Horton

Re: BUG #17564: Planner bug in combination of generate_series(), unnest() and ORDER BY

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Richard Guo <guofenglinux@gmail.com>
Cc: Martijn van Oosterhout <kleptog@gmail.com>, PostgreSQL mailing lists <pgsql-bugs@lists.postgresql.org>
Date: 2022-08-03T14:09:12Z
Lists: pgsql-bugs 
Richard Guo <guofenglinux@gmail.com> writes:
> -               if (IS_SRF_CALL((Node *) em->em_expr))
> +               if (expression_returns_set((Node *) em->em_expr))

Yeah.  I think we need to go further: placed in optimizer.h as
it is, IS_SRF_CALL is just an invitation to writing broken code.
I think the calls in tlist.c are legit because that code is
precisely concerned with breaking down nests of SRFs; but
everywhere else, if you write IS_SRF_CALL you're doing it wrong.
We should take that macro out of the header altogether.

I kind of wonder whether IS_SRF_CALL is a good idea at all.
It'd be more maintainable to have an expression_returns_set_one_level,
or something like that, placed beside expression_returns_set.

			regards, tom lane

Commits

  1. Add proper regression test for the recent SRFs-in-pathkeys problem.

  2. Fix incorrect tests for SRFs in relation_can_be_sorted_early().

  3. Optimize order of GROUP BY keys