On Wed, Apr 2, 2025 at 5:17 AM Alvaro Herrera <alvherre@alvh.no-ip.org> wrote:
> I don't quite love this behavior, but since there have been no
> complaints, I suppose it's okay and we should just do the same for
> not-nulls.
I don't understand the issue. It seems like the pg_dump output shown
here would recreate the catalog state.
> FWIW the part that I think you're not right on, is that constraints on
> partitioned tables never have local definitions. Even if you start with
> a constraint defined locally in the partition, the ATTACH operation will
> change its conislocal flag to false. So you can never "drop" it from
> the partition. For regular inheritance, we don't flip the conislocal
> flag to false, but you're still prevented from "dropping" the constraint
> from the child while the inheritance relationship exists (i.e. you can
> never set conislocal=false in such a case).
Hmm. I think this is different from attislocal/attinhcount.
--
Robert Haas
EDB: http://www.enterprisedb.com
