Rust's Future by Example

Update 2019-01-14: This document is now out of date. I recommend learning about Rust futures another way.

This document is intended to let readers start working with Rust’s Future quickly. Some other useful reading includes:

• The official Tokio documentation
• Zero-cost futures in Rust
• Tokio Internals: Understanding Rust’s asynchronous I/O framework from the bottom up

Future

The Future trait from futures represents an asynchronous operation that can fail or succeed, producing a value either way. It is like an async version of Result. This document assumes that the reader is familiar with Result, which is covered in the second edition of The Rust Programming Language.

One of the most common questions about Future seems to be, “how do I get the value out of it?” The easiest way to do this is to call the wait method. This runs the Future in the current thread, blocking all other work until it is finished.

This is not frequently the best way to run a Future, because no other work can happen until the Future completes, which completely defeats the point of using asynchronous programming in the first place. However, it can be useful in unit tests, when debugging, or at the top level of a simple application.

See the section on reactors for better ways to run a Future.

extern crate futures;
extern crate future_by_example;

fn main() {
    use futures::Future;
    use future_by_example::new_example_future;

    let future = new_example_future();

    let expected = Ok(2);
    assert_eq!(future.wait(), expected);
}

A Future can be modified using many functions analogous to those of Result, such as map, map_err, and then. Here’s map:

extern crate futures;
extern crate future_by_example;

fn main() {
    use futures::Future;
    use future_by_example::new_example_future;

    let future = new_example_future();
    let mapped = future.map(|i| i * 3);

    let expected = Ok(6);
    assert_eq!(mapped.wait(), expected);
}

Like a Result, two Futures can be combined using and_then and or_else:

extern crate futures;
extern crate future_by_example;

fn main() {
    use futures::Future;
    use future_by_example::*;

    let good = new_example_future();
    let bad = new_example_future_err();
    let both = good.and_then(|good| bad);

    let expected = Err(ExampleFutureError::Oops);
    assert_eq!(both.wait(), expected);
}

Future also has a lot of functions that have no analog in Result. Because we’re talking about aynchronous programming, now we have to choose whether we want to run two independent operations one after the other (in sequence), or at the same time (in parallel).

For example, to get the results of two independent Futures, we could use and_then to run them in sequence. However, that strategy is silly, because we are only making progress on one Future at a time. Why not run both at the same time?

Future::join creates a new Future that contains the results of two other Futures. Importantly, both of the input Futures can make progress at the same time. The new Future completes only when both input Futures complete. There’s also join3, join4 and join5 for joining larger numbers of Futures.

extern crate futures;
extern crate future_by_example;

fn main() {
    use futures::Future;
    use futures::future::ok;
    use future_by_example::new_example_future;

    let future1 = new_example_future();
    let future2 = new_example_future();

    let joined = future1.join(future2);
    let (value1, value2) = joined.wait().unwrap();
    assert_eq!(value1, value2);
}

Whereas join completes when both Futures are complete, select returns whichever of two Futures completes first. This is useful for implementing timeouts, among other things. select2 is like select except that the two Futures can have different value types.

Creating a Future

Many libraries return Futures for asynchronous operations such as network calls. Sometimes you may want to create your own Future. Implementing a Future from scratch is difficult, but there are other ways to create futures.

You can easily create a Future from a value that is already available using the ok function. There are similiar err and result methods.

extern crate futures;

fn main() {
    use futures::Future;
    use futures::future::ok;

    // Here I specify the type of the error as (); otherwise the compiler can't
    // infer it
    let future = ok::<_, ()>(String::from("hello"));
    assert_eq!(Ok(String::from("hello")), future.wait());
}

Futures and types

Working with Futures tends to produce complex types. For example, the full type of the expression below is actually:

futures::Map<
  futures::Map<
    futures::Join<
      futures::FutureResult<u64, ()>,
      futures::FutureResult<u64, ()>
    >,
    [closure@src/lib.rs:...]>,
  [closure@src/lib.rs:...]
>

That is, for every transformation, we add another layer to the type of our Future! This can sometimes be confusing. In particular, it can be challenging to identify ways to write out the types that aren’t brittle or verbose.

