maria-developers team mailing list archive
Mailing list archive
Re: MDEV-4978 - Server cursor is broken with blobs in the select list
06.09.2013, в 15:29, Sergei Golubchik <serg@xxxxxxxxxxx> написал(а):
> Hi, Sergey!
> On Sep 06, Sergey Vojtovich wrote:
>> revno: 3878
>> revision-id: svoj@xxxxxxxxxxx-20130906111656-oy2c3pwy2o1q0zqn
>> parent: wlad@xxxxxxxxxxxxxxxx-20130903204512-hgwu77prle0lw7a3
>> committer: Sergey Vojtovich <svoj@xxxxxxxxxxx>
>> branch nick: 5.5-mdev4978
>> timestamp: Fri 2013-09-06 15:16:56 +0400
>> MDEV-4978 - Server cursor is broken with blobs in the select list,
>> ORDER BY does not work
>> Use "dynamic" row format (instead of "block") for MARIA internal
>> temporary tables created for cursors.
>> With "block" row format MARIA may shuffle rows, with "dynamic" row
>> format records are inserted sequentially (there are no gaps in data
>> file while we fill temporary tables).
>> This is needed to preserve row order when scanning materialized cursors.
>> === modified file 'sql/sql_select.cc'
>> --- a/sql/sql_select.cc 2013-09-03 14:41:07 +0000
>> +++ b/sql/sql_select.cc 2013-09-06 11:16:56 +0000
>> @@ -15373,7 +15374,8 @@ create_tmp_table(THD *thd, TMP_TABLE_PAR
>> if (thd->is_fatal_error) // If end of memory
>> goto err; /* purecov: inspected */
>> share->db_record_offset= 1;
>> - table->used_for_duplicate_elimination= (param->sum_func_count == 0 &&
>> + table->used_for_duplicate_elimination= keep_row_order ||
>> + (param->sum_func_count == 0 &&
>> (table->group || table->distinct));
> That's very confusing. Why keep_row_order means that the table is "used
> for duplicate elimination"?
> I had to check the code to see that used_for_duplicate_elimination
> affects the row format in Aria. This is not obvious too.
> And it forces the row format to be DYNAMIC for a very different reason,
> not for preserving row order.
> Perhaps it'd be better to rename used_for_duplicate_elimination to
> "force_dynamic_Aria_row_format" ? A prettier solution would be to
> introduce another flag: table->keep_row_order.
I submitted another patch with the "prettier" solution. Though I tend to think that it'd
be better not to fill TABLE object with things like keep_row_order and
used_for_duplicate_elimination - they can go as an argument to