Top with group on Postgres about prql HOT 5 CLOSED

Thanks @PrettyWood !

Am I correct in thinking that we do want DISTINCT ON in postgres, but there's a bug in ignoring the take 2 transform?

from prql.

I guess so

from prql.

No, we don't want DISTINCT ON since that will pick only the first item in each GROUP BY partition.

The correct behavior would be to pick first two items in each partition. In SQL, this is only achievable by using ROW_NUMBER() and WHERE.

This is the problem here: https://github.com/prql/PRQL/blob/main/prqlc/prql-compiler/src/sql/srq/preprocess.rs#L172-L177

• we are checking if we should take_only_first,
• we correctly apply this condition to DISTINCT,
• we don't apply this condition to DISTINCT ON <-- this is the bug

from prql.

Oh, I didn't notice, you've opened a PR already.

from prql.

Oh, I didn't notice, you've opened a PR already.

Yeah and also realized my suggestion was wrong 😄

from prql.