[SciPy-dev] Problems with building scipy

[Pearu Peterson]

On Thu, 2 Dec 2004, Nils Wagner wrote:

> Now there is only one "failure" in scipy.test() using FORTRAN LAPACK/BLAS.
>
> ======================================================================
> FAIL: check_heev_complex (scipy.lib.lapack.test_lapack.test_flapack_complex)
> ----------------------------------------------------------------------

Fixed in CVS. This failure appeared when using Fortran BLAS.

> Meanwhile, I have also test gcc3.3.4, gcc3.3.2 and gcc3.3.1 
> (SuSE9.2/SuSE9.1 /SuSE9.0). Unfortunately, I got the same segmentation 
> faults as mentioned earlier when using ATLAS 3.6 / 3.7.8. This fact is 
> (at least from my point of view) very unsatisfactory.

Comments:
(i) Have you tried using ATLAS libraries from
       http://www.scipy.org/download/atlasbinaries/linux/
     ?
(ii) Always use ldd on flapack.so to check that system blas/lapack 
libraries are not being used. Also verify the linking command for 
flapack.so and the existence of .a files used in -l.. and that -L..
options are properly ordered so that correct .a files are picked up.

Pearu