[SciPy-user] SciPy with gcc4 and gfortran on OS X

Robert Kern robert.kern at gmail.com
Mon Apr 10 17:37:27 CDT 2006

Zachary Pincus wrote:
> Hi folks,
> I just built the latest scipy svn version on a G5 with gfortran to  
> see how things went.
> Everything build OK, and the tests don't segfault, but all the  
> problems I reported on my G4 persist. Specifically, some tests fail  
> all the time and some fail sporadically (see below). Also, I still  
> get the 'WARNING: clapack module is empty' message, which Robert  
> informed me shouldn't be happening on OS X.

No, the clapack module should be empty if you are using the Accelerate
framework. Although the Accelerate framework provides most of ATLAS, it does not
provide the row-major versions of LAPACK routines that the clapack module wraps.

The cblas module, on the other hand, ideally shouldn't be empty. In order to fix
that, you will need to alter the test in the function generate_pyf() in
Lib/linalg/setup.py (line 79 or so) to take the Accelerate framework into
account as a special case. You will also probably have to adjust the list of
functions in the cblas module to match those provided by the Accelerate
framework. All off my scipy time is going towards scipy.stats at the moment, so
I don't have time to develop and test the appropriate choices here.

> ======================================================================
> FAIL: check_expon (scipy.stats.tests.test_morestats.test_anderson)
> ----------------------------------------------------------------------
> Traceback (most recent call last):
>    File "/Library/Frameworks/Python.framework/Versions/2.4/lib/ 
> python2.4/site-packages/scipy/stats/tests/test_morestats.py", line  
> 57, in check_expon
>      assert_array_less(A, crit[-2:])
>    File "/Library/Frameworks/Python.framework/Versions/2.4/lib/ 
> python2.4/site-packages/numpy/testing/utils.py", line 255, in  
> assert_array_less
>      assert cond,\
> AssertionError:
> Arrays are not less-ordered (mismatch 100.0%):
>          Array 1: 2.1501866413808912
>          Array 2: [ 1.587               1.9339999999999999]

Well, this is a stochastic test; it is *supposed* to fail sporadically. However,
the mismatch in the array shapes is probably indicative of a real bug. The
anderson function is on the list for review, of course.


Robert Kern
robert.kern at gmail.com

"I have come to believe that the whole world is an enigma, a harmless enigma
 that is made terrible by our own mad attempt to interpret it as though it had
 an underlying truth."
  -- Umberto Eco

More information about the SciPy-user mailing list