Travis E. Oliphant wrote:
I think a numpy-ism crept into the code-base (not that I particularly mind as I will welcome the day matplotlib jettisons it's 3-array support).
So will I! I don't have John's saintly patience, although I understand his rationale. My error in this case was failing to test a change I made with all three packages, although I try to do that whenever I think there might be a problem.
In quiver.py, nx.nan is attempted, which raises an error, unless I'm missing the definition of nan somewhere.
Well, it was working at least with the version of numpy I had when I made the change in quiver to use nan internally as a mechanism for supporting masked array input. I have updated to svn, and it works with your change of course, so all is well. I added nan definitions for Numeric and numarray which work on my linux box, but I have not tested them on other platforms; I guess in the worst case, if it doesn't work on Mac or Windows I will have to change quiver.py to use masked arrays internally instead of nans. (The definition I used is "nan = float('nan')".) I used nan here because it considerably simplifies the code, now that nan is supported in the primary backends.