Re: Trixie GCC backports
Hi Matthias,
Matthias Klose <doko@debian.org> hat am 29.09.2026 08:18 MESZ geschrieben:
> On 9/28/26 22:57, Micha Lenk wrote:
> > Am 15. September 2026 15:30:45 MESZ schrieb Emanuele Rocca <ema@debian.org>:
> >> doko and I thought it would be good for stable users to have access to
> >> the latest GCC builds from the various release branches, as they include
> >> a range of fixes and improvements. For example, Trixie has 14.2.0, while
> >> 14.4.0 is available and currently in Forky.
> >>
> >> I have just uploaded 14.4.0-2~bpo13+1, and I plan on working on gcc-12
> >> and gcc-13 in the coming days.
> >
> > To be honest, I don't feel comfortable accepting these uploads. Have you
> > thought through that backports might pick up versioned library dependencies
> > from these builds? If such a backport is installed, it might per versioned
> > dependency update e.g. libstdc++6 from trixie by the version provided in
> > trixie-backports. This will happen unconditionally, for all other packages
> > that depend on it (but happy with the version in trixie) too...
>
> what is the problem?
That I'd like to know too, and that's why I requested more information before
taking a decision.
> if you use a compiler from backports, then a
> potential package build lands in backports as well. Did you even look at
> the symbols files of the runtime libraries?
No, and I also don't know the policy that drives updates to the many GCC
version lines. This is where the expertise of the package maintainers by
far exceeds what the ones deciding whether or not to accept an upload
need to know for this decision.
Yet I checked the number of reverse dependencies. libstdc++6 is picked up
as dependency by 8000 binary packages in trixie. This adds an impact
factor we need to take into account.
Did you follow my example actually, where I tried to outline how a newer
libstdc++6 pulled in by *one* backport (granted, I don't know if this would
ever happen) could impact *a lot* of non-backport packages, simply because
we can only have one version of it installed at a time?
> Also please note that happens with the recent llvm backports as well,
> only that they are even available in the security pocket. So what was
> the argument for accepting these?
Quickly cross-checking, libllvm19 is only picked up as dependency by 57
binary packages -- two magnitudes below that. Should I have cross-checked
a different binary package? Also, LLVM seems to version their libraries
far more aggressively, which prevents such binary package collisions from
trixie vs. trixie-backports and the possible impact nicely.
> Please don't argue with feelings for such issues, but give solid
> technical reasons.
I certainly don't want to take this decision on a package with so many
reverse dependencies easy. Yet I have a feel, and of course I prefer to
decide based on facts, not the feel alone. Collecting such facts is the
whole point of my initial reply.
> I have been backporting GCC versions to Ubuntu LTS
> versions for the past 10 years, without having issues. If there are
> *technical* reasons, please be explicit.
I do not intend to question your expertise. I'd only appreciate if this
expertise could be shared for everyone's benefit so that we all can
follow along why accepting the gcc backports will be a good decision.
I have no problem to take your confidence as a signal to accept the
GCC backports. In the end I'm only operating the backports archive, and
I trust you'll support the uploads you bless in case they cause issues.
Best regards,
Micha
Reply to: