[SciPy-Dev] Many errors/failures running scipy.test() using latest numpy and scipy
Vincent Schut
schut at sarvision.nl
Fri Jul 1 03:59:51 EDT 2011
On 06/30/2011 10:22 PM, Pauli Virtanen wrote:
> On Thu, 30 Jun 2011 13:20:37 +0200, Vincent Schut wrote:
> [clip]
>> anything I could do to diagnose this further?
>
> This is a bit difficult, or at least time-consuming. One thing that
> can be done is to diff the assembler output from Gfortran 4.6 and 4.4
> of the iterative solvers, find the points where it differs, and then
> check if the Fortran code does something suspicious at the
> corresponding point.
>
> If someone else wants to spend some quality time with assembler
> listings before I find time etc. for this, please go ahead :)
>
>> I'd like to have a working scipy again...
>
> These failures have always been there, it's just that now
> we have better test coverage so that we actually catch them...
>
> Pauli
Hmm, my assembler skills are a bit rusty... :-)
I'll just keep gfortran-4.4 installed and as the default fortran
compiler than, for the time being. Just to be sure: I suppose there is
no problem mixing gfortran-4.4 for scipy's fortran code and gcc-4.6 for
the c code parts? As a precaution, I have redone my atlas/blas/lapack
with gfortran-4.4 also, so the entire scientific python stack is
gfortran-4.4 based, now.
Off to the testing failures of scikits-learn, then.
Thanks,
Vincent.
More information about the SciPy-Dev
mailing list