Tom Lane wrote:
>Given the thought that we need both
>transactional and nontransactional state for a sequence, I'm kind of
>inclined to leave the transactional data in pg_class. We could still
>imagine putting the nontransactional state into a new pg_sequence
>catalog indexed by, say, the pg_class OID of the sequences. OTOH I'm
>not sure how much that buys for Peter's problem --- it might be better
>for him just to invent some functions that can grab the required data
>given the sequence OID.
>
>
>
Yes, this seems a lot of lifting for a fairly small need. If there
aren't other advantages, a simple function or two seems a better way to
go, and then there are no legacy problems.
cheers
andrew
