read_csv crashes when run on multiple threads

[mkleehammer]

Running multiple threads each calling read_csv crashes on OS/X. I've seen two traps, which I'll put into an attachment. Sometimes it says "Fatal Python error: GC object already tracked\nAbort trap: 6"" and sometimes "Segmentation fault: 11".

I've put together a small example to reproduce it: https://gist.github.com/4401461

I've also added some of the OS/X crash report in case it isn't crashing on your install.

I'm using the built-in Python 2.7.2 on OS/X. Pandas 0.10.0 was built locally using pip install -U pandas. I don't know if this is new to 0.10.0 since I wasn't threading 0.9 yet.

This might be relevant for the GC already tracked: http://pyrit.wordpress.com/2010/02/18/385/