On 2025-Mar-24, jian he wrote:
> hi.
> you may like the attached. it's based on your idea: attnotnullvalid.
This is quite close to what I was thinking, yeah. I noticed a couple of
bugs however, and ended up cleaning up the whole thing. Here's what I
have so far. I'm not sure the pg_dump bits are okay (apart from what
you report below) -- I think it's losing the constraint names, which is
of course unacceptable.
I think the warnings about creating constraints as valid when
originating as invalid are unnecessary at this point. We should add
those, or not, for all constraint types, not just not-null. That's IMO
a separate discussion.
> I came across a case, not sure if it's a bug.
> CREATE TABLE ttchk (a INTEGER);
> ALTER TABLE ttchk ADD CONSTRAINT cc check (a is NOT NULL) NOT VALID;
> CREATE TABLE ttchk_child(a INTEGER) INHERITS(ttchk);
> ttchk_child's constraint cc will default to valid,
> but pg_dump && pg_restore will make ttchk_child's constraint invalid.
> since it's an existing behavior, so not-null constraint will align with it.
Hmm, yes, such pg_dump behavior would be incorrect. I'll give the
pg_dump code another look tomorrow.
--
Álvaro Herrera 48°01'N 7°57'E — https://www.EnterpriseDB.com/
