Spark sum can overflow

[jinchengchenghh]

No description provided.

[rui-mo]

I found the behaviour may depend on a Spark config to throw or not. Maybe we can use this fix for now, and in the long term, we can consider to implement sum under spark sql folder to resolves the differences.

[rui-mo]

What is the type of reuslt? Maybe accumulated type should be used for the result here.

[rui-mo]

Looks like the accumalted types for int8/int16/int32 are all int64. Maybe we only need to handle int64 here.

          case TypeKind::TINYINT:
            return std::make_unique<T<int8_t, int64_t, int64_t>>(BIGINT());
          case TypeKind::SMALLINT:
            return std::make_unique<T<int16_t, int64_t, int64_t>>(BIGINT());
          case TypeKind::INTEGER:
            return std::make_unique<T<int32_t, int64_t, int64_t>>(BIGINT());
          case TypeKind::BIGINT:
            return std::make_unique<T<int64_t, int64_t, int64_t>>(BIGINT());

[jinchengchenghh]

Fixed

[jinchengchenghh]

Yes, you are right, it is controled by ansi enabled

[jinchengchenghh]

Mark here, after switch to spark function, we should check AverageAggregate and select add numeric overflow format