Re: [XEN PATCH 1/4] automation: Remove clang-8 from Debian unstable container

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Wed, 15 Feb 2023, Andrew Cooper wrote:
> On 15/02/2023 4:31 pm, Andrew Cooper wrote:
> > On 15/02/2023 4:28 pm, Anthony PERARD wrote:
> >> On Wed, Feb 15, 2023 at 12:26:40PM +0000, Andrew Cooper wrote:
> >>> On 15/02/2023 12:02 pm, Anthony PERARD wrote:
> >>>> First, apt complain that it isn't the right way to add keys anymore,
> >>>> but hopefully that's just a warning.
> >>>>
> >>>> Second, we can't install clang-8:
> >>>> The following packages have unmet dependencies:
> >>>>  clang-8 : Depends: libstdc++-8-dev but it is not installable
> >>>>            Depends: libgcc-8-dev but it is not installable
> >>>>            Depends: libobjc-8-dev but it is not installable
> >>>>            Recommends: llvm-8-dev but it is not going to be installed
> >>>>            Recommends: libomp-8-dev but it is not going to be installed
> >>>>  libllvm8 : Depends: libffi7 (>= 3.3~20180313) but it is not installable
> >>>> E: Unable to correct problems, you have held broken packages.
> >>>>
> >>>> clang on Debian unstable is now version 14.0.6.
> >>>>
> >>>> Signed-off-by: Anthony PERARD <anthony.perard@xxxxxxxxxx>
> >>> Yeah that's because clang 8 is way obsolete, and http://apt.llvm.org
> >>> only supports Clang 15 and later now.
> >>>
> >>> In hindsight, it was a mistake to take unstable-llvm-8.list in the first
> >>> place.
> >>>
> >>> Acked-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
> >> I just realize that this change is going to break gitlab testing in the
> >> staging-4* branches. So I guess this mean that we will have to backport
> >> this patch, or at least the gitlab-ci part.
> > I don't think it will.  The currently cached container is still "good"
> > for the older branches to use, but we ought to backport the change
> > nevertheless.
> >
> > Given that the container is un(re)buildable anyway, we're not losing
> > anything, I don't think.
> 
> Ah, so.  Applying just this patch won't break older trees.
> 
> But applying this patch, and patch 3, then rebuilding the the debian
> unstable container will break older branches, because there will no
> longer be anything called clang-8 in the container.
> 
> This demonstrates even more that referencing the compiler by exact
> version is a mistake.  As is using names like unstable.
> 
> As I said up front when we added CI of this form, we were always going
> to retrofit some changes to all branches (including the security-only
> branches) to keep this working.  These chicken are coming back to roost.

Yes. Like you said, I think we should backport this patch (or just the
part of it affecting .gitlab*) to the stable trees.