Currently test_largefile is failing on the Windows buildbots because an fstat() call in Modules/_io/fileio.c is failing. fstat() returns a 32-bit size, but the file being opened is larger than 2GB.

This appears to be a change in the CRT where it would previously succeed with an incorrect value but now returns an error. (While we're here, there is no exception set in this case, so you get "SystemError: NULL result without error in PyObject_Call", but that's not the core issue.)

I can fix this instance, but I suspect we may need to fix this in multiple places. fstat64 (or _fstat64) seems to exist on multiple platforms, and the docs I found (e.g. http://linux.die.net/man/2/fstat64) suggest that EOVERFLOW is always going to occur in this case, so is there any reason not to switch to fstat64 everywhere? Maybe enable it through pyport.h/pyconfig.h and have a #define that is set based on availability?