[144621] trunk/dports/python/py-numpy/Portfile

[Sean Farley]

Joshua Root <jmr at macports.org> writes:

> On 2016-1-15 05:51 , Sean Farley wrote:
>> 
>> Joshua Root <jmr at macports.org> writes:
>> 
>>>> Revision: 144621
>>>>           https://trac.macports.org/changeset/144621
>>>> Author:   sean at macports.org
>>>> Date:     2016-01-13 23:29:24 -0800 (Wed, 13 Jan 2016)
>>>> Log Message:
>>>> -----------
>>>> py-numpy: numpy needs fortran, so require it
>>>
>>> We've been through this before; numpy does not need fortran, its fortran
>>> support is optional.
>> 
>> No, it is not optional. Trying running 'numpy.test()'. Or try compiling
>> with 'atlas +nofortran'. Not having fortran will generate a broken numpy
>> library.
>> 
>> Side note: We should consider getting rid of 'atlas +nofortran'. I
>> haven't found a port that depends on atlas but works without fortran.
>
> What you're saying is that to use fortran with numpy you need to enable
> its fortran support. Well, yes.
>
> If you're just using, say, pyopengl, then no, you don't need fortran
> support.

While a project *might* not need numpy's fortran, numpy expects a
fortran compiler:

"you’ll also need a FORTRAN 77 compiler installed." [1]

This means any dependent of numpy is correct in assuming that the python
compiler wrappers that numpy provide will have fortran. The amount of
headache this solves for us, I believe, far outweighs not having
fortran.

Until we can reliably depend on variants, I don't want a broken numpy
library installed (same with my feelings of removing atlas +nofortran).
It just causes too much headache.

[1] http://docs.scipy.org/doc/numpy-1.10.1/user/install.html