>> As long as a sequence can point to more than
>> one table/column combination.
Doesn't seem like a problem to me --- as far as I understood Ryan,
the new table he's proposing would only contain entries for sequences
created to implement SERIAL keywords. For those, I think there should
indeed be a 1-1 mapping between parent table (+column) and resulting
sequence.
But yeah, don't break usage of ordinary standalone sequences ;-).
Another thing to think about is what it's going to take to dump and
reload this structure in pg_dump. We need to be able to reconstitute
the system tables' contents and the current value of the SERIAL sequence
after a reload.
regards, tom lane