David Prévot <taf...@debian.org> writes:

> Le 29/08/2012 08:27, Julien Cristau a écrit :
>>> Le 27/08/2012 19:03, David Prévot a écrit :
>>>
>>>> Could you please trigger a rebuild, to make the package work again
>>>> (fixing an RC-bug)? Thanks in advance.
>>>>
>>>> nmu python-scientific_2.8-3 . ALL . -m "Rebuild with current stuff 
>>>> (Closes: #685812)"
>
>> - the changelog entry is not understandable
>> - you can't close bugs in binNMUs
>> - I won't schedule a binNMU without knowing why it's needed.  Neither
>>   your request nor the referenced bug help with that.
>
> Thanks Julien for your reply. Documenting on the original bug report
> that the “easy way” is not enough.

The bug in python-netcdf comes from a change in python-numpy. The
last binNMU of python-netcdf having occurred on 2012-03-08, it was built
against python-numpy 1:1.5.1-4, whereas sid/wheezy currently have 1:1.6.2-1.

This is confirmed by the following tests:

- install python-netcdf and python-numpy as currently in sid => the
  testcase provided by Julian Taylor in #685812 fails (as already noted
  by Julian)

- downgrade to python-numpy 1:1.5.1-4 without recompiling python-netcdf
  => the testcase succeeds

- upgrade back to python-numpy 1:1.6.2-1 then recompile python-netcdf =>
  the testcase succeeds (as already noted by Julian)

So my understanding is that the problem comes from some ABI breakage in
python-numpy, which was not followed by a recompilation of python-netcdf.

What I don't understand however is that the python-numpy
maintainers/upstream claim that there was no ABI breakage between
1:1.5.1-4 and 1:1.6.2-1 (see the Provides field of python-numpy). Either
I am missing something or that claim is wrong.

As a side note, a call to dh_numpy should be added to debian/rules in
python-scientific (as noted by lintian); in the present case it would
however not have helped, as evidenced by the above paragraph.

-- 
 .''`.    Sébastien Villemot
: :' :    Debian Maintainer
`. `'     http://www.dynare.org/sebastien
  `-      GPG Key: 4096R/381A7594

Attachment: pgpxtedOz3ung.pgp
Description: PGP signature

Reply via email to