Sergey Vojtovich added a comment - 2015-03-12 13:39 - edited Reply from Kentoku: Please check only this point.   diff -Narup ./008_mariadb-10.0.15-partition_direct_update/sql/item_sum.h ./009_mariadb-10.0.15-dbg/sql/item_sum.h --- ./008_mariadb-10.0.15-partition_direct_update/sql/item_sum.h 2014-11-22 06:02:54.000000000 +0900 +++ ./009_mariadb-10.0.15-dbg/sql/item_sum.h 2014-11-29 17:02:32.059795158 +0900 @@ -1193,6 +1193,7 @@ public: enum Sumfunctype sum_func () const { return UDF_SUM_FUNC; } virtual bool have_field_update(void) const { return 0; }   + virtual void no_rows_in_result() { clear(); } void clear(); bool add(); void reset_field() {};   When I tried using aggregate type UDF, it didn't work. So I tried to add this line into Item_udf_sum class as MySQL 5.1's Item_sum class. I will try to reproduce it.

Sergey Vojtovich added a comment - 2015-03-25 09:29 In simple (non-distinct) case Item_udf_sum::clear() seem to be the only thing done currently: Item_sum::no_rows_in_result() Item_sum::aggregator_clear() Aggregator_simple::clear() Item_udf_sum::clear() Distinct case seem to be impossible with UDF's. So all we have now should be fully compatible with this patch.

Sergey Vojtovich added a comment - 2015-03-25 09:30 Closing per discussion with Kentoku as the problem statement is lost now.