While poking at Michel Dorochevsky's issue, I noticed that if a DELETE
is done in a table that is referenced by many different foreign keys,
we repeat the ri_Check_Pk_Match test over again for each FK. Seems like
it would be nice to avoid this duplicated work. But right now I don't
see any easy way to refactor the RI triggers for it --- the triggers are
independent. Something to think about in future, maybe.
regards, tom lane