Re: Current and upcoming toolchain changes for jessie
On Wed, May 08, 2013 at 01:39:25AM +0200, Matthias Klose wrote:
> Am 07.05.2013 17:48, schrieb Aurelien Jarno:
> > On Tue, May 07, 2013 at 03:25:29PM +0200, Matthias Klose wrote:
> >> Up to today jessie did see updates for the kernel headers, eglibc, and
> >> GCC.
> > What a wonderful coordination with the release team. Quoting the last
> > mail from them on the mailing list:
> > | As for Squeeze, we'd ask that you co-ordinate particularly large
> > | transitions or changes; if your plans involve major toolchain changes or
> > | otherwise have the potential to cause problems in unstable for a long
> > | time (e.g. due to FTBFS issues), please talk to us.
> I think you are over-reacting a bit here. My impression was that the eglibc
> changes were coordinated within the team. So what about the other changes?
They weren't coordinated within the team. Furthermore I don't consider
that eglibc was ready to go to unstable, as it was known that two
architectures were going to FTBFS, without a real try to get that fixed
(for example by contacting the porters).
> - a new version of GCC was uploaded, not as the default, but as a
> separate version. I didn't want to split out the x32 multilibs
> again. I didn't do a test rebuild of the archive, but this upload
> shouldn't cause any additional build failures.
I still don't fully see the point of having this x32 multilibs, given
(a) It goes on the opposite direction of multiarch
(b) It's completely useless on a Debian system, as the kernel doesn't
support x32 binaries.
> There is a mini transition involving ppl and cloog-ppl, however the
> GCC packages are the only reverse dependencies, so this looks fine.
> - binutils was not uploaded, just the intent for the changes
> - I didn't upload linux-libc-dev myself, but until today I didn't
> see any announcement or a test rebuild done by the Debian kernel
> maintainers, so I did add the note about what I did see in multiple
> packages when doing a test rebuild in Ubuntu.
> And I think it's a normal change, and build failures can be reported
> after a test rebuild. You may want to contact the kernel team, if
> you think otherwise.
> As you see in the release announcement, the Debian release team was aware about
> the GCC 4.8 upload. Further I did contact the release team (Adam Barrat) weeks
As far as I know the release team was not aware of the broken eglibc
upload before it happened. They seemed to not really like the fact that
almost every package built against the new libc6 is now getting a
dependency on libc6 >= 2.14, hence blocking all other transitions.
> before the release how to better coordinate the opening of jessie. I assume that
> just didn't happen because people were busy with other things. Yes, seeing
What about contacting the team instead of a random person of the team?
Otherwise you coordinate with one person of the team, not the team.
> people uploading random new library versions which did break at least the ppl
> build did make me a bit nervous, so I wanted to get the required changes in
> place for the upcoming transitions.
> There were rumours that the release would be delayed until May 15, so I tried to
> coordinate toolchain changes to be ready at this time. So, the rumours did
> prove wrong, and Adam Conrad did upload eglibc and gcc packages today.
What's the point of rushing so much these uploads, even if they are not
> >> For the Debian glibc maintainers, Adam Conrad
> > Now I understand why I was not in the loop, I am not member of the team
> > anymore. Glad to learn that in such an email.
> It would help to keep down the sarcasm. My intent was to describe that as an
> coordinated upload of eglibc and GCC. You do have commit rights for some GCC
> ports as well, so you can surely contact me if a port is broken or unusable. But
> according to test results, things look not worse for 4.8 than for 4.7.
It was clearly a coordinated upload, but between Adam Conrad and you,
not between the glibc maintainers and the gcc maintainers.
Aurelien Jarno GPG: 1024D/F1BCDB73