Generic impls of traits don't autoderef

[alexcrichton]

pub struct Foo;

pub trait Bar {
    fn bar(&self);
}

pub trait Baz {}

impl<T: Baz> Bar for T {
    fn bar(&self) {}
}

impl Baz for Foo {}

pub fn foo(t: ~Foo) {
    t.bar() // ~Foo doesn't implement Baz
    (*t).bar() // ok b/c Foo implements Baz
}

fn main() {}

From a code-writer point of view, it'd be nice for t to auto-deref to (*t).bar() to appease the compiler, but I can see that the compiler may not know a whole lot about this.

If we don't get auto-deref here, perhaps a better error message may work. It's a little confusing to see that ~Foo doesn't implement Baz when Foo clearly implements Baz

[emberian]

cc @eddyb

this would be pretty convenient.

[eddyb]

@cmr we have to see how @nikomatsakis intends to improve method lookup.

Though I have to point out that the X doesn't implement Y error comes from vtable checking which comes after type-checking finishes, and I'm pretty sure it has access to concrete types (without type inference variables at least. region inference isn't resolved until after regionck).

That said, if the type information is available at the time of the method lookup, the vtable error can be prevented by searching for another suitable candidate (but this would also complicate simpler cases).

[nikomatsakis]

This actually has nothing to do with auto-deref, but it is addressed by the method lookup algorithm that I'm working on. That is ALMOST ready for publication, just working on the last bits.

[japaric]

@alexcrichton updated example is working with a recent version:

rustc 0.13.0-dev (97a57ec90 2014-11-07 11:31:25 +0000)

I think Box<T> now derefs to &T although Box<T> doesn't implement Deref<T> (#18624)

[alexcrichton]

Looks to be so, thanks @japaric! Flagging as needstest.