Experiment: Let SupportsDunderLT return object.

[randolf-scholz]

Fixes #12562

[srittau]

I'm not a fan of this change. This would decrease the type safety for the common case. For example, TestCase.assertGreater really wants an object that returns a bool and not just any overloaded object.

[JelleZijlstra]

The change here would affect objects that do have an __lt__ or similar method, but where that method doesn't return a bool. I don't think that's likely to cause a lot of real unsafety.

[JelleZijlstra]

This seems right to me but I retriggered mypy-primer to get another look.

[srittau]

I'm still opposed to this. Overloading the comparison operators to return something other than bool is unexpected and using # type: ignore to make this explicit is acceptable.

[jorenham]

I'm still opposed to this. Overloading the comparison operators to return something other than bool is unexpected and using # type: ignore to make this explicit is acceptable.

I'd agree with you if it indeed would be fixeable with one or two # type: ignore comments.
But as I've shown in #12562, this unfortunately isn't the case for user-defined types that return something other than a bool from e.g. __lt__, which currently is rejected when used with e.g. builtins.min.

[randolf-scholz]

I can't help but be reminded of this comment python/mypy#1020 (comment).

As someone who uses numpy daily, I don't find it surprising that < or > might return something other than a bool. If one finds that surprising or a code smell, using a linter might be more appropriate. Imo, the stubs should reflect runtime behavior as best as possible.

[randolf-scholz]

@srittau @JelleZijlstra With this PR seemingly being stuck, can we at least consider revisiting the SupportsBool approach #12939 ? If this weaker modification is agreeable to @srittau, maybe we can merge it and reconsider this PR at a later point.

[github-actions]

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

ibis (https://github.com/ibis-project/ibis)
- ibis/expr/types/relations.py:2924: error: Value of type variable "SupportsRichComparisonT" of "sorted" cannot be "NumericValue | float"  [type-var]

jax (https://github.com/google/jax)
+ jax/_src/pallas/mosaic/pipeline.py:981: error: Unused "type: ignore" comment  [unused-ignore]