matplotlib-devel Mailing List for matplotlib

Hi! Last year, I have been working on making matplotlib compatible with PyPy
here:
https://github.com/shmuller/matplotlib.git
I succeeded to get the svg and pdf backends working, but then had to drop
the ball. My changes involve mostly the C++ parts in src, in particular
replacing some PyCXX (which didn't work with PyPy) with the plain Python C
API. I also needed some minor fixes in .py files.
I understand that it is desirable to remove the PyCXX layer altogether from
matplotlib, and I have come a good deal along this route. I would also be
happy to continue working in this direction.
I have a bit of time now, and so I merged my branch with upstream/master.
The few conflicts were easy to resolve. However, then I tried to run the
test suites (python tests.py), both on the current master and my branch, to
see if I broke anything.
Unfortunately, the test suites randomly segfault on my computer (Macbook Air
2011), at the latest when I try to run them 10 times in a row. Not only on
my branch, but also on upstream/master. I ran the Apple Hardware Test, which
didn't report any issues, so my Mac seems to be fine. I went back to the
state of my branch before merging, which i have been using in production for
a year now without any issues, and noticed that even there the test suite
segfaulted if I try to run it 10 times in a loop. I haven't noticed this
before (and didn't have any issues in production use of matplotlib for a
year). The crashes occur randomly and not always at one specific test.
- Did anyone have this issue before?
- Could someone pull my branch and run the test suite on a different
computer, so that I can exclude that my hardware is at fault after all?
- Could someone please guide me to get my work pulled into the main
repository?
Many thanks,
Stefan
--
View this message in context: http://matplotlib.1069221.n5.nabble.com/Trying-to-prepare-a-pull-request-but-tests-py-keeps-crashing-on-master-tp43680.html
Sent from the matplotlib - devel mailing list archive at Nabble.com.

Once #3277 and #3284 get merged I think we will have squashed all the
reported code bugs from rc1 (both are from me and need a second set of
eyes).
Can someone who understands the freetype code take a look at #3262 to
see if we can relax the version requirement on the installation?
Everything else outstanding is documentation related, if those three
things get taken care of tomorrow, I will tag rc2 tomorrow night.
Tom
-- 
Thomas Caswell
tca...@gm...