1 unstable release

0.1.0 Dec 20, 2021

#9 in #suspend


Used in suspend_fn

Apache-2.0

16KB
456 lines

suspend fn

Disclaimer: this was mostly made as a proof of concept for the proposal below. I haven't tested if there is a performance cost to this macro.

This crate provides a proc-macro that removes the need for the await keyword. For example:

#[suspend_fn]
fn visit_rustlang() -> Result<(), reqwest::Error> {
    let response = reqwest::get("https://www.rust-lang.org")?;
    let text = response.text()?;
    println!("{}", text);
    text.len(); // sync functions work just fine!
    Ok(())
}

the above code is functionally equivalent to:

async fn visit_rustlang() -> Result<(), reqwest::Error> {
    let response = reqwest::get("https://www.rust-lang.org").await?;
    let text = response.text().await?;
    println!("{}", text);
    text.len(); 
    Ok(())
}

suspend blocks

You can also use the suspend! and suspend_move! macros similarlly to async and async move blocks. Note that these might prompt style warnings, so I recommend placing #![allow(unused_parens)] in the crate root.

#![allow(unused_parens)]
suspend! { 
    let response = reqwest::get("https://www.rust-lang.org")?;
    let text = response.text()?;
    println!("{}", text);
    text.len(); 
}

limitations

Currently, the macro does not work inside other macros. For example:

println!("{}", async_fn());

the above code will raise the following error message:

| println!("{}", async_fn()); 
|                ^^^^^^^^^^  `impl Future` cannot be formatted with the default formatter

for such cases you may simply use .await in the macro yourself.

motivation

In a recent blog post discussing the async cacellation problem, it was argued that async destructors would lead to inconsistencies from a langauge design perspective. This is because the compiler would introduce .await calls for async destructors, making some .await calls implicit and others explicit.

As a way to resolve this conflict, it was proposed the removal of the .await syntax entirely. Although I think this would me great from an ergonomics perspective, I would not like to see such a big breaking change. Alternatively, I propose the addition of a new keyword, analogous to async, for the moment let's use Kotlin's suspend.

Then we could have:

suspend fn function() {
    /* 
        implicit await with implicit calls to async destructors
    */
}
async fn function() {
    /* 
        explicit await with explicit calls to async destructors
    */
}

async {
    /* 
        explicit await with explicit calls to async destructors
    */
}

suspend {
    /* 
        implicit await with implicit calls to async destructors
    */
}

Dependencies

~1.5MB
~37K SLoC