> > (1) the "CREATE SEQUENCE foo TABLESPACE disk2" syntax does not seem
> > to be implemented.
>
> This is intentional. Sequences are not large enough to need to be
> pushed around among multiple tablespaces. Also, if we did allow
> sequences to be associated with tablespaces, we'd be precluding other
> implementation changes that are on the wish-list (such as storing all
> sequences in a single system table, instead of needing a separate disk
> file for each one).
That is a point.
As for the semantics, sequences have a tablespace anyway, which is the
default tablespace of the schema as it seems, and it appears in pg_class,
so it is already implemented somewhere, no doubt about that.
As for the syntax, you could decide to ignore the tablespace part of the
syntax if such evolution would require it, maybe with some warning for the
user that part of its query is no more up to date...
It would no more a big deal than dropping "LOCATION" from "CREATE
DATABASE", which is not a upward compatible change and was performed
remorselessly anyway.
> The original patch actually had support for specifying a tablespace for
> a sequence. That was deliberately removed, and it's not going to go
> back in later.
Too bad for elegance and orthogonality.
Thanks for your clear answer anyway;-)
Have a nice day,
--
Fabien Coelho - coelho@cri.ensmp.fr