That is a good question.  While it is true that errors other than 'PyExc_OverflowError', will be mapped onto a 'TypeError' I don't think that is a bad thing.  Any errors that come out of 'PyFloat_AsDouble' should be handled on a case-by-case basis and not blindly passed back out the call chain.  Otherwise, we may end up passing back errors (which are who knows what) that make sense for a caller of 'PyFloat_AsDouble', but not for callers of 'g_set'.

Also, the interface would become variable, meaning that whenever 'PyFloat_AsDouble' introduces new exceptions, then this code would too, which would lead to a somewhat unpredictable interface for callers of 'g_set'.