Pre-RFC: allow conflicting names on import, only complain if used ambiguously

[withoutboats]

@josh

You'd be allowed to use m::S and use n::S, as long as you never used the name S for anything except invoking trait methods on objects that impl one of those traits.

What about using S as a function?

[petrochenkov]

@josh

In this proposal, you would not be allowed to pub use m::S and pub use n::S; that would generate an error. So, ambiguous names should never end up in crate metadata, because you can't export them.

RFC 1560 and RFC 1422 tried hard to avoid treating private imports specially, it feels entirely wrong to me to introduce such a special treatment now. @jseyfried also described what complications arise when we try to avoid reexporting "ambiguity items".

[josh]

Using S as a function would similarly produce an error about the conflicting versions of S.