numpy & non-Apple gcc?

Adam Mercer ram at macports.org
Fri Sep 17 12:35:45 PDT 2010


On Fri, Sep 17, 2010 at 12:14, Michael Dickens <michaelld at macports.org> wrote:
> On Sep 17, 2010, at 12:48 PM, Adam Mercer wrote:
>> Looking at the history, the use of macports gcc was added when atlas
>> was added as a dependency [1]. I believe this was due to compiler ABI
>> differences.
>
> Thanks Adam; seems like maybe MacPorts/GCC/whatever don't have those differences any longer?

As Ryan states I imagine there will always be differences, wether or
not they'll effect us I don't know.

> I see that all of the fortran files are used for testing or in documentation.  Hence, from multiple perspectives I don't see the need for a fortran compiler, or for using gcc43/44/45 whatever.  Apple's GCC seems to work just fine, and it allows +universal to work pretty much out of the release code.

AFAICT the fortran is needed for ATLAS support, I asked upstream about
this a while ago and they told me that we generally didn't need to
specify one, but may need to if we linked against LAPACK or BLAS:

<http://mail.scipy.org/pipermail/numpy-discussion/2008-June/035046.html>

So, if I understand this correctly (which is by no means certain), as
we are linking against atlas we need to specify a fortran compiler?

> Admittedly, I'm a single person running on 10.6.4 x86_64.  Maybe this change won't work on 10.5 or for someone else?  I hope others can test it. - MLD

The addition of atlas as a dependency has caused a lot of headaches.

Cheers

Adam


More information about the macports-users mailing list