Wrapping<_> implements ShX<usize> but wrapping_shX takes u32

[jethrogb]

This is mostly just slightly confusing.

impl ShX for Wrapping<_> (the code refers to #23545, I'm not sure how this is related).

wrapping_shX

[nagisa]

The same way Shr/Shl are implemented for all primitive integrals with all primitive integral types for RHS, Shr/Shl should be implemented ∀ Wrapping<T>, where T: primitive with all primitive integral types for RHS.

[Mark-Simulacrum]

Does anyone know why this is? It was added by @aturon, and it seems it was intended to be implemented for more types: #23549 (comment). I suspect that this is a side-effect of #23545 since it causes inference regressions, but I'm uncertain (and that seems odd, since I'd expect most/all uses of shifting operations to have the types fairly well defined explicitly...).

[clarfonthey]

I ran into this problem when working on #50465; it appears that neither Wrapping(1u32) >> 2 nor Wrapping(1u32) >> Wrapping(2) work today without type ascription.

IMHO, it makes the most sense to get rid of the Shr<usize> and Shl<usize> impls and use Wrapping on both sides, but that would be a breaking change. I assume that there's no way to deprecate trait impls either…

Ideally, we'd want to implement all of the necessary impls for Wrapping >> Wrapping and Wrapping << Wrapping, using all combinations of integer types like the non-wrapping versions, then somehow deprecate the current, lonely usize impl. Ideally, we could outright remove it, but I doubt that'd happen any time soon.

[Enselic]

Triage: I believe this issue can be closed after the disabled impls here have been enabled, which is requires #23545 to be fixed first. Please correct me if that is wrong.