Re: NOT ENFORCED constraint feature

On Tue, Feb 11, 2025 at 12:18 AM Álvaro Herrera <alvherre(at)alvh(dot)no-ip(dot)org> wrote:
>
[...]
>
> > • v13-0001-Add-AlterConstraintStmt-struct-for-ALTER-.-CONST.patch
>
> I think the new node name is wrong, because it makes the code looks as if we support ALTER CONSTRAINT as a statement for this, which is false. (This is a repeat of ReplicaIdentityStmt, which I think is a mistake). I would suggest a name like ATAlterConstraint instead. Perhaps we can use that in the patch for ALTER CONSTRAINT ... SET [NO] INHERIT as well, instead of (as I suggested Suraj) creating a separate subcommand number.

I have renamed AlterConstraintStmt to ATAlterConstraint as per your
suggestion in the attached version. Apart from this, there are no
other changes, as the final behavior is still unclear based on the
discussions so far.

If we don’t want to keep a constraint marked as valid when it is not
enforced, we should revert to the v12 version behavior, where the
constraint is marked invalid when changed to NOT ENFORCED and remains
so even after being changed to ENFORCED, until explicitly validated
using the existing ALTER ... VALIDATE CONSTRAINT command.

However, if we want to provide an option to validate the constraint
immediately when changing it to ENFORCED, we could introduce support
for a syntax like:

ALTER TABLE ... ALTER CONSTRAINT ... [ NOT ] ENFORCED [ (WITH |
WITHOUT) VALIDATION ]

Regards,
Amul