On Wed, 2003-01-22 at 20:37, Tom Lane wrote:
> Rod Taylor <rbt@rbt.ca> writes:
> > Adds in NO MAXVALUE and NO MINVALUE options for create sequence per 200X
> > spec, which will also make alter sequence a touch easier.
>
> What are these supposed to do exactly? There is no such animal as a
> sequence that hasn't got a maxvalue, nor one without a minvalue.
> "Large" does not mean "unlimited".
6) Case:
a) If <sequence generator maxvalue option> is specified, then Case:
i) If NO MAXVALUE is specified, then let SMAX be an
implementation-defined <signed numeric literal> of declared type DT.
ii) Otherwise, let SMAX be <sequence generator max value>.
b) Otherwise, let SMAX be an implementation-defined <signed numeric
literal> of declared type DT.
7) Case:
a) If <sequence generator minvalue option> is specified, then Case:
i) If NO MINVALUE is specified, then let SMIN be an
implementation-defined <signed numeric literal> of declared type DT.
ii) Otherwise, let SMIN be <sequence generator min value>.
b) Otherwise, let SMIN be an implementation-defined <signed numeric
literal> of declared type DT.
> I'm not eager to pick up evidently-poorly-thought-out features from a
> draft spec that is nowhere near being frozen.
These 2 particular items are to spec, with the exception that we don't
stop at the limit of the datatype, int4's have int8 limits.
I think we need an ALTER SEQUENCE command, especially once dependencies
are fully tracked on them. This will include some method of returning
to the defaults.
> Anyway, didn't we decide that converging syntax with 200X was a
> bad idea if we weren't going to converge semantics?
Too late for that. The rest of the create sequence syntax is in-line
with the new spec, with the exception of the above 2 items.
--
Rod Taylor <rbt@rbt.ca>
PGP Key: http://www.rbt.ca/rbtpub.asc
