If you want to hack on pycurl, here's our list of unresolved issues:
* Callback handling for stream seek.
* Add docs to the high-level interface.
* Using certain invalid options, it may be possible to cause a crash.
This is un-Pythonic behaviour, but you somewhere have to draw a
line between efficiency (and feature completeness) and safety.
There _are_ quite a number of internal error checks, but tracking
and catching all possible (deliberate) misuses is not a goal
(and probably impossible anyway, due to the complexity of libcurl).