Diff frommypy_primer, showing the effect of this PR on open source code:

yarl (https://github.com/aio-libs/yarl)+ yarl/_url.py:550: error: Unused "type: ignore" comment  [unused-ignore]

I think this is a good idea. See#363 (yes, three-digit issue number!) - this removes anunreachable warning from the "should be fine" snippet there.

Wow, I overlooked this dinosaur. That's a nice tool you have!

InterpretingNotImplemented asAny seems not to be the only problem. I just played around a little. Mypy does not predict the following crash:

classA1:x=1def__add__(self,o:A1)->float:ifisinstance(o,A1):returnself.x+o.xreturnNotImplementedclassA2:x=2def__add__(self,o:A2)->float:returnself.x+o.xclassB:y=3def__radd__(self,o:A1|A2)->float:returnself.y+o.xA1()+B()# worksA2()+B()# fails  (undetected by Mypy)

I will try to find a fix for this as soon as I have some spare time.

I guess it has never been explicitly reported only because--warn-unreachable is hidden too well, many users aren't even aware that it exists.

Yes, I agree. As soon as--warn-unreachable will be included in--strict, it will likely be discussed more frequently.

I think this makes sense to me, even if it complicates the general idea of what exactly is exempt from unreachability warnings. (I
I can also see this being unnecessary :^). Given nobody has complained, I'm not sure we really need to add this as a feature (since that's ultimately what this is).

It would say it is more of a fix than a feature, becausereturn NotImplemented is reachable, and so the current warning is wrong.

Thanks for the reviews!