enable elision in where-clauses on functions

[nikomatsakis]

The following test case does not presently work, but ought to:

trait Trait<'a> { }

// Supporting `'_` here is #15872:
impl<T> Trait<'_> for T { }

// Supporting `'_` here is this issue:
fn foo<T>(t: T)
where T: Trait<'_>
{ 
}

fn main() { 
  foo(22);
}

This is, I believe, blocked on #15872, because the same work that enables '_ to expand to an early-bound region in that work ought to make supporting '_ in functions fairly straightforward.

[nikomatsakis]

Actually, I thought we maybe didn't want to permit '_ in this position, since we had not yet decided on what it means -- e.g., does it mean "some fresh lifetime name" or does it mean for<'a> T: Trait<'a>?

[nikomatsakis]

It turns out we accidentally started accepting '_ in some of these positions; this was discovered in #54902

[nikomatsakis]

It will become a hard-error again

[jackh726]

Retagging as wg-traits; supporting elision feels trait-y to me

[jackh726]

Visiting for types triage. This feels a bit dated to us. If this is something that is desired in the future, this should probably go throug an RFC.

[nikomatsakis]

A few thoughts here:

I think we need to draft up a coherent RFC about what '_ means in all the places. I am of the opinion that it should mean "add a region to the innermost potential binder spot", which in this case would mean where T: Trait<'_> would mean where T: for<'a> Trait<'a>. I also think impl Foo should follow the same rules, so that T: Trait<'_, impl Debug> would be short for T: for<'a, T: Debug> Trait<'a, T>. But the topic is complex and hence the need for a lang team initiative to pursue