> It's not worth staring at code to try to figure out things like that. > Run it under gdb, examine the relevant data objects when it crashes > and see what's broken (e.g. some pointer is unexpectedly null), then > set a conditional breakpoint that stops execution when that incorrect > condition happens (e.g. the pointer gets set to null), run the program > til the breakpoint triggers, see what other condition X caused the > condition that caused the crash, set another breakpoint that triggers > when X occurs, etc. Work backwards systematically til you find the bug.
Yeah, not 30 seconds after I posted it I found the problem. Under some (funny) conditions the extension would be called with a list instead of a tuple. My debug case (of course) worked fine but the full test would fail. Sorry for the noise, -Daryl -- http://mail.python.org/mailman/listinfo/python-list
