False positive unused_lifetimes warning

[Hawk777]

I tried this code:

#![warn(unused_lifetimes)]

pub struct Thingy<'x, X> {
	x: &'x mut X,
	y: &'static str,
}

impl<'x, X> Thingy<'x, X> {
	pub fn new(x: &'x mut X) -> Self {
		Self{x, y: "hello"}
	}

	pub async fn foo(&mut self) -> &'static str {
		self.y
	}
}

I got a warning “lifetime parameter 'x never used”, on the impl line. But it clearly is used. The warning goes away if I delete the function foo, or make it a regular function instead of an asynchronous function.

Meta

rustc --version --verbose:

rustc 1.46.0
binary: rustc
commit-hash: unknown
commit-date: unknown
host: x86_64-unknown-linux-gnu
release: 1.46.0
LLVM version: 10.0

I saw #61115 and rust-lang/rust-clippy#3988, but those were both closed at least a year ago and this is still happening in 1.46.0 which was just released recently, so I assume it’s not the same bug.

[lcnr]

The issue is that the field x is never marked as used, which is required to mark 'x as used, as x is the only field using 'x.

This is not ideal, but not sure on where exactly we want to lint here. If we have functions f and g with f calling g and g calling f it is probably good to lint for both f and g so I don't think just marking everything used in unused things as used is the right solution here.

[Hawk777]

Then why does the warning not show up if foo is not async? x would be equally unused in that case, but there's no unused lifetime warning.

[Hawk777]

Furthermore, this code, which clearly does use x, still has the unused lifetime warning, but of course not the unused field warning:

#![warn(unused_lifetimes)]

pub struct Thingy<'x> {
	x: &'x u32,
	y: &'static str,
}

impl<'x> Thingy<'x> {
	pub fn new(x: &'x u32) -> Self {
		Self{x, y: "hello"}
	}

	pub async fn foo(&self) -> &'static str {
		self.y
	}

	pub async fn bar(&self) -> u32 {
		*self.x
	}
}

Meanwhile, if I change both async fns to fn, the warning goes away. It seems to be triggered by the presence of any async fn in the impl.

[lcnr]

Yeah, that is an actual bug 👍 thanks

[nikomatsakis]

So this is probably related to the desugaring somehow:

rust/compiler/rustc_ast_lowering/src/lib.rs

Lines 1882 to 1927 in 4d52dc4

	// When we create the opaque type for this async fn, it is going to have
	// to capture all the lifetimes involved in the signature (including in the
	// return type). This is done by introducing lifetime parameters for:
	//
	// - all the explicitly declared lifetimes from the impl and function itself;
	// - all the elided lifetimes in the fn arguments;
	// - all the elided lifetimes in the return type.
	//
	// So for example in this snippet:
	//
	// ```rust
	// impl<'a> Foo<'a> {
	// async fn bar<'b>(&self, x: &'b Vec<f64>, y: &str) -> &u32 {
	// // ^ '0 ^ '1 ^ '2
	// // elided lifetimes used below
	// }
	// }
	// ```
	//
	// we would create an opaque type like:
	//
	// ```
	// type Bar<'a, 'b, '0, '1, '2> = impl Future<Output = &'2 u32>;
	// ```
	//
	// and we would then desugar `bar` to the equivalent of:
	//
	// ```rust
	// impl<'a> Foo<'a> {
	// fn bar<'b, '0, '1>(&'0 self, x: &'b Vec<f64>, y: &'1 str) -> Bar<'a, 'b, '0, '1, '_>
	// }
	// ```
	//
	// Note that the final parameter to `Bar` is `'_`, not `'2` --
	// this is because the elided lifetimes from the return type
	// should be figured out using the ordinary elision rules, and
	// this desugaring achieves that.
	//
	// The variable `input_lifetimes_count` tracks the number of
	// lifetime parameters to the opaque type *not counting* those
	// lifetimes elided in the return type. This includes those
	// that are explicitly declared (`in_scope_lifetimes`) and
	// those elided lifetimes we found in the arguments (current
	// content of `lifetimes_to_define`). Next, we will process
	// the return type, which will cause `lifetimes_to_define` to
	// grow.

but I don't exactly know how yet. Maybe applying the desugaring to the example will help to see it.

[Hawk777]

I can also trigger a single_use_lifetimes warning with a free function:

#![warn(single_use_lifetimes)]
pub async fn foo<'a>(x: &'a u32) -> &'a u32 {
	x
}

Again, this only happens with an async fn, not with a regular fn, so I suspect it’s probably the same underlying cause (though if you disagree I’m happy to file a separate ticket).

[ghost]

I also encountered this bug, and if I add #[allow(unused_lifetimes)] on the async fn (instead of the impl), the warning disappears, in both my code and the code of this issue.
I also read #78522. Is that the same bug? The warnings are both fired on the async fn itself.

[Vooblin]

@hyd-dev Yes, it is the same bug and I fixed it in this #79689 PR.

[Vooblin]

@rustbot claim

[tmandry]

Presumably fixed in #79689, leaving this open to try the original example in the next nightly.

[tmandry]

Triage: Confirmed that this is fixed in nightly.