Lifetime inference misses what seems like a simple case

[zwarich]

I distilled this from an example in some actual code I have, although I'm not entirely sure that the distilled example reflects the problem in my original code. The program

struct A;

struct B<'a> {
    a: &'a A,
    b: &'a A,
}

impl <'a> B<'a> {
    fn foo<'b>(&'b self) {
        B {
            a: self.a,
            b: self.b,
        };
    }
}

fn main() { }

fails to compile with this error:

test.rs:10:9: 13:10 error: cannot infer an appropriate lifetime for lifetime parameter `'a due to conflicting requirements
test.rs:10         B {
test.rs:11             a: self.a,
test.rs:12             b: self.b,
test.rs:13         };
test.rs:12:16: 12:22 note: first, the lifetime must be contained by the expression at 12:15...
test.rs:12             b: self.b,
                          ^~~~~~
test.rs:12:16: 12:22 note: ...so that automatically reference is valid at the time of borrow
test.rs:12             b: self.b,
                          ^~~~~~
test.rs:11:16: 11:22 note: but, the lifetime must also be contained by the expression at 11:15...
test.rs:11             a: self.a,
                          ^~~~~~
test.rs:11:16: 11:22 note: ...so that automatically reference is valid at the time of borrow
test.rs:11             a: self.a,
                          ^~~~~~

If I change the program slightly so that foo returns a B<'b> then everything works fine:

struct A;

struct B<'a> {
    a: &'a A,
    b: &'a A,
}

impl <'a> B<'a> {
    fn foo<'b>(&'b self) -> B<'b> {
        B {
            a: self.a,
            b: self.b,
        }
    }
}

fn main() { }

[zwarich]

@nikomatsakis @pnkfelix Any clue what's going on here?

[nikomatsakis]

I'm not sure, I have to investigate.

[Ryman]

This is now fixed: http://is.gd/DjKFae

[jooert]

Triage: A regression test has been added in #31527; can be closed now.