overwriting
the current JOIN structure with the saved copy.
Because of this we should pay extra care of not freeing up helper structures
that are referenced by the original contents of the JOIN. We can check for
this by making sure the "current" join is not the temporary copy, e.g.
!tmp_join || tmp_join != join
tmp_join is used when the JOIN needs to be "reusable" (e.g. in a
subquery that gets re-executed several times) and we know will use
temporary tables for materialization. The materialization to a
temporary table overwrites the JOIN structure to point to the
temporary table after the materialization is done. This is where
tmp_join is used : it's a copy of the JOIN before the
materialization and is used in restoring before re-execution by
overwriting the current JOIN structure with the saved copy.
Because of this we should pay extra care of not freeing up helper
structures that are referenced by the original contents of the
JOIN. We can check for this by making sure the "current" join is
not the temporary copy, e.g. !tmp_join || tmp_join != join
We should free these sub-structures at JOIN::destroy() if the "current" join
has a copy is not that copy.
We should free these sub-structures at JOIN::destroy() if the
"current" join has a copy is not that copy.
*/
JOIN *tmp_join;
ROLLUP rollup; ///< Used with rollup
@@ -512,6 +514,7 @@ public:
optimized= 0;
cond_equal= 0;
group_optimized_away= 0;
no_rows_in_result_called= 0;
all_fields= fields_arg;
if (&fields_list != &fields_arg) /* Avoid valgrind-warning */