dead_code lint wrongly warns about "never used" functions that are, in fact used

[glandium]

Code

mod ffi {
    use super::*;

    extern "C" {
        pub fn DomPromise_AddRef(promise: *const Promise);
        pub fn DomPromise_Release(promise: *const Promise);
    }
}

#[repr(C)]
#[allow(unused)]
pub struct Promise {
    private: [u8; 0],
    __nosync: ::std::marker::PhantomData<::std::rc::Rc<u8>>,
}

pub unsafe trait RefCounted {
    unsafe fn addref(&self);
    unsafe fn release(&self);
}

unsafe impl RefCounted for Promise {
    unsafe fn addref(&self) {
        ffi::DomPromise_AddRef(self)
    }
    unsafe fn release(&self) {
        ffi::DomPromise_Release(self)
    }
}

Output with 1.79.0:

nothing

Output with 1.80.0:

warning: function `DomPromise_AddRef` is never used
 --> src/lib.rs:5:16
  |
5 |         pub fn DomPromise_AddRef(promise: *const Promise);
  |                ^^^^^^^^^^^^^^^^^
  |
  = note: `#[warn(dead_code)]` on by default

warning: function `DomPromise_Release` is never used
 --> src/lib.rs:6:16
  |
6 |         pub fn DomPromise_Release(promise: *const Promise);
  |                ^^^^^^^^^^^^^^^^^^

warning: `playground` (lib) generated 2 warnings

The original code didn't have the #[allow(unused)], which was added because of the "never constructed" dead_code lint, which I guess would be #126169. Adding a constructor does make the function never used errors go away, so it seems #[allow(unused)] doesn't have enough power.

[apiraino]

WG-prioritization assigning priority (Zulip discussion).

@rustbot label -I-prioritize +P-medium

[apiraino]

bisected to rust-lang-ci@c137109

cc @mu001999

[mu001999]

like #126169

[glandium]

#126169 is about the struct itself, and there's a allow(unused). This is about the side effects.

[mu001999]

#126169 is about the struct itself, and there's a allow(unused). This is about the side effects.

Oh, I see.

@rustbot claim