[SciPy-dev] SciPy SVN compilation on SnowLeopard

[David Cournapeau]

Pierre GM wrote:
>
> I didn't try to recompile scipy w/ numpy r7387 yet, but this latter  
> already raises a second issue that maybe related to my scipy  
> installation problems.
> I tried to recompile some fortran libraries that I need for a project,  
> and it fails (log available here: http://pastebin.com/m7d38b395).  
> Looks like lipo can't figure the architecture of some of the temp  
> files... Any pointer would be deeply appreciated.
>   

I would try to fix the compilation problem first - the .o file, if it
exists at all (which I doubt), would certainly be garbage since the code
cannot be parsed by gcc.

>
>   
>> If all you do is using the system python (the one from apple), then
>> you only need to set up FFLAGS because gfortran does not target x86_64
>> by default.
>>     
>
> Actually, I'm using Python2.6.2 compiled from sources (via homebrew,  
> figured I could give it a try). What other flag should I set ?
>   

Then I don't know, it will depend on the flags you set yourself (or the
default ones). But python does not support 64 bits arch on mac os X yet
very well - the apple python uses some apple specific patches.

I think that for the time being, unless you really need it, you should
avoid using non-apple python if you care about 64 bits.

> Now, does the issue raised over here [http://www.nabble.com/-MacPorts---19397%3A-scipy-not-completely-universal-td23174931.html#a23174931 
> ] relevant to our problem ? 

No, because it was fixed some time ago in numpy.

cheers,

David