Support move expressions in coroutine closures

[TaKO8Ki]

This adds move(expr) support for coroutine closures.

• Support for move expressions in coroutine closures
• Support for move expressions in async blocks

RFC: rust-lang/rfcs#3968
Tracking issue: #155050
Project goal:

• Ergonomic ref-counting goals#107
• https://rust-lang.github.io/rust-project-goals/2026/ergonomic-rc.html

r? @nikomatsakis

[rustbot]

Some changes occurred to MIR optimizations

cc @rust-lang/wg-mir-opt

[rustbot]

nikomatsakis is currently at their maximum review capacity.
They may take a while to respond.

[nikomatsakis]

Hmm. Something seems off here. I think we need more tests around nested-move and async blocks, async closures, and gen blocks. For example I'd expect:

let c: Arc<String> = Default::default();
let future = async {
    let f = async {
        move(c.clone());
    };
    assert!( /* c ref count is 2 */ );
    f.await;
    assert!( /* c ref count is 1 */ );
    drop(c);
};

let c: Arc<String> = Default::default();
let future = async {
    let f = async {
        move(c.clone());
    };
    f.await;
    drop(c);
};
println!("{c}"); // <-- ERROR: c is moved

let c: Arc<String> = Default::default();
let future = async {
    let f = async {
        move(move(c.clone()));
    };
    assert!( /* c ref count is 3 */ );
    f.await;
    assert!( /* c ref count is 2 */ );
    drop(c);
};
println!("{c}"); // OK, prints ""

View changes since this review

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[nikomatsakis]

The behavior around async closures still seems off to me

View changes since this review

[nikomatsakis]

✅ These match my expectations.

[nikomatsakis]

❌ But this does not. I expected the move to execute when the closure was created. In other words, I expect these to be equivalent

let created = Cell::new(0);
let c = || {
    let n = move({ created.set(created.get() + 1); created.get() });
    n
};
assert_eq!(created.get(), 1);
assert_eq!(c(), 1);
assert_eq!(c(), 1);
assert_eq!(created.get(), 1);

and

let created = Cell::new(0);
let c = async || {
    let n = move({ created.set(created.get() + 1); created.get() });
    n
};
assert_eq!(created.get(), 1);
assert_eq!(c().await, 1);
assert_eq!(c().await, 1);
assert_eq!(created.get(), 1);

I think we are leaking the "desugaring" of async closures here somehow.

[nikomatsakis]

✅ This matches my expectations

[nikomatsakis]

✅ As does this

[nikomatsakis]

✅ As does this

[nikomatsakis]

✅ As does this

[nikomatsakis]

✅ These errors match my expectations

[nikomatsakis]

✅ As does this

[nikomatsakis]

✅ As does this

[nikomatsakis]

✅ Interesting example