Bug with numpy 1.7.0

[guyer]

As reported [http://thread.gmane.org/gmane.comp.python.fipy/2963 on the list]:

I think I found a bug with numpy 1.7.0 and Fipy (latest git).
The best example is the circle diffusion problem in which I can't
initialize the viewer. I try to debug the problem and is seems to be
related to numpy.take for which indices should be integers. It
complains to cast from float64 to int64 !

The best workaround I found is to tweak the tools/numerix.py and
modifiy the following lines :

596: taken = a.take(indices.astype('int64'), axis=axis)   
601: taken = MA.take(a,MA.filled(indices,0).astype('int64'),axis=axis) 
620: taken = NUMERIX.take(a, indices.astype('int64'),axis=axis)
622: taken = MA.take(a, indices.astype('int64'),axis=axis)

I used Gentoo and all the python package are on the edge :
dev-python/numpy-1.7.0
dev-python/matplotlib-1.2.0-r2
sci-libs/scipy-0.11.0-r1

I hope it will help !!!

Imported from trac ticket #539, created by guyer on 03-18-2013 at 11:32, last modified: 03-20-2013 at 15:51

[guyer]

The issue is because of fipy/meshes/gmshMesh.py@28de7a8107fb#L859. If maxVerts-len(v) is zero, then concatenation is with an empty list, but NumPy 1.7 apparently casts each of the concatenates to an array before concatenating them. Compare:

>>> import numpy as np
>>> np.__version__
'1.7.0'
>>> np.concatenate(([1, 2, 3], []))
array([ 1.,  2.,  3.])

with

>>> import numpy as np
>>> np.__version__
'1.6.1'
>>> np.concatenate(([1, 2, 3], []))
array([1, 2, 3])

The solution is to manually cast the (possibly empty) list of [-1] to an array of int before concatenation.

Trac comment by guyer on 03-19-2013 at 10:38

[guyer]

Pull request

The following changes since commit 8c51478a3e8c30618b3a96c132612a46735c7548:

  Merge branch 'ticket496-FIPY_DISPLAY_MATRIX_is_broken' into develop (2013-02-11 12:18:26 -0500)

are available in the git repository at:


  ssh://git@code.matforge.org/nist/fipy.git ticket539

for you to fetch changes up to 9ae4801ab0d2f314a7b51dafcc155eb142da0b2d:

  Fix for issue #373 (2013-03-19 10:40:29 -0400)

----------------------------------------------------------------
Jonathan Guyer (1):
Fix for issue #373

 fipy/meshes/gmshMesh.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Trac comment by guyer on 03-19-2013 at 13:16

[wd15]

Tests pass on Buildbot and in a virtualenv with numpy 1.7. Merged back to develop with [0c8a5a6/fipy].

Trac comment by wd15 on 03-20-2013 at 12:59