On Mon, 10 Mar 2025 at 19:11, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>
> I wrote:
> > Yeah, I think we can likely get away with that. We cannot back-patch
> > the changes that added relid to the outfuncs/readfuncs representation,
> > which means that the RTE's relid won't propagate to parallel workers,
> > but I don't see why they'd need it. We only need that info to get
> > as far as planning. I've not tried though.
>
> OK, the attached patch for v15 passes check-world, with or without
> force_parallel_mode. I'm inclined to commit the rewriteHandler.c
> and parsenodes.h bits in a separate patch for commit log visibility.
>
That looks good to me, on a quick read-through.
However, that's not quite the end of it -- preprocess_function_rtes()
/ inline_set_returning_function() can turn a function RTE into a
subquery RTE, leading to a similar problem:
create table foo (a int, b int);
insert into foo values (1,2);
create or replace function f() returns setof foo as
$$ select * from foo offset 0 $$ language sql stable;
update foo set b = f.b from f() as f(a,b) where f.a = foo.a returning f;
ERROR: attribute 3 of type record has wrong type
DETAIL: Table has type record, but query expects foo.
I tried looking for other places that change an RTE's rtekind, and
haven't managed to find any other problems, but may have missed
something.
Regards,
Dean
