Type inference breakage in 1.80 has not been handled well

[josh]

@jplatte

Maybe traits would have to opt into single-applicable-impl-affects-inference behavior starting Edition 2027 or somesuch?

That seems likely, yes; some discussion on Zulip led to the same suggestion.

@jplatte

Specifically, we depend on type inference figuring out the first generic parameter T from the fact that only a single of the Handler impls applying to any specific type you'd want to pass to a routing function. Usually, the handler type is a closure but there's also a use for non-closure types like Layered implementing Handler.

Interesting. It seems like there's enough information there for the compiler to infer the type without using the only-one-impl rule, but that's a good use case to verify when checking that.