Holding a non-Send Copy type across a yield should be allowed

[jyn514]

For non-Send types with trivial drop, we still consider them held across the generator, even with -Zdrop-tracking:

async fn trivial_drop() {
    // uncomment to make this program compile
    // {
    let x: *const usize = &0; //~ ERROR has type `*const usize` which is not `Send`
    // }
    async {}.await;
}

fn assert_send<T: Send>(_: T) {}

fn main() {
    assert_send(trivial_drop());
}

There's no reason to do this; *const T has no drop impl and it's not possible to read or write it if it's not explicitly used after the yield.

Originally posted by @jyn514 in #98754 (comment)

[jyn514]

cc @tmandry, not sure if people have explicitly thought about this or it just happens to be forbidden by the current implementation.

[eholk]

We talked about this issue in triage last time and this time (because I forgot to mark it as triaged the first time).

Thanks for the PR! I'll take a closer look at it soon. It looks simpler than I was expecting, so that's a pleasant surprise. I was afraid fixing this issue would need a liveness pass, but maybe we already have the information we need?

The first time we talked about this there was some question about whether we want this behavior. They main concerns were around backwards compatibility and how complex we want the algorithm here to be. Although if it turns out we already have the analysis we need, I think that makes the case stronger for accepting this. At any rate, it's probably good to loop in T-Lang.

@rustbot label +AsyncAwait-Triaged

[joshtriplett]

Seems fine to me from a lang perspective. I do think this needs a lang FCP for the behavior change, though.

[joshtriplett]

@rfcbot merge

[rfcbot]

Team member @joshtriplett has proposed to merge this. The next step is review by the rest of the tagged team members:

• @cramertj
• @joshtriplett
• @nikomatsakis
• @pnkfelix
• @scottmcm

Concerns:

• early-storage-dead-of-stack-value (Holding a non-Send Copy type across a yield should be allowed #99104 (comment))
• needs-exact-proposed-rule (Holding a non-Send Copy type across a yield should be allowed #99104 (comment))

Once a majority of reviewers approve (and at most 2 approvals are outstanding), this will enter its final comment period. If you spot a major issue that hasn't been raised at any point in this process, please speak up!

See this document for info about what commands tagged team members can give me.

[scottmcm]

What's the definition of "holding" here? Since there's no use after the await, I'm unsure what's complaining.

EDIT: Oh, I guess this is Copy as a marker for "promises not to need any Dropping later"?

[joshtriplett]

Confirming something: this is using Copy as a proxy for "definitely not Drop", right? This is not actually proposing to copy the value?

(If we were copying the value, we should have the lint about large copies. But if we're not copying the value, that's not an issue.)