mysterious test_axes/fill_units failure; buildbot images back online

Development matplotlib-devel
1

Hi All,

I just spent an hour or two tracking down some problems with the buildbot test images that crept in unnoticed with the pdf backend testing. I think I now fixed all the issues with the buildbot testing, which required a few changes to the MPL source and a few on the buildbot server to build the image overview pages correctly. Anyhow, the image failure overview page is back to displaying useful information at http://mpl.code.astraw.com/overview.html

All this was prompted by a buildbot failure that occurred starting at r7958 or r7959, although these changes seem unlikely to be the cause of the issue. So, I'm a little mystified as to the actual cause, but it again looks like some freetype problem. Mike, do you think it could somehow be related to your recent font work?

-Andrew

2

Andrew Straw wrote:

I'm a little mystified as to the actual cause, but it again looks like some freetype problem. Mike, do you think it could somehow be related to your recent font work?

OK, in the absence of a fix, I just checked in the images that were being generated. They didn't look too bad -- probably just a weird freetype font hinting difference between the original and re-generated images.

Anyhow, now all the tests are passing again, and therefore we shouldn't have any more severe bugs like that in r7985 creeping in unnoticed. I take this as evidence that it's good to fix (or disable) minor failing tests right away so they don't obscure more severe failures.