Horton

Re: longfin and tamandua aren't too happy but I'm not sure why

Peter Geoghegan <pg@bowt.ie>

View thread View on postgr.es
From: Peter Geoghegan <pg@bowt.ie>
To: Thomas Munro <thomas.munro@gmail.com>
Cc: Robert Haas <robertmhaas@gmail.com>, Tom Lane <tgl@sss.pgh.pa.us>, Dilip Kumar <dilipbalaut@gmail.com>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>, Justin Pryzby <pryzby@telsasoft.com>
Date: 2022-09-28T19:49:36Z
Lists: pgsql-hackers 
On Wed, Sep 28, 2022 at 12:32 PM Thomas Munro <thomas.munro@gmail.com> wrote:
> I don't agree with this.  The build farm clearly has more ways to
> break than CI, because it has more CPUs, compilers, operating systems,
> combinations of configure options and rolls of the timing dice, but CI
> now catches a lot and, importantly, *before* it reaches the 'farm and
> everyone starts shouting a lot of stuff at you that you already knew,
> because it's impacting their work.

Right. I really don't can't imagine how CI could be seen as anything
less than a very significant improvement. It wasn't that long ago that
commits that certain kinds of work that used OS facilities would
routinely break Windows in some completely predictable way. Just
breaking every single Windows buildfarm animal was almost a routine
occurrence. It was normal. Remember that?

Of course it is also true that anything that breaks the buildfarm
today will be disproportionately difficult and subtle. You really do
have 2 buildfarms to break -- it's just that one of those buildfarms
can be broken and fixed without it bothering anybody else, which is
typically enough to prevent breaking the real buildfarm. But only if
you actually check both!

-- 
Peter Geoghegan

Commits

  1. meson: Split 'main' suite into 'regress' and 'isolation'

  2. ci: Add 32bit build and test

  3. Fix InitializeRelfilenumberMap for 05d4cbf9b6ba708858984b01ca0fc56d59d4ec7c

  4. Fix alignment problems with SharedInvalSmgrMsg.

  5. In BufTagGetForkNum, cast to the correct type.

  6. Increase width of RelFileNumbers from 32 bits to 56 bits.

  7. Harden pg_filenode_relation test against concurrent DROP TABLE.