[SciPy-User] [msvc] scipy/numpy build with visual studio and intel fortran compiler

[David]

On 04/29/2011 03:49 PM, Eloi Gaudry wrote:
> On 04/29/2011 01:37 AM, Christoph Gohlke wrote:
>>
>> On 4/28/2011 2:26 PM, Eloi Gaudry wrote:
>>> well, i somehow messed up with my setup. using the enclosed patches, i
>>> can build numpy and scipy using msvc8/9 and intel visual fortran
>>> compiler 11.1 (you need to apply the patches and use the matching
>>> build_*msvc.x*.bat file).
>>> 'hope this would help people that need to stick to visual studio
>>> compilers...
>> I have no such problems building numpy 1.5.1 and scipy 0.9 from the
>> official source releases using Visual Studio 2008 (msvc9) and Intel
>> ifort/MKL 11.1. No need to patch sources.
> christoph,
> you're right, those patches are useless (except the one targeting at
> scipy/sparse/linalg/dsolve/SuperLU/SRC/lsame.c) : those are obsolete
> modifications used for building scipy with a bad fcompiler setup, i
> shouldn't have sent those.
>> Using Visual Studio 2005 (msvc8) to build Python extensions is not a
>> good idea unless you target a custom CPython built with that compiler.
> i build python and python extensions using msvc8 for various reasons.
> what make you think this is a bad idea ?

You need to use the exact same C library for both python interpreters 
and any extension used afterwards. The only way to do that is to use the 
exact same compiler to build python and all the extensions you want to 
use. Python 2.6 and above official releases are built with MSVC 2008.

You are almost guaranteed to have weird crashes if you don't take that 
precaution while building numpy,

cheers,

David