Using extra lifetime parameter when implementing a trait is allowed

[bjorn3]

trait Abc {
    fn abc<'tcx>(&self);
}

struct MyAbc;

impl Abc for MyAbc {
    //     vv this lifetime parameter doesn't exist in the original trait
    fn abc<'a, 'tcx>(&self) {}
}

(Playground)

Errors:

   Compiling playground v0.0.1 (/playground)
warning: struct is never constructed: `MyAbc`
 --> src/lib.rs:5:1
  |
5 | struct MyAbc;
  | ^^^^^^^^^^^^^
  |
  = note: #[warn(dead_code)] on by default

    Finished dev [unoptimized + debuginfo] target(s) in 0.40s

[jonas-schievink]

This might be expected. Methods in an impl are matched with the trait method using subtyping, so effectively there's these two method types:

• for<'self, 'tcx> fn(&'self Self)
• for<'self, 'a, 'tcx> fn(&'self Self)

And to me it doesn't look terribly wrong to consider those compatible (or even equivalent).

[bjorn3]

And to me it doesn't look terribly wrong to consider those compatible (or even equivalent).

It's just surprising :)

[pnkfelix]

T-compiler triage: assuming this is nominated for T-lang. Leaving unprioritized.

[eddyb]

It's linted by unused_lifetimes (if you turn that lint on, e.g. #![deny(unused_lifetimes)]), at least.

[pnkfelix]

pre-triage: removing T-compiler under assumption that this is the lang teams issue, not compiler's.

[nikomatsakis]

this is indeed intended behavior. I am therefore closing in favor of this bug on the reference repository rust-lang/reference#637. Thanks!