Re: contrib/pg_visibility fails regression under CLOBBER_CACHE_ALWAYS

On Mon, Jun 7, 2021 at 11:15 AM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>
> husky just reported $SUBJECT:
>
> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=husky&dt=2021-06-05%2013%3A42%3A17
>
> and I find I can reproduce that locally:
>
> diff -U3 /home/postgres/pgsql/contrib/pg_visibility/expected/pg_visibility.out /home/postgres/pgsql/contrib/pg_visibility/results/pg_visibility.out
> --- /home/postgres/pgsql/contrib/pg_visibility/expected/pg_visibility.out 2021-01-20 11:12:24.854346717 -0500
> +++ /home/postgres/pgsql/contrib/pg_visibility/results/pg_visibility.out 2021-06-06 22:12:07.948890104 -0400
> @@ -215,7 +215,8 @@
> 0 | f | f
> 1 | f | f
> 2 | t | t
> -(3 rows)
> + 3 | t | t
> +(4 rows)
>
> select * from pg_check_frozen('copyfreeze');
> t_ctid
> @@ -235,7 +236,8 @@
> 0 | t | t
> 1 | f | f
> 2 | t | t
> -(3 rows)
> + 3 | t | t
> +(4 rows)
>
> select * from pg_check_frozen('copyfreeze');
> t_ctid
>
>
> The test cases that are failing date back to January (7db0cd2145f),
> so I think this is some side-effect of a recent commit, but I have
> no idea which one.

It seems like the recent revert (8e03eb92e9a) is relevant.

After committing 7db0cd2145f we had the same regression test failure
in January[1]. Then we fixed that issue by 39b66a91b. But since we
recently reverted most of 39b66a91b, the same issue happened again.

Regards,

[1] https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=hyrax&dt=2021-01-19+20%3A27%3A46

--
Masahiko Sawada
EDB: https://www.enterprisedb.com/