Re: SRF patch (was Re: [HACKERS] troubleshooting pointers)

Joe Conway <mail@joeconway.com> writes:
> Here's the patch, per my post to HACKERS.
> It builds cleanly on my dev box, and passes all regression tests.

I've committed this with some revisions.  The VIEW cases you were
worried about seem to work now.  I think you'll find that
single-FROM-item cases generally work, and it's time to start worrying
about joins (ie, rescans).

Parameters also need thought.  This should be rejected:

regression=# select * from foo, foot(fooid) z where foo.f2 = z.f2;
server closed the connection unexpectedly

On the other hand, IMHO this should work:

regression=# select * from foo where f2 in
regression-# (select f2 from foot(foo.fooid) z where z.fooid = foo.fooid);
server closed the connection unexpectedly

and here again rescanning is going to be critical.

            regards, tom lane

PS: test case for above:

create table foo(fooid int, f2 int);
insert into foo values(1, 11);
insert into foo values(2, 22);
insert into foo values(1, 111);

create function foot(int) returns setof foo as '
select * from foo where fooid = $1' language sql;