Jens Wilke <jens.wilke@affinitas.de> writes:
> Thanks Tom, yes, the index is named
> Indexes:
> "concurrently" btree (ulq_guid)
> In the 8.4 cluster and 9.0.4's pg_dumpall dumps it as
> CREATE INDEX concurrently ON foo USING btree (ulq_guid);
> That's it.
Oh, fun. We knew that not reserving that keyword was going to cause
some problems.
> But shouldn't pg_upgrade be able to handle this?
It's not pg_upgrade's fault; it's pg_dump that's failing to reproduce
the state of the source database.
I'm inclined to think that maybe we should hack pg_dump to forcibly
quote "concurrently" in this context, even though it doesn't do so
anywhere else since the word isn't reserved.
regards, tom lane