dead code lint is incorrectly reporting code as being unused with static methods

[erickt]

I was playing with passing around a trait that only has static methods. It works okay, but it appears that this confuses the dead code lint. Here's an example:

trait Visitor<R> {
    fn foo();
}

fn visit<V: Visitor<R>, R>() {
    Visitor::<R>::foo()
}

struct V;

impl Visitor<()> for V {
    fn foo() { println!("hi") }
}

pub fn main() {
    visit()
}

Which warns with:

serde2.rs:9:1: 9:10 warning: struct is never used: `V`, #[warn(dead_code)] on by default
serde2.rs:9 struct V;
            ^~~~~~~~~

As best as I can tell, Rust is able to infer that there's only one implementation possible and uses it, but this information isn't conveyed to the lint, so it marks the struct as being unused.

[ghost]

This should be as simple as visiting param substitutions for method calls in dead.rs.

[TimNN]

Adjusting the code for current rust, this no longer produces any warnings (although I did have to use V explicitly at one place, but I don't think this can be avoided):

trait Visitor<R> {
    fn foo();
}

fn visit<V: Visitor<R>, R>() {
    V::foo()
}

struct V;

impl Visitor<()> for V {
    fn foo() { println!("hi") }
}

fn main() {
    visit::<V, ()>()
}

[apasel422]

I'm not sure that @TimNN's updated code exhibits the original issue, because it explicitly uses V.

[TimNN]

I was thinking about that when I adapted the code. Do you know a way to avoid explicitly mentioning V?

[nagisa]

From the duplicate issue:

struct A;
trait As {
    fn banana();
}

impl As for A {
    fn banana() { panic!("banana"); }
}

trait Bs<T: As> {
    fn do_it(&self) { T::banana() }
}

struct B;

impl Bs<A> for B {}

fn main() {
    Bs::do_it(&B)
}

When compiled, the dead_code lint fires, claiming that A is not used.

This code, when run, panics with banana, thus proving that A is, in fact, used.

---

Removing E-needstest, because the issue is not resolved.

[Mark-Simulacrum]

Closing in favor of #18290.