Close inference failure

[Kimundi]

Error:

<anon>:51:16: 51:83 error: cannot determine a type for this bounded type parameter: unconstrained type
<anon>:51     assert_eq!(match_indices(s, |c: char| c == 'b').collect::<Vec<(uint, uint)>>(),
                         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
<std macros>:1:1: 14:2 note: in expansion of assert_eq!
<anon>:51:5: 52:40 note: expansion site
playpen: application terminated with error code 101

Code:

trait Matcher {
    fn next_match(&mut self) -> Option<(uint, uint)>;
}
struct CharPredMatcher<'a, 'b> {
    str: &'a str,
    pred: |char|:'b -> bool
}
impl<'a, 'b> Matcher for CharPredMatcher<'a, 'b> {
    fn next_match(&mut self) -> Option<(uint, uint)> {
        None
    }
}

/////////////////////////

trait IntoMatcher<'a, T> {
    fn into_matcher(self, &'a str) -> T;
}
impl<'a, 'b> IntoMatcher<'a, CharPredMatcher<'a, 'b>> for |char|:'b -> bool {
    fn into_matcher(self, s: &'a str) -> CharPredMatcher<'a, 'b> {
        CharPredMatcher {
            str: s,
            pred: self
        }
    }
}

/////////////////////////

struct MatchIndices<M> {
    matcher: M
}

impl<M: Matcher> Iterator<(uint, uint)> for MatchIndices<M> {
    fn next(&mut self) -> Option<(uint, uint)> {
        self.matcher.next_match()
    }
}

fn match_indices<'a, M, T: IntoMatcher<'a, M>>(s: &'a str, from: T) -> MatchIndices<M> {
    let string_matcher = from.into_matcher(s);
    MatchIndices { matcher: string_matcher }
}

/////////////////////////

fn main() {
    let s = "abcbdef";
    assert_eq!(match_indices(s, |c: char| c == 'b').collect::<Vec<(uint, uint)>>(),
               vec![(1u, 2u), (3, 4)]);
}

It works if the closure gets rooted in its own variable like this:

fn main() {
    let s = "abcbdef";
    let f = |c: char| c == 'b';
    assert_eq!(match_indices(s, f).collect::<Vec<(uint, uint)>>(),
               vec![(1u, 2u), (3, 4)]);
}

cc @nikomatsakis

[nikomatsakis]

I don't know precisely what's going on but I do know that we do some funky things with closure types. Probably this is an interaction with our over-reliance on expected types somehow, though it's not yet quite obvious to me.

[ghost]

This is fine now but needs a test case.