Negative interaction between &self, auto-ref, and @T implementations #3585

@erickt stumbled across a negative interaction between the current auto-ref/auto-deref rules, explicit self, and implementations for @T.

If you write code like this:

trait Foo {
    fn foo(&self) -> ~str;
}

impl<T: Foo> @T: Foo {
    fn foo(&self) -> ~str {
        fmt!("@%s", (**self).foo())
    }
}

impl uint: Foo {
    fn foo(&self) -> ~str {
        fmt!("%u", *self)
    }
}

fn main() {
    let x = @3u;
    assert x.foo() == ~"@3";
}

You get an assertion failure. What happens is that auto-ref derefs the @uint to uint and then auto-refs that to &uint. This is because the current rule is "auto-deref until you can't, then auto-ref". It would be easy enough to make it try auto-refing at each stage. Probably it should, this seems like an important use case. This is a fairly small change to middle/typeck/check/method.rs.