[r+] Implement fully qualified UFCS expressions.

[eddyb]

Working towards #16293, this adds support for <T as Trait>::method in expressions.

[rust-highfive]

r? @nikomatsakis

(rust_highfive has picked a reviewer for you, use r? to override)

[eddyb]

@nikomatsakis I almost want to separate ast::Path into a Vec<ast::PathSegment> and a terminal ast::PathSegment - most of the code out there that touches paths uses .last().unwrap() and having to produce ast::Path from ast::QPath isn't really optimal (but I chose it instead of duplicating lots of resolve and typeck logic).

[nikomatsakis]

r+. This looks great. The one thing I'd like to see is a more diverse array of tests, for example:

1. A run-pass test where the trait has a type parameter (<Foo as Trait<X>>::item()).
2. A run-pass test where the trait has a region parameter.
3. A compile-fail test where the self type that is provided doesn't match what the user provides (<i32 as Trait>::method(3_u32)).

[nikomatsakis]

@eddyb regarding Path, I personally would like to refactor it into something recursive, rather than the current vector of path segments. I don't think the vector is particularly helpful. Having something recursive (a cons list, basically) would also make trivial to extract the last name, as you say.

[huonw]

Is it worth having some tests with default type parameters (in traits and types)?

Also, does using _ work? e.g. <Option<_> as PartialEq>::eq(&Some(1u8), &None).

[eddyb]

@huonw yes it does - also, the implementation results in <_ as Foo>::bar being identical to Foo::bar, type-wise.

[eddyb]

I don't think I handle <T as PolyTrait>::foo correctly, given trait PolyTrait<T> - that is, it works, just like PolyTrait::foo does, where the type param is optional.
I guess I'll just use instantiate_trait_ref, which will require that all type params are present, and add a compile-fail test for this case.

[nikomatsakis]

as we said on IRC, I actually expected <T as PolyTrait>::foo to make fresh type variables and work. But I could probably go either way on this, given that we have _.