> The former. If C allows it then what's the point of special-casing
> Py_MIN and Py_MAX to disallow it?

"C allows it" includes cases like "C allows an the result to be
implementation-defined, or an implementation-defined signal to be
raised", and indeed, some compilers do raise signals in the cases
where they are allowed to.

I'd rather not have code in the Python implementation that raises
implementation-defined signals on some systems and gives
implementation-defined results on other systems.

> Again, if this is a serious issue (which I don't think it is)

I agree.

> it would be better handled by choosing the appropriate compiler options.

It might well be that this *is* the appropriate compiler option (even
though it's not a compiler command line flag, but a compiler language
extension).

In any case, there seems to be agrement that this is not a serious
issue (Victor said he didn't catch any new errors when using this),
so I'm rejecting the patch.