Carl Meyer

Thanks for the report!

It looks to me like mypy and pyrefly are both OK with both versions (with `Self` and with `M`). Pyright errors on both versions. I agree that the two versions...

Re-titled the issue, since I think the diagnostic on the Self version is correct -- the only mystery/bug here is why the explicit-TypeVar version does not behave the same as...

> The critical takeaway from this equivalence is that `self` and `other` must be of the **exact same type** . In a concrete implementation, this means `type(self)` must be identical...

@kamalfarahani I agree with you, I think that type checkers (including ty) are behaving wrongly here: neither `IntSum` nor `StringConcat` should be considered to implement the `Semigroup` protocol in your...

Early on we had discussed the possibility of adding a special case that when an `__init__` parameter is annotated, and that parameter is assigned directly to an instance attribute in...

I'm going to put this issue into stable release because I do think we should consider ways to reduce the number of cases where we union `Unknown` into an attribute...

@bradleyharden thanks for the feedback. This is the conclusion I've been reaching as well; I think the gradual-guarantee approach is theoretically interesting but doesn't serve most users best in practice....

The reason this issue is separate from #1240 was to track the idea mentioned above that we might special-case assignments to un-annotated implicit instance attributes directly from annotated `__init__` arguments....

We have already disabled `possibly-unresolved-reference` and `possibly-missing-import` by default, due to the likelihood of false positives when libraries do complex conditional imports. (The latter being disabled by default is I...