Sergey Petrunya wrote:
> Hi!
>
> I can't offer any testcases but I think this patch has several issues. See
> below.
(...)
>> + */
>> + while ((item= it++))
>> + {
>> + field_tab= item->field->table->reginfo.join_tab;
>> + if (!(field_tab->sj_strategy == SJ_OPT_MATERIALIZE ||
>> + field_tab->sj_strategy == SJ_OPT_MATERIALIZE_SCAN))
> This is a wrong way to check if a field is inside SJ-Materialization nest. The
> condition is true only for the first table in SJ-Materialization nest, while we
> need to catch *any* SJ-Mat-inner table.
>
> the correct way to check this is as follows:
>
> field_tab->pos_in_table_list->embedding &&
> field_tab->pos_in_table_list->embedding->sj_mat &&
> field_tab->pos_in_table_list->embedding->sj_mat->is_used
>
Sergey's way of checking if the table is inside a SJM nest, together with this
rewrite of the following for-look:
/*
It's a field from an materialized semi-join. We can substitute it only
for a field from the same semi-join.
*/
JOIN_TAB *first;
JOIN *join= field_tab->join;
uint tab_idx= field_tab - field_tab->join->join_tab;
/* Find first table of this semi-join. */
for (int i=tab_idx; i >= join->const_tables; i--)
{
if (join->best_positions[i].sj_strategy == SJ_OPT_MATERIALIZE_LOOKUP ||
join->best_positions[i].sj_strategy == SJ_OPT_MATERIALIZE_SCAN)
{
//We have found the first table inside this semi-join nest
first= join->join_tab + i;
break;
}
}
fixes BUG#50019 as well. Note that
for (int i=tab_idx; i >= join->const_tables; i--) {}
does not stop when i gets below 0 if const_tables == 0. You'll probably have to
cast join->const_tables to int or even better: change the type of const_tables
to int since we're unlikely to encounter that many tables anyway ;)
I suggest BUG#50019 should be fixed as part of this bug and be closed as a
duplicate.
--
Jørgen Løland
