[Tim Peters] ... >> The most common cause for "impossible exceptions" > is flawed C code in an extension that fails to >> check a Python C API call for an error return.
[Mark English] > Yes, I use a lot of C modules which I wrote. Then you know where to start looking <wink>. > It could certainly be one of them (although which one could be an > interesting challenge to find out). It may even have something to do > with moving to the VC7.1 compiler. Hard to say in advance. Since you have VC7.1, that opens a world of possiblities if you recompile Python. There are extra checks in ceval.c in a debug build, trying to catch "a call lied about whether it raised an exception" soon(er) after it happens. Rebuilding all your extensions in debug mode too could be a PITA, so you could get the same effect by #define'ing CHECKEXC in a release build. But a debug build has all sorts of extra checks that can help, and you could get into the debugger then and set a breakpoint on the unexpected OverflowError ... a quick check of the source shows that "long int too large to convert to int" is produced only by longobject.c's PyLong_AsLong() function. There's no way we _should_ be getting into that function by applying min() to floats, of course. > Thank you very much for your suggestion. Without it I would never have > thought to look at that code. Nobody ever does <wink>. -- http://mail.python.org/mailman/listinfo/python-list
