Rust: `async` blocks in a const-context error on nightly, but not on stable

[DutchGhost]

The following code compiles on stable, but not on nightly. I know this is realy bogus code, and I'm not sure it should be rejected:

fn regress() -> [(); { core::mem::ManuallyDrop::new(async { 0 }); 4 }] {
    todo!()
}

On nightly, the error message is:

error: `from_generator` is not yet stable as a const fn
 --> src/lib.rs:2:60
  |
2 | fn regress() -> [(); { core::mem::ManuallyDrop::new(async { 0 }); 4 }] {
  |                                                            ^^^^^^^^^
  |
  = help: add `#![feature(gen_future)]` to the crate attributes to enable

error: aborting due to previous error

error: could not compile `playground`

To learn more, run the command again with --verbose.

[lcnr]

cc @ecstatic-morse in case this is related to what you were doing recently

[lzutao]

searched nightlies: from nightly-2020-08-20 to nightly-2020-09-30
regressed nightly: nightly-2020-09-24
searched commits: from https://github.com/rust-lang/rust/commit/0da58007451a154da2480160429e1604a1f5f0ec to https://github.com/rust-lang/rust/commit/8b4085359ae798dedb05c95ad42520557bd25320
regressed commit: https://github.com/rust-lang/rust/commit/f6d59207ad56e24a2bfefa3544de48a0f6491363

bisected with cargo-bisect-rustc v0.5.2

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

cargo bisect-rustc 2020-08-20 --end 2020-09-30 --preserve --regress=error --without-cargo 

---

That's #76850

[ecstatic-morse]

cc @rust-lang/wg-const-eval This code was allowed erroneously; async-await has not been stabilized in a const-context. I believe we have exceptions to the backwards compatibility rules for cases like this? Actually doing anything with the async block (.await, dropping it) will cause an error during const-checking, so this is useless as far as I know.

This also works in a regular const initializer:

const _: i32 = { core::mem::ManuallyDrop::new(async { 0 }); 4 };

[camelid]

Hmm, the span in the error shown above seems wrong:

error: `from_generator` is not yet stable as a const fn
 --> src/lib.rs:2:60
  |
2 | fn regress() -> [(); { core::mem::ManuallyDrop::new(async { 0 }); 4 }] {
  |                                                            ^^^^^^^^^
  |
  = help: add `#![feature(gen_future)]` to the crate attributes to enable

It's pointing at the section from the space before the 0 to the space after the 4? But the error from putting it into a fresh playground looks correct:

error: `from_generator` is not yet stable as a const fn
 --> src/lib.rs:1:53
  |
1 | fn regress() -> [(); { core::mem::ManuallyDrop::new(async { 0 }); 4 }] {
  |                                                     ^^^^^^^^^^^
  |
  = help: add `#![feature(gen_future)]` to the crate attributes to enable

@DutchGhost is that how the error appeared to you when you ran into it?

[camelid]

Temporarily assigning P-medium per the prioritization working group discussion.

[oli-obk]

Yea this was never intended to get stabilized and since it's only possible in code that doesn't actually do anything, this is unlikely to occur in practice. I don't see a reasonable way to allow this. Note the same effect is possible with std::mem::forget.

[camelid]

So should we remove regression-from-stable-to-nightly? It sounds like it's not a regression; it's a bug fix.

[RalfJung]

I agree that this falls under "accidental stabilization", and is not a regression.

This also works in a regular const initializer:

Still works, or used to work? If it still works that arguably is a bug,

[ecstatic-morse]

The second. You can try it in the playground.

I'm assigning myself to give a better error message in these cases, but this will remain an error going forward.

[DutchGhost]

Since the PR improving the error message for cases like this has been merged, this issue can be closed?