Re: Using contrib modules in check (Re: pgsql: Fix BRIN to use SnapshotAny during summarization)

Christoph Berg <myon@debian.org> writes:
> Re: Tom Lane 2015-08-07 <928.1438900846@sss.pgh.pa.us>
>> I don't really think we need this isolation test at all, but if we do,
>> please fix it to not rely on any extensions.  Perhaps looking at
>> pg_relation_size or some such would do?  Or you could just issue
>> a query that should use the index, and see if it finds the rows it
>> ought to.

> this breaks the Debian package builds as well because we run
> check-world as a build step.
> Any chance for a fix/workaround so the nightly master/head builds will
> succeed again?

I was waiting for Alvaro to deal with this, but perhaps he's on summer
vacation or something.  I will remove the isolation test until he has
time to address it more fully.

However, we did learn something valuable from the fact that all the
-DCLOBBER_CACHE_ALWAYS critters failed on it: per my earlier message,
brin_page_items() is unsafe against a relcache flush on the index.
I'll put that on the 9.5 open items list.

(If I were tasked with fixing it, I'd be tempted to rewrite it to do
all the work in one call and return a tuplestore; the alternative
seems to be to try to keep the index open across multiple calls,
which would be a mess.)

            regards, tom lane