Thread

Re: plan shape work

Robert Haas <robertmhaas@gmail.com> — 2025-12-08T19:06:35Z

On Mon, Dec 1, 2025 at 3:31 PM Robert Haas <robertmhaas@gmail.com> wrote:
> So I'm wondering if we ought to step back and rethink a bit.
> subplanNames ensures that we assign a different name to every
> PlannerInfo, but it doesn't give you any help finding the PlannerInfo
> given the name, or enumerating all of the PlannerInfo objects that
> exist. If we went back to allroots, and just fixed the problem of
> temporary entries creeping into the list, the core code wouldn't
> really notice the difference, but I think extensions would have an
> easier time.
>
> Thoughts?

Here's a couple of patches to bring back allroots. The first one adds
allroots, makes choose_plan_name() use it, and adds assertions that
the contents of allroots and subplanNames correspond. The second one
removes subplanNames. I imagine that these would be collapsed for
commit, but testing 0001 without 0002 seems useful for peace of mind.

-- 
Robert Haas
EDB: http://www.enterprisedb.com