Re: Patch to add a primary key using an existing index

Gurjeet Singh <singh(dot)gurjeet(at)gmail(dot)com> writes:
> Attached is the updated patch with doc changes and test cases.

Applied with assorted corrections. Aside from the refactoring I wanted,
there were various oversights.

> I have consciously disallowed the ability to specify storage_parameters
> using the WITH clause, if somebody thinks it is wise to allow that and is
> needed, I can do that.

AFAICS, WITH would be supplied at the time of index creation; it's not
appropriate to include it here, any more than INDEX TABLESPACE.

A point that may or may not have gotten discussed back when is that it's
important that the result of this process be dumpable by pg_dump, ie
there not be any hidden discrepancies between the state after ADD
CONSTRAINT USING INDEX and the state you'd get from straight ADD
CONSTRAINT, because the latter is the syntax pg_dump is going to emit.
ADD CONSTRAINT can handle WITH and INDEX TABLESPACE, so carrying those
over from the original index specification is no problem, but
non-default index opclasses or sort ordering options would be a big
problem. That would in particular completely break pg_upgrade, because
the on-disk index wouldn't match the catalog entries created by running
pg_dump. I added some code to check and disallow non-default opclass
and options.

regards, tom lane