[Python-modules-team] Bug#448530: A workaround to this problem.

[Ondrej Certik]

On Dec 22, 2007 5:40 PM, Kumar Appaiah <akumar at ee.iitm.ac.in> wrote:
> Dear Jan,
>
> With Ondrej's help, I've managed to find a workaround to this
> problem. Rather, let me say that this issue isn't numpy's, but more of
> Atlas'.
>
> The issue you menion here happens when we link numpy against
> Atlas. This now causes us to depend on the BLAS provided by Atlas
> instead of the reference BLAS, i.e. we depend on libatlas.so.3 instead
> of refblas3. I tried the build with plain refblas, and Ondrej and I
> can confirm that this problem goes away. So, I am now patching the
> build system to avoid building against Atlas (which is the right way).
>
> However, this still doesn't explain why linking against Atlas causes a
> problem, and Ondrej and I have decided to test it out and leisure, and
> if necessary, file a bug against Atlas.
>
> Thanks!
>
> Kumar

Thanks Kumar for fixing this!


> P.S. The Atlas in Debian is soooo old now! *sigh*.

Yes, it really sucks.

Ondrej