On Mon, Jul 25, 2022 at 11:49:50AM +0200, Alvaro Herrera wrote:
> On 2022-Jul-23, Michael Paquier wrote:
>> As the problem comes down to the fact that INDEX/TABLE, SCHEMA and
>> DATABASE/SYSTEM need to handle names for different object types each,
>> I think that we could do something like the attached, removing one
>> block on the way at the cost of an extra parser node.
>
> Yeah, looks good. I propose to also test the error for reindexing a
> different database, which is currently uncovered, as attached.
Good idea.
>> By the way, it seems that 83011ce also broke the case of "REINDEX
>> DATABASE CONCURRENTLY", where the parser missed the addition of a
>> DefElem for "concurrently" in this case.
>
> Wow.
For this one, we have a gap in the test, actually. It seems to me
that we'd better make sure that the OID of the indexes rebuilt
concurrently is changed. There is a REINDEX DATABASE CONCURRENTLY
already in the TAP tests, and the only thing that would be needed for
the job is an extra query that compares the OID saved before the
reindex with the one in the catalogs after the fact..
--
Michael