Travis Oliphant <oliphant@ee.byu.edu> writes: ...
Thanks for the information. This looks like a genuine bug. You should not be getting this much of a difference. Is the problem related to gcc 3.4.2? Is anybody using that particular compiler? The problem is the statlib.swilk function. Apparently, It is not returning the correct value on your system. This function is in Fortran and so either gcc 3.4.2 is compiling it differently or the f2py-generated interface is not functioning correctly.
OK, the problem seems to be related to the gcc version. When I compile scipy CVS with gcc 3.3.1 "python -c 'import scipy ; scipy.test()'" finishes without error, even with
python Python 2.4a3 (#1, Sep 9 2004, 20:49:30) [GCC 3.4.2] on linux2
Could you manually run the test that is failing and give the output of stats.shapiro(x2) using the x2 in the test_morestats.py file near line 40?
I get 0.959027051926 1.0 from print w,pw Regards Berthold -- bhoel@web.de / http://starship.python.net/crew/bhoel/