Re: pg_dump: use ALTER TABLE for PKs

Philip Warner <pjw@rhyme.com.au> writes:
> I can't really think of a case where the index OID would be less than the
> table OID, but it might be good to put the table OID in the deps list; this
> will force the PK to go after the table in all cases. Did I mention I'm
> paranoid?

After OID-counter wraparound it'd be quite easy to have such a case.

To the extent that pg_dump is depending on OID comparison to tell it
something about creation order, it's really broken and should be fixed.
The dependencies mechanism should eventually be extended to cover all
these relationships.
        regards, tom lane