lowering: extend temporary lifetimes around await

[davidtwco]

Fixes #63832.

r? @matthewjasper
cc @nikomatsakis

[Centril]

Could you please also update the doc comment on fn lower_expr_await to accurately reflect the new desugaring and then follow up with a PR against the reference (See rust-lang/reference#635 for @nikomatsakis's initial PR.)?

[davidtwco]

Thanks @Centril, fixed those comments and opened rust-lang/reference#674.

[matthewjasper]

@bors r+

[bors]

📌 Commit 63fad69 has been approved by matthewjasper

[RalfJung]

Is it expected that this is a breaking change? A Miri testcase is failing now:

2019-09-10T19:17:04.0564321Z normalized stderr:
2019-09-10T19:17:04.0564553Z error[E0597]: `x` does not live long enough
2019-09-10T19:17:04.0565934Z   --> $DIR/async-fn.rs:18:13
2019-09-10T19:17:04.0566063Z    |
2019-09-10T19:17:04.0566130Z 18 |     async { x + y }.await
2019-09-10T19:17:04.0566458Z    |     --------^------
2019-09-10T19:17:04.0566539Z    |     |     | |
2019-09-10T19:17:04.0566641Z    |     |     | borrowed value does not live long enough
2019-09-10T19:17:04.0566727Z    |     |     value captured here by generator
2019-09-10T19:17:04.0566873Z    |     a temporary with access to the borrow is created here ...
2019-09-10T19:17:04.0566971Z 19 | }
2019-09-10T19:17:04.0567254Z    | -
2019-09-10T19:17:04.0567323Z    | |
2019-09-10T19:17:04.0567409Z    | `x` dropped here while still borrowed
2019-09-10T19:17:04.0567524Z    | ... and the borrow might be used here, when that temporary is dropped and runs the destructor for type `impl std::future::Future`
2019-09-10T19:17:04.0567653Z    |
2019-09-10T19:17:04.0568341Z    = note: The temporary is part of an expression at the end of a block. Consider forcing this temporary to be dropped sooner, before the block's local variables are dropped. For example, you could save the expression's value in a new local variable `x` and then make `x` be the expression at the end of the block.
2019-09-10T19:17:04.0568497Z 
2019-09-10T19:17:04.0568622Z error[E0597]: `y` does not live long enough
2019-09-10T19:17:04.0568922Z   --> $DIR/async-fn.rs:18:17
2019-09-10T19:17:04.0569038Z    |
2019-09-10T19:17:04.0569103Z 18 |     async { x + y }.await
2019-09-10T19:17:04.0569731Z    |     ------------^--
2019-09-10T19:17:04.0569811Z    |     |     |     |
2019-09-10T19:17:04.0569908Z    |     |     |     borrowed value does not live long enough
2019-09-10T19:17:04.0569999Z    |     |     value captured here by generator
2019-09-10T19:17:04.0570108Z    |     a temporary with access to the borrow is created here ...
2019-09-10T19:17:04.0570189Z 19 | }
2019-09-10T19:17:04.0570464Z    | -
2019-09-10T19:17:04.0570534Z    | |
2019-09-10T19:17:04.0570619Z    | `y` dropped here while still borrowed
2019-09-10T19:17:04.0570726Z    | ... and the borrow might be used here, when that temporary is dropped and runs the destructor for type `impl std::future::Future`
2019-09-10T19:17:04.0570849Z    |
2019-09-10T19:17:04.0571489Z    = note: The temporary is part of an expression at the end of a block. Consider forcing this temporary to be dropped sooner, before the block's local variables are dropped. For example, you could save the expression's value in a new local variable `x` and then make `x` be the expression at the end of the block.
2019-09-10T19:17:04.0571685Z 
2019-09-10T19:17:04.0571753Z error: aborting due to 2 previous errors

[yoshuawuyts]

Not sure if this was intended to cause breakage, but we had to patch our code in async-std likely because of this PR: https://github.com/async-rs/async-std/pull/182/files.

before

 peer.send(format!("from {}: {}\n", from, msg)).await?

after

let msg = format!("from {}: {}\n", from, msg);
peer.send(msg).await?

[seanmonstar]

So are the above examples a regression?

[Centril]

Depends... can you create a minimal reproducer for before/after?

[tmandry]

This causes errors to crop up all over the Fuchsia code base. Based on #63832 (comment), it seems like this might have been expected, however.

[seanmonstar]

@Centril The change in miri is quite succinct, and the "before" doesn't appear to be bad... while the "after" seems odd? https://github.com/rust-lang/miri/pull/946/files#diff-ecc42d2008d431ab52ee698995abd41b

Is something more needed?

Inlined:

Before:

async fn add(x: u32, y: u32) -> u32 {
    async { x + y }.await
}

After:

async fn add(x: u32, y: u32) -> u32 {
    let a = async { x + y };
    a.await
}

[davidtwco]

I've confirmed locally that this patch did cause these regressions. I'm not sure what the best way to proceed here is, @nikomatsakis @Centril @matthewjasper?

[pimeys]

Seems that this patch broke my test server with Tide.

https://github.com/pimeys/blocking_test/

Works with 2019-09-10, broken in 2019-09-11.