Tor Didriksen a écrit, Le 16.05.2011 15:01:
> On 2011-05-16 14:44, Guilhem Bichot wrote:
>> So, the only usage of Array I found is in sql_select.h:
>> Array<Item_exists_subselect> sj_subselects;
>> which could be translated to using your new class with "object type"
>> == "pointer to Item_exists_subselect" ...?
>
> indeed:
>
> @@ -4219,11 +4219,14 @@
> - prefer correlated subqueries over uncorrelated;
> - prefer subqueries that have greater number of outer tables;
> */
> - sj_subselects.sort(subq_sj_candidate_cmp);
> + my_qsort(sj_subselects.begin(),
> + sj_subselects.size(), sj_subselects.element_size(),
> + reinterpret_cast<qsort_cmp>(subq_sj_candidate_cmp));
> + // sj_subselects.sort(subq_sj_candidate_cmp);
can delete the line after //
> === modified file 'sql/sql_select.h'
> --- sql/sql_select.h 2011-05-16 11:48:23 +0000
> +++ sql/sql_select.h 2011-05-16 12:57:28 +0000
> @@ -56,6 +56,7 @@
> sj_pred_no(UINT_MAX)
> {}
>
> +
> Key_use(TABLE *table_arg, Item *val_arg, table_map used_tables_arg,
> uint key_arg, uint keypart_arg, uint optimize_arg,
> key_part_map keypart_map_arg, ha_rows ref_table_rows_arg,
> @@ -1888,7 +1889,7 @@
> bool union_part; ///< this subselect is part of union
> bool optimized; ///< flag to avoid double optimization in EXPLAIN
>
> - Array<Item_exists_subselect> sj_subselects;
> + Mem_root_array<Item_exists_subselect*, true> sj_subselects;
>
> /* Temporary tables used to weed-out semi-join duplicates */
> List<TABLE> sj_tmp_tables;
> @@ -1913,7 +1914,7 @@
> select_result *result_arg)
> : keyuse(thd_arg->mem_root),
> fields_list(fields_arg),
> - sj_subselects(thd_arg->mem_root, 4)
> + sj_subselects(thd_arg->mem_root)
The only thing I wonder is: the old code required a mem_root parameter
in two calls: constructor, and append(); so it was theoretically
possible that different mem_roots would be used accross the life of the
array (there is the "statement preparation" mem_root, the "statement
execution" mem_root)...
However, the only append() is in JOIN::prepare(), which happens at
"statement preparation", so I guess the THD's mem_root is then the same
as during construction (which is when JOIN is allocated).
Using two different mem_roots is also unlikely, as freeing one mem_root
could make the array half-freed, so I imagine this doesn't happen now.
Ok to push (as separate patch). Please also delete the Array
declaration, no need to tempt someone to use it again.
