[SciPy-user] Probs with scipy.test() and seg faults

[Pearu Peterson]

On Fri, 6 Jun 2003, J Analytis wrote:

> Hey,
> I've copied the two relevant sections from the scipy.test() output below. 

Thanks.

> The 
> first is to do with finding clapack, but has corrected the problem by using 
> flapack.
> The errors are related to incomplete BLAS sources. INSTALL.txt says I can cure 
> the prob by using ATLAS or the official release of BLAS libraries. But I'm 
> using atlas 3.5.2 so I should be ok.

These undefined srotmg_ errors indicate that ATLAS was not found.

> How do i find out if scipy is finding the ATLAS libraries correctly (assuming 
> that i've installed atlas properly!)? Is ATLAS incomplete? (liblapack.a is 
> about 5MB, and I wasn't able to find it using the command 
> python scipy_core/scipy_distutils/system.py
> as suggested in the release notes.)

If system_info.py does not show that it detected ATLAS libraries then
also scipy installation cannot find ATLAS.

Where did you install ATLAS libraries, what is its directory contents? 
You can use ATLAS environment variable to indicate the location of ATLAS
libraries.

> Also do I need to remove the existing scipy if I need to rebuild/reinstall it?

No. However removing scipy build directory might be a good idea before 
rebuilding scipy, also
  rm -f Lib/linalg/{clapack,flapack,cblas,fblas}.pyf
might be required, especially when having issues with finding ATLAS
libraries. See also comments in the header of Lib/linalg/setup_linalg.py.

Pearu