eigh hangup

[Numpty]

numpy version
1.0.4

sys.version
2.5 (r25:51908, Sep 19 2006, 09:52:17) [MSC v.1310 32 bit (Intel)]

import numpy

cov = numpy.array([
    [ 77.70273908,   3.51489954,  15.64602427],
    [  3.51489954,  88.97013878,  -1.07431931],
    [ 15.64602427,  -1.07431931,  98.18223512],
])
# or any array I've tried


# This never returns (not within minutes anyway)
vals, vecs = numpy.linalg.eigh(cov)

eig() works and using the gui examples tool from the LAPACK 3.1.1 Windows installation, it seems that the DSYEVD routine mentioned in the docstring succeeds with these inputs.

[nils]

I cannot reproduce this behavior with

>>> numpy.__version__
'1.0.5.dev4854'

on linux.

>>> vals
array([  68.60568999,   89.57756725,  106.67185574])
>>> vecs
array([[ 0.86777139,  0.1334124 ,  0.47872116],
       [-0.17432834,  0.98379858,  0.04183273],
       [-0.46538419, -0.11975591,  0.87696983]])

[cdavid]

This is most likely due to the way numpy 1.0.4 was built (see #628, #652). I add a regression test in svn r4903, and for 1.0.5, I think we will be more careful about the blas/lapack used for the windows binaries.