[SciPy-user] Problems installing scipy in SuSE 10.2

[David Cournapeau]

Christian K wrote:
> David Cournapeau wrote:
>> Christian K wrote:
>>> gfortran 4.0.xxx had some real problems, see this thread:
>>> http://thread.gmane.org/gmane.comp.python.scientific.user/6889
>>> starting from 4.1 I did not encounter any problems. However I remember having
>>> read somewhere, that it is not recommended to build atlas with gfortran.
>> Disclaimer: I am not a fortran developer, I know nothing about fortran, 
>> only about packaging software compiled with fortran compiler and related 
>> problems.
>>
>> For atlas, it is hardly a problem I think, because atlas is a C library. 
>
> Ture, then I think it was that it was recomended to use gcc3 instead of 4 :) 
Oh yes, but this is a totally different issue: this is because gcc 3 is 
(was ?) better than gcc 4 for fpu code on x86 archs (not true anymore 
for core 2 duo). Also note that it is recommended to *link* the objects 
together with the default compiler of the platform, that is: compile the 
kernel objects files with gcc 3, and then, build the whole library using 
gcc 4. This is precisely to avoid special problems arising with ABI 
issues between different compilers.
> Me
> neither have ever written a single line of fortran, but sometimes I wrap some
> fortran code with f2py.
> I used SuSE for many years but once I tried ubuntu I realised that all libs and
> packages I was fighting with to get them compiled (like many other SuSE users
> apparently) are part of the distribution and just work. Keep that in mind the
> next time you plan to upgrade your system...
Well, as far as scipy and its dependencies are concerned, from my 
limited experience (I do not use openSUSE or fedora outside testing), 
ubuntu and debian are *much* better than FC or openSUSE. You can build 
numpy and scipy without compiling anything else.

David