Re: Support EXCEPT for TABLES IN SCHEMA publications

On Fri, Aug 14, 2026 at 11:29 AM shveta malik <shveta(dot)malik(at)gmail(dot)com> wrote:
>
> On Thu, Aug 13, 2026 at 12:52 PM Peter Smith <smithpb2250(at)gmail(dot)com> wrote:
> >
> >
> > 2b.
> > /In the partition case/For partitions/
> >
> > AFAICT this case is referring to something like: "FOR TABLE part,
> > TABLES IN SCHEMA EXCEPT (part_root)"
> >
> > But, isn't that just a variation of the 1st case issue? e.g. where
> > table "part" is not yet visible for later lookup of "root", then you
> > wont be able to check integrity of the partition tree regardless of
> > the up/down traversal logic, so I wasn't sure why this 2nd case was
> > separately mentioned at all.
>
> The second point is different from the first. Consider this case:
>
> CREATE PUBLICATION pub1 FOR s2.tab_part;
> ALTER PUBLICATION pub1 ADD TABLES IN SCHEMA s2 EXCEPT (TABLE tab_root);
>
> Here, the partition entry in pg_publication_rel is visible to the
> second command in publication_add_relation() and in
> check_publication_add_relation(). But the checks there are not
> sufficient to identify the error. If we try to detect the error while
> adding tab_root in publication_add_relation(),
> we would need to perform a full descendant search to determine whether
> any of its descendants are already present in pg_publication_rel. This
> downward traversal is what we are trying to avoid in
> publication_add_relation(). Geenrally we rely on ancestor-lookup and
> we want to stick to that instead of introducing a new logic.
> Thus the logic in CheckExceptNotInTableList() is needed here. It reads
> all explicitly added entries from pg_publication_rel, looks up their
> ancestors to find the root, and checks the EXCEPT entries against that
> root.
>
> This is my understanding, let's wait for Nisha's comments as well.
>

Yes, your understanding is correct, that is precisely the reason.

> I feel the comment about the second case should be moved to patch002,
> where it actually makes sense and explains why it is needed even after
> point 1.
>

Agreed. I’ll move the 2nd point to patch-002, as it relates to the
ALTER PUBLICATION command.

--
Thanks,
Nisha