Re: [PATCH] ci: bump debian-11 job to debian-12

[Patrick Steinhardt <ps@pks.im>]

On Wed, Oct 07, 2026 at 01:44:09PM -0700, Junio C Hamano wrote:
> Jeff King <peff@peff.net> writes:
> 
> > BTW, I noticed that the linux32 build is using ubuntu 20.04, which has
> > been out of LTS for a year. But bumping isn't really an option; they
> > dropped i386 platform support, and so has Debian.
> >
> > I'm mostly inclined to leave it unless/until it starts creating
> > headaches. To some degree, if we cannot even find an image to test
> > again, it might not be an important enough platform to care about. But I
> > can also imagine there is a long tail of oddball 32-bit platforms that
> > Git does run on (like small ARM chips), and it's nice to at least have
> > some coverage. Possibly there's an ARM image we could use (looks like
> > armhf?).
> >
> > We also seem to use 20.04 for linux-TEST-vars. On the surface there's no
> > reason it couldn't be using ubuntu-latest, though I think this may be
> > one of those cases where it's doing double duty as "test exotic configs"
> > and "test on an older platform". But might be worth bumping to the
> > oldest in-scope LTS.
> >
> > All out of scope for this patch, and mostly I'm inclined to ignore it
> > for now until we hit problems (and then decide if it's worth
> > accommodating or if old systems are too old).
> 
> I am getting annoyed enough to see that the lack of 20.04 is finally
> giving failures more often than it used to.  And am planning to
> suggest to:
> 
>  * drop linux32 job
> 
>  * update linux-TEST-vars to run with ubuntu:rolling like everybody
>    else with the default version of gcc
> 
> I personally do not see much value in the test-vars job in that
> enabling all exotic configs all at once would not match use patterns
> of any real world users, which may likely to enable only some but
> not all of them, and for that reason am also tempted to propose to
> just remove it at the same time as we remove linux32 job.

I think it'd be great to continue exercising i386, and Debian still has
supported images for that.

Ialso think that we should keep the TEST-vars job. It has catched
regressions several times for me in the past. Sure, the combination of
flags in typically not exercised. But I think that by itself is not a
good enough argument to drop this entirely.

But, oh well, you probably just want to snipe somebody into doing this.
So fine, I'll bite and will send patches later today.

Patrick