[Numpy-discussion] Compiling for free on Windows32

Pauli Virtanen pav at iki.fi
Wed Apr 15 15:54:30 EDT 2009


Wed, 15 Apr 2009 15:42:05 -0400, josef.pktd wrote:

> On Wed, Apr 15, 2009 at 3:17 PM, David Cournapeau <cournape at gmail.com>
> wrote:
>> On Thu, Apr 16, 2009 at 1:17 AM, Fadhley Salim
>> <Fadhley.Salim at uk.calyon.com> wrote:
>>> I've been asked to provide Numpy & Scipy as python egg files.
>>> Unfortunately Numpy and Scipy do not make official releases of their
>>> product in .egg form for a Win32 platform - that means if I want eggs
>>> then I have to compile them myself.
>>
>> Yes.
>>
>>
>>> At the moment my employer provides Visual Studio.Net 2003, which will
>>> compile no version of Numpy later than 1.1.1 - every version released
>>> subsequently cannot be compiled with VS2003. I've tried.
>>
>> Yes, this is a visual studio bug.
>>
>>
>>> What I'd really like is some other compiler I can use, perhaps for
>>> free, but at a push as a free time-limited trial. I would use that to
>>> compile the eggs. Is anybody aware of another compiler that I can
>>> download and use without paying anything and will definitely compile
>>> Numpy on Windows?
>>
>> The compilers that are regularly used to compile python on windows are
>> the mingw compilers. If you only need numpy and not scipy, there are no
>> dependencies, and only the C compiler is needed. It should compile out
>> of the box.
>>
>> Otherwise, you will need to compile your own blas/lapack, which is
>> complicated on windows. The blas/lapack sources are to be downloaded
>> from netlib.org:
>>
>> http://netlib.org/lapack/
>>
>> You should not use the version 3.2 (3.1.1 is fine, and is the version
>> used to build the numpy binaries).
>>
>> cheers,
>>
>> David
> 
> numpy compiles fine with mingw, but I never managed to build an egg. I'm
> not able to get the --compiler=mingw32 to be recognized.

I think it might work if you separate the compiling and bdist_egg steps:

	setupegg.py build --compiler=mingw32 bdist_egg

Does this help?

-- 
Pauli Virtanen




More information about the NumPy-Discussion mailing list