Horton

Thread

Commits

  1. SET NOT NULL: Call object-alter hook only after the catalog change

  2. Add pg_constraint rows for not-null constraints

  1. Order of InvokeObjectPostAlterHook within ATExecSetNotNull

    Artur Zakirov <zaartur@gmail.com> — 2026-03-18T16:21:06Z

    Hello hackers!

The commit 14e87ff added pg_constraint rows for not-null [1]. The
consequence of the patch is that it changed the order of
InvokeObjectPostAlterHook() call and update of "attnotnull" field in
the catalog. Before the update was before the call of the hook [2].
And now "attnotnull" is updated after the call of
InvokeObjectPostAlterHook() [3].

This is a noticeable change to extensions which use
"object_access_hook" since they see a bit of stale information about
the attribute.

Is this an intended behavior or rather a bug?

1 - https://postgr.es/m/202408310358.sdhumtyuy2ht@alvherre.pgsql
2 - https://github.com/postgres/postgres/blob/REL_17_9/src/backend/commands/tablecmds.c#L7818
3 - https://github.com/postgres/postgres/blob/REL_18_1/src/backend/commands/tablecmds.c#L8043-L8047

-- 
Kind regards,
Artur

  2. Re: Order of InvokeObjectPostAlterHook within ATExecSetNotNull

    Álvaro Herrera <alvherre@kurilemu.de> — 2026-03-18T16:34:36Z

    On 2026-Mar-18, Artur Zakirov wrote:

> Hello hackers!
> 
> The commit 14e87ff added pg_constraint rows for not-null [1]. The
> consequence of the patch is that it changed the order of
> InvokeObjectPostAlterHook() call and update of "attnotnull" field in
> the catalog. Before the update was before the call of the hook [2].
> And now "attnotnull" is updated after the call of
> InvokeObjectPostAlterHook() [3].
> 
> This is a noticeable change to extensions which use
> "object_access_hook" since they see a bit of stale information about
> the attribute.
> 
> Is this an intended behavior or rather a bug?

This was not intentional.  Want to submit a patch?

-- 
Álvaro Herrera               48°01'N 7°57'E  —  https://www.EnterpriseDB.com/

  3. Re: Order of InvokeObjectPostAlterHook within ATExecSetNotNull

    Artur Zakirov <zaartur@gmail.com> — 2026-03-18T16:55:25Z

    On Wed, 18 Mar 2026 at 17:34, Álvaro Herrera <alvherre@kurilemu.de> wrote:
> >
> > Is this an intended behavior or rather a bug?
>
> This was not intentional.  Want to submit a patch?
>
> --
> Álvaro Herrera               48°01'N 7°57'E  —  https://www.EnterpriseDB.com/

Thank you for the quick response. Sure, I attached the patch for the
main branch and for REL_18_STABLE just in case, although the main
patch also should apply to REL_18_STABLE.
I think we might need to backpatch it to PG 18 too where the change
was introduced.

-- 
Kind regards,
Artur

  4. Re: Order of InvokeObjectPostAlterHook within ATExecSetNotNull

    Álvaro Herrera <alvherre@kurilemu.de> — 2026-03-20T13:45:29Z

    On 2026-Mar-18, Artur Zakirov wrote:

> Thank you for the quick response. Sure, I attached the patch for the
> main branch and for REL_18_STABLE just in case, although the main
> patch also should apply to REL_18_STABLE.
> I think we might need to backpatch it to PG 18 too where the change
> was introduced.

Thanks, this is clearly my mistake -- pushed to both branches.

(I didn't use the patch you sent for 18, just the one for master, which
I then cherry-picked into 18.  I would say that it's not necessary to
submit a version specific to a branch when the one for master
cherry-picks cleanly.)

-- 
Álvaro Herrera         PostgreSQL Developer  —  https://www.EnterpriseDB.com/