Updated archive rebuild results

[Matthias Klose]

On 04.12.2010 20:53, Scott Howard wrote:
> On Sat, Dec 4, 2010 at 1:41 PM, Evan Broder<evan at ebroder.net>  wrote:
>> On Sat, Dec 4, 2010 at 8:17 AM, Scott Howard<showard314 at gmail.com>  wrote:
>>> Those two data points, combined, leads me to believe that there are
>>> 400+ packages that are failing to build in Natty. Because of the
>>> Debian freeze, we should not expect fixes to end up in unstable and
>>> thus propagate to Ubuntu. Ubuntu would have to create diffs to fix
>>> those FTBFS.
>>
>> I think your assumptions are a little off. Plenty of DDs are more than
>> happy to upload fixes for binutils-gold errors into unstable. I've
>> fixed at least one FTBFS where I knew the Debian maintainer was
>> responsive by sending the patch off, waiting for it to be uploaded to
>> unstable, and then syncing it back into Debian. No Ubuntu diff needed.
>
> My problem isn't with the diff as much as there is work that needs to
> be done, and it needs to be initiated from within Ubuntu since these
> are not RC bugs to Debian until after squeeze is released. I'm not
> saying there isn't coordination, just that if Ubuntu waits for Debian,
> it won't get done until after Squeeze is released because DD/DMs
> shouldn't upload to unstable unless there is an RC bug fix during
> freeze. These FTBFS are not RC bugs to Debian Squeeze (but are for
> Debian Weezy). Exactly as you said, you had to do the work to fix the
> FTBFS - which is great, we need more of that. Diff or no-diff, we need
> to do that 400 more times before natty.

So how long do you wait?  Until after these issues are all fixed in Debian? 
Surely that is the least amount of work, but do we want to wait that long?  If 
changes are made in Debian and Ubuntu starts to merge these changes, then Ubuntu 
still has to go ahead for all these reports for the next Ubuntu release (if this 
release is before the next Debian release).

Looking at our current toolchain "diff":

  - GCC-4.5: Natty started to use GCC-4.5.  Issues are reported in Debian
    as [1], these are reproducible by installing gcc/g++ and gcc-4.5/g++-4.5
    from experimental).

This was planned at the maverick UDS. Better code generation in 4.5 than in 4.4.

  - Build failures with the Gold linker, or with ld --no-add-needed.
    Reported as , reproducible by installing gcc/g++ and gcc-4.5/g++-4.5
    from experimental or by installing binutils-gold from experimental.

Planned and decided at the natty UDS. It's a correctness issue, and it will 
enable us to further experiment/test with gold.

  - Build failures with ld --as-needed. Issues found at [3].

Discussed at the natty UDS.  Primary reason is to get dependencies down a bit 
and ease library transitions (Ubuntu doesn't have binNMU's).  Before release 
there's always a lot of cruft to clean up in NBS; these changes should reduce 
the amount of clean-up work.

  - Build failures with glibc-2.12, currently not tracked with user tags,
    although I cannot remember more than a few issues.

Not (yet) changed in natty.

  - Build failures with more recent kernel headers, currently not tracked
    with user tags.

Changes not tested, fallout will be seen in packages using kernel headers. I 
don't track these explicitly.

Large-scale changes always make more work than a simple library transition, but 
there are reasons to make these changes because of the benefits.  You cannot 
completely avoid this kind of work, even if Debian starts but Ubuntu needs to 
finish earlier, so it's more predicatable to start these large-scale changes at 
the beginning of a release cycle.

There will be one more large-scale change for natty, making python2.7 the 
default.  Delaying that for 11.10 would make this version available more than 
twelve months after the 2.7 release, which is a bit late. At least Ubuntu uses 
Python for many distro-specific developments.  So we'll make the 2.7 switch 
still in this year.

> I'm a DM that received a bug report from ubuntu with a patch, which I
> immediately uploaded to experimental and requested sync to ubuntu.
> Even if every maintainer does that, there are still 400+ packages that
> need to be reviewed, patched, uploaded to debian (most likely
> experimental), and manually synced to Ubuntu (since it is in
> experimental). It is doable, we just need to be aware of it and get
> started. My point is that if we don't put an organized effort into it,
> a large number of packages won't be fixed in time for natty.

About 100 packages are now fixed in natty, even with the holidays coming, these 
looks promising to get these fixed until the feature freeze.

   Matthias