[pymvpa] Trouble building from source code

[Malcolm Tobias]

I'm attempting to install PyMPVA on our linux cluster.  Instead of using a 
packaged version of the software (which would have to be installed on all of 
the nodes) I've installed a separate Python instance under a shared file system 
and have used this Python to install the necessary pre-reqs.  When I'm done, 
my PyMPVA  build is flakey (it gets the wrong answer when compared against pre-
built PyMPVA packages and often prints out lots of warnings like "Warning: 
divide by zero encountered in power" and "Warning: Extremely bad integrand 
behavior occurs at some points of the integration interval.")

I'm doing this on a RHEL 5.5 system and using gfortran 4.1.2.
I initially started with Python 2.4.6, but also repeated this process starting 
with a much newer version Python 2.6.5.  After Python, I installed Numpy 
1.5.0b1, SciPy 0.8.0, Nifticlib 2.0.0, PyNIfTI 0.20090303.1 ( I had to do some 
hacks to get this to build properly.  It needed help finding the Nifti 
libraries and some of the compression libraries) and finally PyMPVA 0.4.4.
At each point I tried to run any tests that come with the various packages.
Things looked fairly reasonable (this is with Python 2.4.6):

Numpy:

  [root at login001 nose-0.11.2]# python -c 'import numpy; numpy.test()'
  ...
  Warning: divide by zero encountered in log
  Ran 2945 tests in 15.034s

  OK (KNOWNFAIL=4)

SciPy:

  [root at login001 nose-0.11.2]# python
  >>> import scipy
  >>> scipy.test()
  ...
  Warning: invalid value encountered in multiply
  Warning: divide by zero encountered in log
  Warning: invalid value encountered in multiply
  ...
  SSSS......SSSS...................................................F.Segmentation 
fault

[root at login001 pynifti-0.20090303.1]# make test
.........................
Ran 2 tests in 2.636s

FAILED (failures=1)
make: *** [testmanual] Error 1

and with Python 2.6.5:

[root at login002 src]# python -c 'import numpy; numpy.test()'
...
...........Warning: divide by zero encountered in log
...........................................................................................................................................................................................................................................
----------------------------------------------------------------------
Ran 2966 tests in 13.843s


>>> import scipy
>>> scipy.test()
...
Ran 4405 tests in 47.563s

OK (KNOWNFAIL=13, SKIP=38)
<nose.result.TextTestResult run=4405 errors=0 failures=0>


Any suggestions on what to try next?
Any tips on building from source?  Are some versions of Python easier to work 
with?

Malcolm

-- 
Malcolm Tobias
314.362.1594