Hi!
Please see my comments inline.
on 15.04.2009 16:00 Anurag Shekhar wrote:
> #At file:///home/anurag/mysqlsrc/mysql-5.0-bugteam-39918/ based on
> revid:sergey.glukhov@stripped
>
> 2725 Anurag Shekhar 2009-04-15
> Bug #39918 memory (heap) engine crashing with b-tree index and DELETE with seg
> fault
>
> While evaluating self join with delete server uses two instance of handler
> (with both holding reference to same file). During query evaluation last_pos
> is cached in handler to improve search performance. But in the mean while
> second handler is used to delete the node last_pos is holding. This makes
> the last_pos an invalid and when accessed results in segmentation fault.
Please explain the real problem in more general way (as Svoj suggested).
> === modified file 'mysql-test/t/heap_btree.test'
> --- a/mysql-test/t/heap_btree.test 2007-10-19 20:04:30 +0000
> +++ b/mysql-test/t/heap_btree.test 2009-04-15 11:00:45 +0000
> @@ -253,5 +253,13 @@ insert into t1 values (1, 1), (3, 3), (2
> select * from t1 where a is null;
> drop table t1;
>
> +#
> +# bug#39918 - memory (heap) engine crashing while executing self join with delete
> +#
would be nice to
--echo #
--echo # Bug#39918 - memory (heap) engine crashing while executing self join with delete
--echo #
to have the header in the result file as well.
> === modified file 'sql/sql_delete.cc'
> --- a/sql/sql_delete.cc 2009-03-27 05:19:50 +0000
> +++ b/sql/sql_delete.cc 2009-04-15 11:00:45 +0000
> @@ -503,6 +503,11 @@ int mysql_multi_delete_prepare(THD *thd)
> }
> }
> }
> + /*
> + Reset the exclude flag to false so it doesn't interfare
> + with further calls to unique_table
> + */
> + lex->select_lex.exclude_from_table_unique_test= FALSE;
> DBUG_RETURN(FALSE);
> }
>
> @@ -538,11 +543,23 @@ multi_delete::initialize_tables(JOIN *jo
> DBUG_RETURN(1);
>
> table_map tables_to_delete_from=0;
> + delete_while_scanning= 1;
> for (walk= delete_tables; walk; walk= walk->next_local)
> + {
> tables_to_delete_from|= walk->table->map;
> + if (unique_table (thd, walk, join->tables_list, false))
a needless space after "unique_table".
the above "if" should be changed to
if (delete_while_scanning &&
unique_table(...))
> + {
> + /*
> + If the table we are going to delete from appears
> + in join, we need to defer delete. So the delete
> + doesn't interfers with the scaning of results.
> + */
the comment should be aligned properly.
Thanks,
Ramil.
