bpo-39350: Fix fractions for int subclasses

[vstinner]

Fix regression in fractions.Fraction if the numerator and/or the
denominator is an int subclass.

https://bugs.python.org/issue39350

[vstinner]

That's an alternative fix for https://bugs.python.org/issue39350

See PR #18309 to discussion (the other proposed fix).

[vstinner]

I tested manually that fractions.Fraction works again with numpy.int64: it works as expected.

See: https://bugs.python.org/issue39350#msg361483

[mdickinson]

The change LGTM, but we should make it clear in the Misc/NEWS entry that there's an actual behaviour change here from 3.8: it's not simply a case of a regression being introduced and then fixed.

[vstinner]

@mdickinson: I completed the documentation, would you mind to review it?

@mdickinson: You approved my PR. Does it mean that you are ok to always divide the numerator and denominator with an int, which is a Python 3.9 incompatible change? I'm fine with that, since I propose the change. But I prefer to confirm with you, since you proposed PR #18309 :-)

[mdickinson]

Still LGTM, especially with the versionchanged update. Thank you!

Does it mean that you are ok to always divide the numerator and denominator with an int, which is a Python 3.9 incompatible change?

@vstinner Yes, I'm personally fine with that change; I don't know how other interested parties might feel about it, and there seems a risk that people won't even notice that there's a behaviour change proposed and implemented without a dedicated b.p.o. issue. OTOH, I think it ought to be relatively uncontroversial, so perhaps we don't need that issue and discussion.

[vstinner]

@vstinner Yes, I'm personally fine with that change; I don't know how other interested parties might feel about it, and there seems a risk that people won't even notice that there's a behaviour change proposed and implemented without a dedicated b.p.o. issue. OTOH, I think it ought to be relatively uncontroversial, so perhaps we don't need that issue and discussion.

I really liked the discussion :-) It was required to discuss how other types are handled. At the beginning, I didn't think about types other than int. At least, my PR now adds a test for an int subclass :-)