expand unions in order to solve generic overloads more precisely

[rouge8]

Summary

Apologies I don't quite know how to describe this other than an example:

In https://play.ty.dev/af42dbca-2851-49a9-beb2-4ef0b56c7a1b, ty thinks y is int | None while mypy (correctly) thinks it's int. I think this is related to this mypy PR: python/mypy#15846

ty infers the correct type without the overloads, but mypy does not ☹️ .

Version

ty 0.0.14 (Homebrew 2026-01-27)

[carljm]

Pyright and pyrefly both seem to behave the same way we do here. I don't think our behavior is wrong, since T in the second overload can solve to int | None. But I see why it's a useful feature to expand the union in this case, in order to solve T to a more precise type.

Thanks for the report!

[dhruvmanila]

A possible solution would be to have some heuristic to detect this kind of scenarios (an argument with union type that matches a parameter with generic type variable?) and invoke argument type expansion because, at least in this case, it should then resolve to int. More precisely, it would resolve to NoReturn | int which is then simplified to int.

Edit: Carl mentioned the same thing 😅