[GAT] Associated type lifetime on regular type

[c410-f3r]

After playing with Generic Associated Types, the following snippet was successfully created:

#![feature(generic_associated_types)]

pub trait AsRef2 {
  type Output<'a> where Self: 'a;

  fn as_ref2<'a>(&'a self) -> Self::Output<'a>;
}

impl<T> AsRef2 for Vec<T> {
  type Output<'a> where Self: 'a = &'a [T];

  fn as_ref2<'a>(&'a self) -> Self::Output<'a> {
    &self[..]
  }
}

#[derive(Debug)]
struct Foo<T>(T);
#[derive(Debug)]
struct FooRef<'a, U>(&'a [U]);

impl<U> AsRef2 for Foo<Vec<U>> {
  type Output<'a> where Self: 'a = FooRef<'a, U>;

  fn as_ref2<'a>(&'a self) -> Self::Output<'a> {
    FooRef(self.0.as_ref2())
  }
}

fn main() {
    let foo = Foo(vec![1, 2, 3]);
    dbg!(foo.as_ref2());
}

But then I was wondering if it is possible to generalize the above statement for every Foo<T> with T: AsRef2. Here is what was accomplished so far:

#![feature(generic_associated_types)]

pub trait AsRef2 {
  type Output<'a> where Self: 'a;

  fn as_ref2<'a>(&'a self) -> Self::Output<'a>;
}

impl<T> AsRef2 for Vec<T> {
  type Output<'a> where Self: 'a = &'a [T];

  fn as_ref2<'a>(&'a self) -> Self::Output<'a> {
    &self[..]
  }
}

#[derive(Debug)]
struct Foo<T>(T);
#[derive(Debug)]
struct FooRef<'a, U>(&'a [U]);

impl<'b, T, U> AsRef2 for Foo<T>
where
    // * `for<'b, 'c> T: AsRef2<Output<'b> = &'c [U]>>` does not work
    //
    // * `U` is unconstrained but should be allowed in this context because `Output` is
    // an associated type
    T: AsRef2<Output<'b> = &'b [U]>,
    U: 'b
{
  type Output<'a> where Self: 'a = FooRef<'a, U>;

  fn as_ref2<'a>(&'a self) -> Self::Output<'a> {
    FooRef(self.0.as_ref2())
  }
}

fn main() {
    let foo = Foo(vec![1, 2, 3]);
    dbg!(foo.as_ref2());
}

Any clues? Feel free to close this issue if duplicated.

[jackh726]

Short blurb here while I'm looking at this and remember the problem: in the GATs case, 'b is not considered constrained inT: AsRef2<Output<'b> = &'b [U]>, U isn't constrained.

[jackh726]

GATs issue triage: not blocking. @nikomatsakis is supposed to post a comment at some point with some more details, but the general gist is to be able to do something like for<'b> T: AsRef2<Output<'b> = &'b [U]>, we need some sort of implied bounds or additional reasoning. Not going to really be possible for initial GATs release; kind of blocked on Chalk.

[c410-f3r]

Closing in favor of #88526