> If you break out the following two patch items:
>
> > > - psql displays foreign keys (\d output)
> > > - Foreign key triggers are autonamed based on the constraint
name
>
> I'm sure that part of the patch will get committed (so long as it's
good),
> as we'd agreed already to come up with such a patch (except I never
got
Yes, isconstraint triggers are ignored. Triggers were easy to toss
the names around as they inherit the name of the constraint. The
constraint was autonamed.
> > > - pg_dump uses ALTER TABLE / ADD FOREIGN KEY
>
> The item above is trouble because it makes adding foreign keys from
dumps
> very slow on large tables. The advantage of the CREATE CONSRAINT
TRIGGER
> approach is that it doesn't actually _check_ the constraint.
>
> My earlier suggestion was to create a 'SET CONSTRAINTS UNCHECKED;'
sort of
> transaction-only function that would make ADD FOREIGN KEY _not_
check
> constraints. I can't remember what the repsonse to that was, but we
need
> something...
No choice. Using CREATE TRIGGER drops all dependency information with
a pgdump / restore -- not to mention the actual foreign key. If you
want to wrap it with the above suggestion feel free. I've got a few
other things to clean up before I could go into that.