Bernd Helmle <mailings@oopsware.de> writes:
> What i can try is to record the inheritance information only in case of
> attinhcount > 0. This would make maintenance of the pg_constraint records
> for NOT NULL columns a little complicater though. Another thing we should
> consider is that Peter's functional dependency patch was supposed to rely
> on this feature (1), once it gets done. Not sure this still holds true....
Oh, right, that's a killer argument. Finishing that patch still
requires that NOT NULL constraints have pg_constraint OIDs assigned,
which means they *have to* have pg_constraint rows to carry the OIDs.
So forget the whole thing; we'll just eat the space penalty.
regards, tom lane