In order to help the Rust compiler do type inference, below we have specify the type of expected. It’s much terser than writing the full type out, and adding another operation won’t break compilation.

extern crate futures;

fn main() {
    use futures::future::ok;
    use futures::Future;

    let expected: Result<u64, ()> = Ok(6);
    assert_eq!(
        ok(5).join(ok(7)).map(|(x, y)| x + y).map(|z| z / 2).wait(),
        expected
    )
}

Alternatively, we can make use of _ to let the Rust compiler infer types for us.

extern crate futures;

fn main() {
    use futures::future::ok;
    use futures::Future;
    use futures::Map;

    let expected: Result<_, ()> = Ok(6);
    let twelve: Map<_, _> = ok(5).join(ok(7)).map(|(x, y)| x + y);
    assert_eq!(twelve.map(|z| z / 2).wait(), expected)
}

Rust requires that all types in function signatures are specified.

One way to achieve this for functions that return Futures is to specify the full return type in the function signature. However, specifying the exact type can be verbose, brittle, and difficult.

It would be nice to be able to define a function like this:

fn make_twelve() -> Future<Item=u64, Error=()> {
    unimplemented!()
}

However, the compiler doesn’t like that:

error[E0277]: the trait bound `futures::Future<Item=u64, Error=()>: std::marker::Sized` is not satisfied
   --> src/lib.rs:119:13
    |
119 |         let twelve = make_twelve();
    |             ^^^^^^ `futures::Future<Item=u64, Error=()>` does not have a constant size known at compile-time
    |
    = help: the trait `std::marker::Sized` is not implemented for `futures::Future<Item=u64, Error=()>`
    = note: all local variables must have a statically known size

This can be solved by wrapping the return type in a Box. One day, this will be solved in a more elegant way with the currently unstable impl Trait functionality.

extern crate futures;

fn main() {
    use futures::Future;
    use futures::future::ok;

    fn make_twelve() -> Box<Future<Item=u64, Error=()>> {

        ok(5).join(ok(7)).map(|(x, y)| x + y).boxed()
    }

    let twelve = make_twelve();
    assert_eq!(twelve.map(|z| z / 2).wait(), Ok(6))
}

Unlike functions, closures do not require all types in their signatures to be explicitly defined, so they don’t need to be wrapped in a Box.

extern crate futures;

fn main() {
    use futures::Future;

    let make_twelve = || {
        use futures::future::ok;

        // We don't need to put our `Future` inside of a `Box` here.
        ok(5).join(ok(7)).map(|(x, y)| x + y)
    };

    let expected: Result<u64, ()> = Ok(6);
    let twelve = make_twelve();
    assert_eq!(twelve.map(|z| z / 2).wait(), expected)
}

A more powerful way to run Futures

Composing a bunch of Futures into a single Future and calling wait on it is a simple and easy method as long as you only need to run a single Future at a time. However, if you only need to run a single Future at a time, perhaps you don’t need the futures crate in the first place! The futures crate promises to efficiently juggle many concurrent tasks, so let’s see how that might work.

The tokio-core crate has a struct called Core which can run multiple Futures concurrently. Core::run runs a Future, returning its value. Unlike Future::wait, though, it allows the Core to make progress on executing other Future objects while run running. The Future in Core::run is the main event loop, and it may request that new Futures be run by calling Handle::spawn. Note that the Futures run by spawn don’t get to return a value; they exist only to perform side effects.

extern crate futures;
extern crate tokio_core;

fn main() {
    use tokio_core::reactor::Core;
    use futures::future::lazy;

    let mut core = Core::new().unwrap();
    let handle = core.handle();
    let future = lazy(|| {
        handle.spawn(lazy(|| {
            Ok(()) // Ok(()) implements FromResult
        }));
        Ok(2)
    });
    let expected: Result<_, ()> = Ok(2usize);
    assert_eq!(core.run(future), expected);
}

License: MIT/Apache-2.0