I am finding a very strange error.
I have a program that integrates a partial differential equation using
fortran code in an f2py module. Data in arrays held in the f2py module
is updated in the fortran code. The module arrays are exposed to
python (but not changed by python code). After each increment of
timestep the array is plotted using matplotlib, and a figure saved
with savefig, so as to create an animation.
What I am finding is that the fortran module array is becoming
corrupted with NaNs.
If, however, I simply comment out the call to savefig in my python
code, there is no corruption of the module arrays.
WxAgg backend, Intel Mac OS 10.5.6, python 2.5.2, matplotlib svn
revision 6739, numpy 1.3.0.dev6297
I have the same problem with GtkAgg python 2.5.1 + matplotlib svn
6757.+ numpy 1.3.0.dev6297 on suse linux 10.1, amd_64.