[SciPy-Dev] OSX build failing at gfortran link

[Ralf Gommers]

On Tue, Oct 12, 2010 at 10:58 PM, Ralf Gommers
<ralf.gommers at googlemail.com>wrote:

> Hi Matthew,
>
> On Tue, Oct 12, 2010 at 7:33 AM, Matthew Brett <matthew.brett at gmail.com>wrote:
>
>> Hi Ralf,
>>
>> >> Looking at numpy/distutils/fcompiler/gnu.py, I see that the function
>> >> ``_can_target`` cycles through the arch flags seeing whether they (on
>> >> their own) raise an error when compiling an empty .f file.    I guess
>> >> the problem here is that, although gfortran will compile a file with
>> >> '-arch x86_64' - it can't link a python extension with '-arch x86_64'.
>> >>  Would it make any sense to get the -arch flags somehow from the
>> >> default C linking flags?  Is that sensible?  Is it easy?
>> >
>> > I think it makes sense but I'm not sure it's easy. I'll try to figure
>> that
>> > out before 1.5.1rc1, if it's not easy then the patch from #1399 should
>> just
>> > be reverted. Mea culpa again for causing this problem.
>>
>> I had a go - always unpleasant with distutils.  I think this works:
>>
>> http://github.com/matthew-brett/numpy/compare/numpy:master...farchs-from-c
>>
>> At least the standard 'CXX=/usr/bin/g++-4.0 python setup.py install'
>> routine gave a working scipy passing all tests.
>>
>> Does it look OK to you?  If so I'll submit a pull request to numpy.
>>
>> Thanks for diving in! That works for me, and looks like the right thing to
> do. The methods you removed are clearly marked as private in the base class
> and not used anywhere else as far as I can tell, but it would be good to get
> a confirmation from someone familiar with the history / use cases of
> fcompiler that this is fine.
>

David W.F. tested on two non-standard Python installs, all good:
http://projects.scipy.org/numpy/ticket/1399#comment:8. So time to apply your
fix I think.

Ralf
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20101014/56387eb5/attachment.html>