[SciPy-dev] SciPy SELinux and import bugs?

[Tom Loredo]

Hi folks-

I've just "upgraded" to RHEL 5 and am going through the chore of
installing a ton of software that I previously used under Fedora,
including Python/NumPy/SciPy.

I've gotten numpy-1.0.3.1 installed (with atlas & fftw) and it passes
all tests.  But with scipy-0.6.0, scipy.test(1,1) gives me an
"Illegal instruction" error, and crashes out of Python.  The first
such error was an SELinux violation due to a text relocation
requirement in _fftpack.so (whatever that means!).  I executed
the "chcon" context change command SELinux suggested as a workaround.
But scipy.test(1,1) still crashes with "Illegal instruction" at
apparently the same place, though now without raising an SELinux
violation (at least, not one that appears in the SELinux troubleshooter).

At this point I thought I'd give the current SVN SciPy (r3369) a try.
It installs fine, but I get an error on import:

>>> import scipy
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python2.5/site-packages/scipy/__init__.py", line 63, in <module>
    pkgload(verbose=SCIPY_IMPORT_VERBOSE,postpone=True)
  File "/usr/local/lib/python2.5/site-packages/numpy/_import_tools.py", line 177, in __call__
    self._init_info_modules(packages or None)
  File "/usr/local/lib/python2.5/site-packages/numpy/_import_tools.py", line 100, in _init_info_modules
    self._init_info_modules(getattr(info_module,'depends',[]))
  File "/usr/local/lib/python2.5/site-packages/numpy/_import_tools.py", line 72, in _init_info_modules
    exec 'import %s.info as info' % (package_name)
  File "<string>", line 1, in <module>
  File "/home/inference/loredo/python/scripts/special.py", line 97, in <module>
    def factrl(n, ntop=0, prev=N.ones((33),N.Float)):
AttributeError: 'module' object has no attribute 'Float'

It is finding an old script of mine that happens to be named "special"
that I haven't used and thus haven't upgraded from Numeric to NumPy.
But that's beside the point---I think it must be a bug that SciPy's
import mechanism is finding it.

Any suggestions for dealing with either issue (Illegal instruction
with 0.6.0, or the import issue for r3369) would be appreciated!

-Tom Loredo

PS:  Regarding the text relocation issue in _fftpack.so, SELinux 
points to this web page for hints on how to fix it:

http://people.redhat.com/drepper/selinux-mem.html


-------------------------------------------------
This mail sent through IMP: http://horde.org/imp/