Re: pg_sequence catalog

Andres Freund <andres(at)anarazel(dot)de> writes:
> On September 5, 2016 7:26:42 AM PDT, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> The main problem I can see with this is that serial columns will have
>> default expressions that are written out as
>> "nextval('foo_f1_seq'::regclass)". I do not think we can afford to
>> break dumps containing that, but I'm not sure how to get the regclass
>> cast replaced with a regsequence cast.

> Why not just continue having a pgclass entry, but no relfilenode?

Yeah, maybe. I was hoping to dispense with the pg_attribute rows, but
maybe that's not enough overhead to worry about.

In this viewpoint, we'd keep the sequence-specific data in a pg_sequence
catalog. pg_sequence rows would be extensions of the associated pg_class
rows in much the same way that pg_index rows extend the pg_class entries
for indexes. We should supply a view pg_sequences that performs the
implied join, and encourage users to select from that rather than directly
from pg_sequence (compare pg_indexes view).

regards, tom lane