unhelpful error message for missing (first) self param on traiit method

[jbclements]

This code:

trait CtxtFn{
    pub fn f9(uint) -> uint;
}

pub fn use_ctxtfn<T : CtxtFn>(cf: T) -> uint {
    cf.f9(342)
}

...signals the error message:

error: type `T` does not implement any method in scope named `f9`

...which is kind of crazy, because T is bounded by CtxtFn, and I can see the definition of f9.
<before reading on: can you see the problem?>

I tore my hair out for quite a while before realizing I'd forgotten the "self" parameter on the decl of f9. This is just a papercut, but it seems like a big one to me. Fixing this--that is, signalling an error like "method f9 is declared without a self parameter"--would presumably require a pass in resolve which scanned for methods without a self parameter.

[jdm]

I too lost a bunch of time to this papercut recently.

[metajack]

A complete test case:

trait CtxtFn {
    fn f9(uint) -> uint;
}

fn use_ctxtfn<T: CtxtFn>(cf: T) -> uint {
    cf.f9(342)
}

fn main() {
}

This is still an issue. Since multiple rust devs have hit this themselves, nominating for production ready.

[catamorphism]

Accepted for production-ready

[jdm]

The relevant code is in check_method_call (http://mxr.mozilla.org/rust/source/src/librustc/middle/typeck/check/mod.rs#1916). I think performing another check for a static trait method that otherwise matches would make a ton of sense here. Perhaps using lookup_in_trait (http://mxr.mozilla.org/rust/source/src/librustc/middle/typeck/check/method.rs#122) which has an optional self expression argument?

[pnkfelix]

Assigning P-low, not 1.0 blocker.