regression: parameter type may not live long enough

[Mark-Simulacrum]

[INFO] [stdout] error[E0310]: the parameter type `EWA` may not live long enough
[INFO] [stdout]    --> src/worker_pool.rs:109:1
[INFO] [stdout]     |
[INFO] [stdout] 109 | / {
[INFO] [stdout] 110 | |     let mut handle_vec = Vec::with_capacity(default_worker);
[INFO] [stdout] 111 | |
[INFO] [stdout] 112 | |     let mut worker_sender_vec = Vec::with_capacity(default_worker);
[INFO] [stdout] ...   |
[INFO] [stdout] 128 | |     }
[INFO] [stdout] 129 | | }
[INFO] [stdout]     | | ^
[INFO] [stdout]     | | |
[INFO] [stdout]     | |_the parameter type `EWA` must be valid for the static lifetime...
[INFO] [stdout]     |   ...so that the type `EWA` will meet its required lifetime bounds
[INFO] [stdout]     |
[INFO] [stdout] help: consider adding an explicit lifetime bound
[INFO] [stdout]     |
[INFO] [stdout] 108 |     EWA: Clone + 'static,
[INFO] [stdout]     |                +++++++++

Smells similar to #123279, but maybe not.

• https://crater-reports.s3.amazonaws.com/beta-1.78-3/beta-2024-03-30/gh/shishanyue.RW-RJR-Server/log.txt

[compiler-errors]

searched toolchains b6d2d84 through 7d3702e

---

Regression in 09bc67b

---

Attempting to search unrolled perf builds
ERROR: couldn't find perf build comment

searched nightlies: from nightly-2024-02-01 to nightly-2024-04-01
regressed nightly: nightly-2024-03-07
searched commit range: b6d2d84...7d3702e
regressed commit: 09bc67b

bisected with cargo-bisect-rustc v0.6.8

Host triple: x86_64-unknown-linux-gnu
Reproduce with:

cargo bisect-rustc --start=2024-02-01 

[compiler-errors]

I can't tell upon a quick search whether this was an expected regression from #121679/#115008. cc @lcnr @aliemjay

[aliemjay]

This is expected. This is the culprit bound:
https://github.com/shishanyue/RW-RJR-Server/blob/0343d1021fbbd3282e29377491b531d5a7bc98eb/src/worker_pool.rs#L23

Minimized:

struct WorkerPool<D: 'static>(Option<D>);

fn new_worker_pool<D>() -> impl Sized {
    async {
        WorkerPool(None::<D>)
    }
}

What is annoying though is the error span which points to the entire async block because, when inside the async block, we draw implied bounds from its output type, allowing us to assume D: 'static, which then fails to be proven in the parent fn. See #115173 for a partial fix.

[aliemjay]

Ironically, this is accepted now and since forever because the rvalue never get the chance to be checked for well-formedness:

struct WorkerPool<D: 'static>(Option<D>);

fn new_worker_pool<D>() {
    let _ = async {
        WorkerPool(None::<D>)
    };
}

[riking]

Upstream fix is shishanyue/RW-RJR-Server#13

[lcnr]

closing as expected and we have provided a PR to fix the issue

[apiraino]

@aliemjay @lcnr anything here that should be noted in the release notes?

[lcnr]

added a comment to #121679 (comment)