Re: [PATCH v1] [BUG #19516] Skip whole-row projection shortcut for OLD/NEW returning type

Dean Rasheed <dean.a.rasheed@gmail.com>

From: Dean Rasheed <dean.a.rasheed@gmail.com>
To: Marko Grujic <marko.grujic@enterprisedb.com>
Cc: pgsql-hackers@lists.postgresql.org, Marko Grujic <markoog@gmail.com>
Date: 2026-06-10T12:47:32Z
Lists: pgsql-hackers
On Wed, 10 Jun 2026 at 13:23, Marko Grujic
<marko.grujic@enterprisedb.com> wrote:
>
> Agreed that this is a better solution than the present patch.
>
> That said what do you think about retrofitting GetNSItemByRangeTablePosn to accept VarReturningType too (other callers can pass VAR_RETURNING_DEFAULT)?

I don't like changing GetNSItemByRangeTablePosn() in back-branches
because some external code might be using it, though I didn't find any
examples on PGXN.

Some users of GetNSItemByRangeTablePosn() look fine, so there's no
need to change them -- for example, the MERGE parsing code, which
isn't starting from a Var, and isn't processing something that could
be in a RETURNING list.

OTOH, I think ExpandRowReference() does need updating -- at least the
comment suggests that (old.*).* will go through it, rather than
ParseComplexProjection(), so wouldn't be fixed by your original patch.

The one I'm unsure about is coerce_record_to_complex(). I can't manage
to come up with an example that breaks it, but it certainly looks like
it should be updated.

Regards,
Dean



Commits

  1. Fix parsing of parenthesised OLD/NEW in RETURNING list.