> Vadim, Philip changed that part of pg_dump on my advice. The idea was
> to try to do the right thing for sequences when loading schema only or
> data only. Analogously to loading data into a pre-existing table, we
> felt that a data dump ought to be able to restore the current state of
> an already-existing sequence object. Hence it should use setval().
Tables have many records but sequences single one.
So, I don't see what would be wrong if we would drop/recreate sequences
in data-only mode - result would be the same as with setval: required
state of sequence. Ok, ok - sequence' OID would be different.
...
> My inclination is to leave pg_dump as it stands, and change
> do_setval's error check. We could rip out the check entirely, or we
> could modify the code so that a setval() is allowed for a sequence
> with cache > 1 only if it's the new three-parameter form of setval().
> That would allow pg_dump to do its thing without changing the behavior
> for existing applications. Also, we can certainly make setval() flush
> any cached nextval assignments that the current backend is holding, even
> though we have no easy way to clean out cached values in other backends.
>
> Comments?
I don't object any approach.
Vadim
