Trait method resolution is overly eager for nested pointers & types

[huonw]

&T has an impl of Clone for all T (i.e. (&T).clone() should always work), however having &TypeThatConditionallyImplementsClone<T> is not always clonable, because (it appears that) the compiler will lock itself into using the Clone impl for the inner type, but find that T doesn't satisfy the requirements and so error.

#[allow(dead_code)];

struct NoClone;

#[cfg(error)]
fn err() {
    let x = Some(NoClone);
    let y = &x;
    let _: &Option<NoClone> = y.clone();
}

fn ok() {
    let x = NoClone;
    let y = &x;
    let _: &NoClone = y.clone();
}

fn main() { }

Without --cfg error it compiles fine, but with it:

pointer-option-method.rs:9:31: 9:41 error: failed to find an implementation of trait std::clone::Clone for NoClone
pointer-option-method.rs:9     let _: &Option<NoClone> = y.clone();
                                                         ^~~~~~~~~~

[reem]

Triage. This arises with basically anything that implements Deref including Arc and Box<SomethingClone>.

[japaric]

@huonw example is working with a recent version:

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

[Gankra]

The DList iterator that claims to not be able to derive clone seems to be able to now. Marking needs test.