On Tue, Sep 27, 2022 at 5:29 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> ... also, lapwing's not too happy [1]. The alter_table test
> expects this to yield zero rows, but it doesn't:
>
> SELECT m.* FROM filenode_mapping m LEFT JOIN pg_class c ON c.oid = m.oid
> WHERE c.oid IS NOT NULL OR m.mapped_oid IS NOT NULL;
>
> I've reproduced that symptom in a 32-bit FreeBSD VM building with clang,
> so I suspect that it'll occur on any 32-bit build. mamba is a couple
> hours away from offering a confirmatory data point, though.
>
> (BTW, is that test case sane at all? I'm bemused by the symmetrical
> NOT NULL tests on a fundamentally not-symmetrical left join; what
> are those supposed to accomplish? Also, the fact that it doesn't
> deign to show any fields from "c" is sure making it hard to tell
> what's wrong.)
This was added by:
commit f3fdd257a430ff581090740570af9f266bb893e3
Author: Noah Misch <noah@leadboat.com>
Date: Fri Jun 13 19:57:59 2014 -0400
Harden pg_filenode_relation test against concurrent DROP TABLE.
Per buildfarm member prairiedog. Back-patch to 9.4, where the test was
introduced.
Reviewed by Tom Lane.
There seems to be a comment in that commit which explains the intent
of those funny-looking NULL tests.
--
Robert Haas
EDB: http://www.enterprisedb.com