A strange processing methods visibility

[bozaro]

rustc 1.0.0-nightly (2b01a37 2015-02-21) (built 2015-02-22)

I create file with content:

use std::hash::SipHasher;
use std::hash::Hasher;

fn main() {
}

fn compile_fail() {
    let hasher = SipHasher::new();
    hasher.write(&[0, 1, 2]);
}

fn compile_work() {
    let sip_hasher = SipHasher::new();
    let hasher: &mut Hasher = &mut sip_hasher;
    hasher.write(&[0, 1, 2]);
}

And have compilation errors:

private.rs:9:2: 9:26 error: method `write` is private
private.rs:9    hasher.write(&[0, 1, 2]);
                ^~~~~~~~~~~~~~~~~~~~~~~~
error: aborting due to previous error

I do not understand what is happening due to this error, because this cast is essentially doing nothing.

[edwardw]

SipHasher has both a private write method and a public one implemented in Hasher trait for it. The former shadows the later. Probably related to #12808.

[edwardw]

This is rather confusing from user's point of view. Because by definition, a private method doesn't show up in the rustdoc. So looking up rustdoc doesn't help diagnosing this particular error.

I wonder what is the optimal solution here. Maybe an help message of some sort?

[steveklabnik]

I've created a self-contianed example of this behavior:

mod foo {
    pub struct Foo;

    impl Foo {
        fn bar(&self) {
            println!("private bar in foo");
        }
    }

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

    impl Baz for Foo {
        fn bar(&self) {
            println!("public bar in foo from Baz");
        }
    }
}

fn main() {
    let mut f = foo::Foo;
    f.bar(); // doesn't work

    let b: &mut foo::Baz = &mut f;

    b.bar(); // works
}

/cc @rust-lang/lang . I am not sure that this is fixible? It would seem like a big change to modify these rules.

[aturon]

Nominating for discussion at next lang team meeting.

[nikomatsakis]

As @edwardw said, this is (to my mind) the same "FIXME" as #12808 (which just talks about fields). I think we can probably change this, because we'd just be making more code compile, but I've not thought super hard about that.

[nikomatsakis]

In @rust-lang/lang meeting we decided that this should indeed follow the same lookup rules as #12808.

[nikomatsakis]

triage: P-medium

[jseyfried]

@nikomatsakis This is fixed in the above PR -- I'll add @steveklabnik's example as a test.