On Mon, 14 May 2001, Tom Lane wrote:
> Philip Warner <pjw@rhyme.com.au> writes:
> > It is worth considering skipping the entire 'copy to children' approach?
> > Something like:
> > pg_constraints(constraint_id, constraint_name, constraint_details....)
> > pg_relation_constraints(rel_id, constraint_id)
> > Then, when we drop constraint 'FRED', the relevant rows of these tables are
> > deleted. There is only ever one copy of the constraint definition.
>
> This would work if we abandon the idea that a table cannot have
> multiple constraints of the same name (which seems like an unnecessary
> restriction to me anyway).
I'm not sure it would. You could have two constraint_ids with the same
name still as long as there's no constraint on constraint_name, both would
presumably be deleted on a drop. Since rel_id is only part of
pg_relation_constraints you wouldn't want the constraint_name to be forced
unique (barring the spec definition) anyway, so there'd be nothing to
prevent you from naming all your constraints FRED, just you'd have a
better way to refer to a particular constraint than its name internally.
