Trait methods can return &T types without lifetimes

[erickt]

It doesn't seem to matter whether or not an abstract trait method returns a &T type without a region:

trait A {
    fn foo(&self) -> ∫
}

struct B {
    x: int,
}

/*
impl A for B {
    fn foo(&self) -> &int { &self.x }
}
*/

fn main() {
}

Compiles fine. However, once impl A for B is uncommented, it properly errors with:

foo.rs:2:28: 2:40 error: cannot infer an appropriate lifetime due to conflicting requirements
foo.rs:2     fn foo(&self) -> &int { self.bar() }
                                     ^~~~~~~~~~~~
foo.rs:2:26: 2:40 note: first, the lifetime cannot outlive the anonymous lifetime #1 defined on the block at 2:26...
foo.rs:2     fn foo(&self) -> &int { self.bar() }
                                   ^~~~~~~~~~~~~~
foo.rs:2:28: 2:33 note: ...due to the following expression
foo.rs:2     fn foo(&self) -> &int { self.bar() }
                                     ^~~~~
foo.rs:2:26: 2:40 note: but, the lifetime must be valid for the anonymous lifetime #2 defined on the block at 2:26...
foo.rs:2     fn foo(&self) -> &int { self.bar() }
                                   ^~~~~~~~~~~~~~
foo.rs:2:28: 2:40 note: ...due to the following expression
foo.rs:2     fn foo(&self) -> &int { self.bar() }
                                     ^~~~~~~~~~~~

What's interesting is that the impl cannot correct for the error in the trait. Changing it to this:

impl A for B {
    fn foo<'a>(&'a self) -> &'a int { &self.x }
}

Results in this error:

foo.rs:10:4: 10:47 error: method `foo` has an incompatible type: expected concrete lifetime, but found bound lifetime parameter &
foo.rs:10     fn foo<'a>(&'a self) -> &'a int { &self.x }
              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

I feel we should error out on the trait method or else we risk someone creating an unimplementable trait.

[bluss]

Compare with issue #8309; &int is effectively the same as &'static int here, which makes it implementable, but relatively useless.

[nikomatsakis]

Also a dup of #2478 I suspect

[flaper87]

Triage bump

Still an issue and the example in the report is still valid.

[alexcrichton]

This is now an error or expected with lifetime elision, so closing.