Re: pg_dump versus SERIAL, round N

Andreas Pflug <pgadmin(at)pse-consulting(dot)de> writes:
> I basically doubt the concept of a single owner. I'd expect a sequence
> to be dropped from cascaded table dropping, if that was the last usage
> and dependencies existed. This would probably mean "multiple owners".

That's not going to happen without extensive revisions to our dependency
mechanisms, which I am not about to undertake now. And I don't see the
point anyway. If you did have a sequence being used to feed multiple
tables, why would you want it to go away if the number of tables dropped
transiently to zero? If you then want to add back another table being
fed by that sequence, you've lost the state of the sequence. That's the
same sort of corner case that prompted us to allow zero-column tables,
ie, the table can continue to exist even if it momentarily has no
columns.

I see SERIAL as a simple shorthand for a common case, not some
hydra-headed beast ...

regards, tom lane