Re: alter_table regression test problem

On 2013-11-07 00:30:55 +0100, Andres Freund wrote:
> On 2013-11-07 00:17:34 +0100, Andres Freund wrote:
> > On 2013-11-06 17:00:40 -0300, Alvaro Herrera wrote:
> > > Kevin Grittner wrote:
> > >
> > > > That makes for a pretty simple test for git bisect, even if
> > > > everything including initdb is painfully slow with
> > > > CLOBBER_CACHE_ALWAYS.
> > >
> > > Most likely culprit is f01d1ae3a104019d6d68aeff85c4816a275130b3
> >
> > Well, that test tests the functionality added in that commit, so sure,
> > it can't be before that. What confuses me is that relfilenodemap has
> > survived quite some CLOBBER_CACHE_ALWAYS runs in the buildfarm since:
> > http://buildfarm.postgresql.org/cgi-bin/show_history.pl?nm=jaguarundi&br=HEAD
> >
> > Did you compile with CLOBBER_CACHE_ALWAYS or CLOBBER_CACHE_RECURSIVELY?
>
> Either way, the code is completely and utterly broken in the face of
> cache invalidations that are received when it does its internal
> heap_open(RelationRelationId). I can't have been thinking straight when
> I wrote the invalidation logic.

Ok, I've attached a fix for this, which unfortunately is not all that
small.
Could either of you commit it?

Greetings,

Andres Freund

--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services