Async/Await: Confusing error message when using non-Send type in async fn

[dbrgn]

With this Cargo.toml:

[package]
name = "errmsg"
version = "0.1.0"
edition = "2018"

[dependencies]
futures-preview = { version = "0.3.0-alpha.19", features = ["async-await"] }
tokio = "0.2.0-alpha.6"

...and this main.rs:

use std::rc::Rc;
use std::time::Duration;

use tokio::timer::delay;

async fn wait() {
    let rc = Rc::new(0);
    println!("Before sleep");
    delay(tokio::clock::now() + Duration::from_millis(500)).await;
    println!("After sleep");
}

#[tokio::main]
async fn main() -> Result<(), std::io::Error> {
    println!("Start");
    tokio::spawn(wait());
    println!("End");
    Ok(())
}

...I get this error message:

   Compiling errmsg v0.1.0 (/tmp/errmsg)
error[E0277]: `std::rc::Rc<i32>` cannot be sent between threads safely
   --> src/main.rs:16:5
    |
16  |     tokio::spawn(wait());
    |     ^^^^^^^^^^^^ `std::rc::Rc<i32>` cannot be sent between threads safely
    |
   ::: /home/danilo/.cargo/registry/src/github.com-1ecc6299db9ec823/tokio-0.2.0-alpha.6/src/executor.rs:100:40
    |
100 |     F: Future<Output = ()> + 'static + Send,
    |                                        ---- required by this bound in `tokio::executor::spawn`
    |
    = help: within `impl std::future::Future`, the trait `std::marker::Send` is not implemented for `std::rc::Rc<i32>`
    = note: required because it appears within the type `{std::rc::Rc<i32>, fn(std::time::Instant) -> tokio_timer::delay::Delay {tokio_timer::delay}, fn() -> std::time::Instant {tokio_timer::clock::now}, std::time::Instant, fn(u64) -> std::time::Duration {std::time::Duration::from_millis}, u64, std::time::Duration, tokio_timer::delay::Delay, ()}`
    = note: required because it appears within the type `[static generator@src/main.rs:6:17: 11:2 {std::rc::Rc<i32>, fn(std::time::Instant) -> tokio_timer::delay::Delay {tokio_timer::delay}, fn() -> std::time::Instant {tokio_timer::clock::now}, std::time::Instant, fn(u64) -> std::time::Duration {std::time::Duration::from_millis}, u64, std::time::Duration, tokio_timer::delay::Delay, ()}]`
    = note: required because it appears within the type `std::future::GenFuture<[static generator@src/main.rs:6:17: 11:2 {std::rc::Rc<i32>, fn(std::time::Instant) -> tokio_timer::delay::Delay {tokio_timer::delay}, fn() -> std::time::Instant {tokio_timer::clock::now}, std::time::Instant, fn(u64) -> std::time::Duration {std::time::Duration::from_millis}, u64, std::time::Duration, tokio_timer::delay::Delay, ()}]>`
    = note: required because it appears within the type `impl std::future::Future`
    = note: required because it appears within the type `impl std::future::Future`

If I comment out the unused rc assignment, then everything works fine.

If comment out the delay(..).await statement, it works too!

If I put the rc assignment in a block, it compiles as well (because the rc is dropped by the time await is called).

From what I know about the implementation of async-await, I think this is because a state machine is created that keeps the local variables across multiple await-points, right? In any case, the error message is quite confusing, especially when the await call and the non-Send type are not related at all.

It seems that we're back at a pre-NLL situation with the borrow checker in async functions.

In summary, two issues:

• The error message is quite confusing and not all too helpful, expecially if the user doesn't know anything about the internal implementation of async-await
• The rc could be dropped before the await call, which would resolve this issue

[nikomatsakis]

Can you clarify @dbrgn whether you are running this on nightly or on beta? We've been doing work on exactly this error message, but it may not be triggering in your particular case.

[davidtwco]

Assigning to myself to check if #65345 addresses this.

[dbrgn]

@nikomatsakis Sorry for forgetting to mention this. I was on 1.39.0-beta.6.

[davidtwco]

I can confirm that with #65345 that the error message is improved in this case:

error: future cannot be sent between threads safely
   --> src/main.rs:16:5
    |
16  |     tokio::spawn(wait());
    |     ^^^^^^^^^^^^ future returned by `wait` is not `Send`
    |
   ::: /home/david/.cargo/registry/src/github.com-1ecc6299db9ec823/tokio-0.2.0-alpha.6/src/executor.rs:100:40
    |
100 |     F: Future<Output = ()> + 'static + Send,
    |                                        ---- required by this bound in `tokio::executor::spawn`
    |
    = help: within `impl std::future::Future`, the trait `std::marker::Send` is not implemented for `std::rc::Rc<i32>`
note: future is not `Send` as this value is used across an await
   --> src/main.rs:9:5
    |
7   |     let rc = Rc::new(0);
    |         -- has type `std::rc::Rc<i32>`
8   |     println!("Before sleep");
9   |     delay(tokio::clock::now() + Duration::from_millis(500)).await;
    |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ await occurs here, with `rc` maybe used later
10  |     println!("After sleep");
11  | }
    | - `rc` is later dropped here

error: aborting due to previous error

[dbrgn]

@davidtwco thanks for checking! This is indeed much better.

Now the question is whether the rc could automatically be dropped before the await call if it's not being used afterwards... But that should probably be handled in a separate issue.

[davidtwco]

Now the question is whether the rc could automatically be dropped before the await call if it's not being used afterwards... But that should probably be handled in a separate issue.

There was discussion about that in Zulip which might be of interest to you.

I wonder if there is something we could do in this diagnostic when the destructor is the reason that the type lives across the await point to help explain this.

[nikomatsakis]

@davidtwco I definitely think that destructors are worth trying to "special case" -- we do some similar things in the NLL logic, as I recall.