Replace usage of ResumeTy in async lowering with Context

[Swatinem]

Replaces using ResumeTy / get_context in favor of using &'static mut Context<'_>.

Usage of the 'static lifetime here is technically "cheating", and replaces the raw pointer in ResumeTy and the get_context fn that pulls the correct lifetimes out of thin air.

fixes #104828 and #104321 (comment)

r? @oli-obk

[rustbot]

Hey! It looks like you've submitted a new PR for the library teams!

If this PR contains changes to any rust-lang/rust public library APIs then please comment with @rustbot label +T-libs-api -T-libs to tag it appropriately. If this PR contains changes to any unstable APIs please edit the PR description to add a link to the relevant API Change Proposal or create one if you haven't already. If you're unsure where your change falls no worries, just leave it as is and the reviewer will take a look and make a decision to forward on if necessary.

Examples of T-libs-api changes:

• Stabilizing library features
• Introducing insta-stable changes such as new implementations of existing stable traits on existing stable types
• Introducing new or changing existing unstable library APIs (excluding permanently unstable features / features without a tracking issue)
• Changing public documentation in ways that create new stability guarantees
• Changing observable runtime behavior of library APIs

[oli-obk]

since they are on the generator argument, why would they get captured?

ah this is #68923

Please reference that issue here, as we'll lose the connection now that ResumeTy is dead code

[Swatinem]

👍🏻 I slightly reworded that comment, referencing the given issue, and have also rebased the PR.

[oli-obk]

@bors r+

[bors]

📌 Commit 9dbb189 has been approved by oli-obk

It is now in the queue for this repository.

[JohnTitor]

Failed in rollup: #105335 (comment)
Seems this needs a rebase, @bors r-

[Swatinem]

Looks like this was conflicting with #105180.

I rebased, but I seem to have two mir-opt failures locally on Mac, though I don’t think they are related.

[mir-opt] src/test/mir-opt/inline/inline_generator.rs
[mir-opt] src/test/mir-opt/issues/issue_59352.rs

[oli-obk]

@bors r+

[bors]

📌 Commit cf031a3 has been approved by oli-obk

It is now in the queue for this repository.

[compiler-errors]

Heads up that this probably broke #105501.

@khuey is checking if reverting this PR makes the code compile again (that they can't share) -- This probably has to do with the same old generator-interior lifetime erasure issue, since now we have a type that has two lifetimes in the generator interior 😅.

Unclear which lifetime in &'static Context<'_> is the culprit here that's making the higher-ranked lifetime error -- we probably should be storing this as a *mut Context<'static> and transmuting it when needed :/

Anyways, at this rate, this will unfortunately probably make its way to beta, but we'll probably see regressions in the crater run after beta is cut.

[khuey]

Reverting this does indeed fix my problem.

[Swatinem]

@khuey can you do me a favor and try it with this snippet here:

https://github.com/rust-lang/rust/pull/104833/files#diff-b24a095026827d389d25a6913f9426e6cea3c0ef9b348a12a834d3a5919b8687R2720-R2743

That turns the &'static mut Context<'_> into &'static mut Context<'static>. If that doesn’t help, maybe using a *mut as @compiler-errors suggested may be worth another try.

[khuey]

Using that hunk doesn't change anything.

[Swatinem]

Oh well :-(
Thanks for checking.

I think reverting back to ResumeTy it is then, as that is a basically a Send + Sync wrapper around *mut Context<'static>.

I hope adding a #[repr(transparent)] for it, and maybe with a few changes to cg_clif to make it see through #[repr(transparent)] types may help it along to not reject it.

In the meantime, it would be nice to have a minimal repro and maybe a regression test for #105501 so that does not regress again. Either way, I should be able to open a revert PR tomorrow evening or monday.

[bjorn3]

I hope adding a #[repr(transparent)] for it, and maybe with a few changes to cg_clif to make it see through #[repr(transparent)] types may help it along to not reject it.

Please do a similar fix to #105082 instead and make sure that the generated MIR matches the function signature. Mismatches between the function signature and the MIR are bound to give issues apart from this type compatibility. For example without #105082 a projection on the return place went completely haywire causing an assignment of an i32 value to a () place which was correctly rejected with a crash.

[Fabian-Gruenbichler]

this also causes a regression for the following (reduced from actual production code ;)) code that works fine on stable and beta:

use futures::future::FutureExt;

fn main() {
    let _ = async {
        async {
            async {}.await; // await inside a `.catch_unwind()` currently fails
        }
        .catch_unwind().await
    };
}

Compiling playground v0.0.1 (/playground)
error[[E0277]](https://doc.rust-lang.org/nightly/error-index.html#E0277): the type `&mut Context<'_>` may not be safely transferred across an unwind boundary
   --> src/main.rs:5:9
    |
5   | /         async {
6   | |             async {}.await; // await inside a `.catch_unwind()` currently fails
7   | |         }
    | |         ^
    | |         |
    | |_________`&mut Context<'_>` may not be safely transferred across an unwind boundary
    |           within this `[async block@src/main.rs:5:9: 7:10]`
    |
    = help: within `[async block@src/main.rs:5:9: 7:10]`, the trait `UnwindSafe` is not implemented for `&mut Context<'_>`
    = note: `UnwindSafe` is implemented for `&std::task::Context<'_>`, but not for `&mut std::task::Context<'_>`
note: future does not implement `UnwindSafe` as this value is used across an await
   --> src/main.rs:6:21
    |
5   | /         async {
6   | |             async {}.await; // await inside a `.catch_unwind()` currently fails
    | |                     ^^^^^^ await occurs here, with the value maybe used later
7   | |         }
    | |         -
    | |         |
    | |_________the value is later dropped here
    |           has type `&mut Context<'_>` which does not implement `UnwindSafe`
note: required by a bound in `futures::FutureExt::catch_unwind`
   --> /playground/.cargo/registry/src/github.com-1ecc6299db9ec823/futures-util-0.3.25/src/future/future/mod.rs:433:23
    |
433 |         Self: Sized + ::std::panic::UnwindSafe,
    |                       ^^^^^^^^^^^^^^^^^^^^^^^^ required by this bound in `FutureExt::catch_unwind`

For more information about this error, try `rustc --explain E0277`.
error: could not compile `playground` due to previous error

https://play.rust-lang.org/?version=nightly&mode=debug&edition=2021&gist=6da04ccbbdc8828f0872d970b0074a20

broken on current nightly as well as on a default build of 71ec145
works again on a default build of 71ec145 with cf031a3 from this PR reverted

thanks @Blub for reporting and reducing the reproducer!

[Swatinem]

Further reduced:

fn is_unwindsafe(_: impl std::panic::UnwindSafe) {}

fn main() {
    is_unwindsafe(async {
        // this needs an inner await point
        async {}.await;
    });
}

Thanks for the repro. I will work on a fix later today.