Horton

Re: Teaching planner to short-circuit empty UNION/EXCEPT/INTERSECT inputs

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

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: David Rowley <dgrowleyml@gmail.com>
Cc: PostgreSQL Developers <pgsql-hackers@lists.postgresql.org>
Date: 2025-10-07T01:48:35Z
Lists: pgsql-hackers 
David Rowley <dgrowleyml@gmail.com> writes:
> The reason I didn't go down the route of SETOP_VAR was that it's still
> a hack, it's just making it look a bit more official.  I suppose the
> correct way to fix all this and get rid of the varno==0 stuff forever
> is to have a proper top-level RTE for the top-level set operation and
> make it so each child is an OTHER_MEMBER rel at that query level. It
> felt like going a bit too far to do something like that to fix this
> bug, so I didn't explore that further.

Yeah, I think "more RTEs" is the ultimate solution here, but it's
never risen to the top of my to-do list either.  I was kind of
thinking about an RTE per set-op child, though.  Not sure if one
for the top-level op, or one for an intermediate op, would help;
though it's certainly possible it would.

			regards, tom lane

Commits

  1. Fix UNION planner estimate_num_groups with varno==0

  2. Fix possible usage of incorrect UPPERREL_SETOP RelOptInfo

  3. Teach planner to short-circuit EXCEPT/INTERSECT with dummy inputs

  4. Fix incorrect targetlist in dummy UNIONs

  5. Teach UNION planner to remove dummy inputs