[SciPy-user] scipy.test() from trunk segfaults

David Cournapeau david@ar.media.kyoto-u.ac...
Sat Feb 9 05:51:36 CST 2008


Scott Beardsley wrote:
> I'm having a problem building/testing scipy from the svn trunk. I built 
> the ATLAS libraries from scratch (including the dynamic libs). All seems 
> to go great with the numpy install (also the trunk). numpy tests all 
> finish successfully. I've tried searching the archives and found a 
> message about adding libg2c and libm but that didn't seem to work. As 
> you can see from the log below scipy.test() segfaults:
It looks like you did not compile atlas correctly. The problem is that 
by default, ATLAS will pick up gfortran if you have it, whereas numpy 
and scipy will pick up g77 first if you have it. Since you are using RH 
3, which uses g77 for all its fortran code, you should not use gfortran 
*at all* for anything related to numpy. g77 and gfortran are *not* ABI 
compatible, and mixing code from one with the other is just asking for 
trouble. Reconfigure and rebuilt atlas with the option "-C if g77". Make 
sure that none of the softwares you are using to build numpy is using 
gfortran.

Also, I noticed that you are using some alpha softwares as libraries 
(fftw 3.2). Please use released softwares first to see if you can 
reproduce the bug.

cheers,

David


More information about the SciPy-user mailing list