Represent lifetimes as Names instead of Idents

[edwardw]

Closes #7743.

[huonw]

FWIW, do we really want lifetimes to be non-hygienic? i.e. should

macro_rules! foo ( ($t: ty) => { fn foo<'a>(a: &'a int, b: $t) { ... } })

foo!(&'a int)

be valid?

(This is particularly important if we ever get macros being able to expand to methods (#4621), in which case the above with &'a self instead of a: &'a int would do weird/bad things with

impl<'a> Foo<'a> {
     foo!(&'a int) // meant to be Foo's 'a.
}

)

(Sorry to be annoying about hygiene...)

[huonw]

That said, reverting this patch shouldn't be particularly hard if we do implement hygiene for them later...

[pnkfelix]

I'm inclined to agree with @huonw here: As his example illustrates, a macro can expand into a fn that binds a lifetime. This leads me to think that lifetimes should respect hygiene. (I guess I mention this only because huon has already said that macros-expanding into methods are a potential future motivation, but I would think that the current motivations are sufficient.)

I do not understand the motivation outlined in the description on #7743: "This will allow them to be safely compared without fear of runtime-fail." Is this just talking about when writing rustc internal code? Or writing macro expanders?

In any case I'd prefer runtime-fail over having active footguns that are hidden by lack of hygiene.

@jbclements can you elaborate?

[edwardw]

If a decision is made, I can always revert this patch. One caveat is since #12338 has turned lifetime renaming on, we may hit ast.rs#L67.