> Robert Haas <robertmhaas@gmail.com> writes:
> > On Sun, Aug 4, 2013 at 6:20 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> >> I think it's time to bite the bullet and *not* pass back completed paths.
> >> What's looking more attractive now is to just pass back the top-level
> >> RelOptInfo ("final_rel" in query_planner()).
>
> > I tend to think this is a pretty good plan.
>
> I looked around a little more and noted that this would complicate the
> special-case handling of an empty join tree (viz, "SELECT 2+2"). Right now
> query_planner() just has to make the appropriate Result path and it's done.
> We'd have to create a dummy RelOptInfo representing an empty set of relations,
> which is a bit weird but probably not too unreasonable when all's said and
done.
I think this is reasonable.
Thanks,
Best regards,
Etsuro Fujita