Awesome

async-csp

Communicating sequential processes for use with ES2016's async/await syntax.

Here's GoLang's ping/pong example in async-csp flavor:

import Channel from 'async-csp'

async function sleep(duration) {
    return new Promise(resolve => setTimeout(resolve, duration))
}

async function player(name, table) {
    while (true) {
        let ball = await table.take();
        if (ball === Channel.DONE) {
            console.log(`${name}: table's gone!`);
            break;
        }
        ball.hits++;
        console.log(`${name}! Hits: ${ball.hits}`);
        await sleep(100);
        await table.put(ball);
    }
}

async function pingPong() {
    console.log('Opening ping-pong channel!');
    let table = new Channel();

    player('ping', table);
    player('pong', table);

    console.log('Serving ball...');
    let ball = {hits: 0};
    await table.put(ball);
    await sleep(1000);

    console.log('Closing ping-pong channel...');
    table.close();

    await table.done();
    console.log('Channel is fully closed!');
    console.log(`Ball was hit ${ball.hits} times!`);
}

pingPong()

Sometimes the output of this example is

Opening ping-pong channel!
Serving ball...
ping! Hits: 1
pong! Hits: 2
ping! Hits: 3
pong! Hits: 4
ping! Hits: 5
pong! Hits: 6
ping! Hits: 7
pong! Hits: 8
ping! Hits: 9
Closing ping-pong channel...
pong: table's gone!
Channel is fully closed!
Ball was hit 9 times!
ping: table's gone!

and sometimes it's

Opening ping-pong channel!
Serving ball...
ping! Hits: 1
pong! Hits: 2
ping! Hits: 3
pong! Hits: 4
ping! Hits: 5
pong! Hits: 6
ping! Hits: 7
pong! Hits: 8
ping! Hits: 9
pong! Hits: 10
Closing ping-pong channel...
ping: table's gone!
Channel is fully closed!
Ball was hit 10 times!
pong: table's gone!

Sometimes the ball is hit 9 times, and sometimes 10! This is due to the nature of asynchronicity which is nicely depicted in this example.

Installation

npm install async-csp

Default Task

• Install node.js
• Clone the async-csp project
• Run npm install
• Run gulp
  • Executes tests
  • Cleans dist
  • Lints source
  • Builds source
  • Watches source and tests

Examples

Examples can be found here. To run any example, make sure the default task has been successfully run once (or at least npm install), then run node index.js from the root folder of the example.

Usage

Note: All of the code pieces below are assumed to be executed from an async context, so await is available at the base level. To read more about these methods, see this proposal for async/await in ES7.

Data Flow

A Channel is a container which makes use of Promises to handle the incoming and outgoing flow of data.

To put a value on a Channel use Channel#put(), and to take a value from the channel use Channel#take().

By default, the promise returned from Channel#put() will not resolve until its value is taken from the channel, and the promise returned from Channel#take() will not resolve until a value can be taken from the channel.

import Channel from 'async-csp';

let channel = new Channel();

async function puts(ch) {
    await ch.put(1); // resolves when the first ch.take() is executed
    await ch.put(2); // resolves when the second ch.take() is executed
    await ch.put(3); // resolves when the third ch.take() is executed
}

async function takes(ch) {
    console.log(await ch.take()); // resolves to 1, from the first ch.put()
    console.log(await ch.take()); // resolves to 2, from the second ch.put()
    console.log(await ch.take()); // resolves to 3, from the third ch.put()
}

puts(channel);
takes(channel);

Buffering

A Channel can be created with a buffer for receiving puts. Essentially, this means a put can resolve while space is available on the buffer, even if no take is waiting to receive a value. As soon as the buffer becomes full, put will begin blocking again until a take clears a space from the buffer.

To create a Channel with a buffer, pass in a Number as the first argument to the constructor.

import Channel, { timeout } from 'async-csp';

let channel = new Channel(2); // buffer size of 2

async function puts(ch) {
    await ch.put(1); //=> this can resolve immediately, taking one space on the buffer
    console.log('after put 1'); // fires immediately
    await ch.put(2); //=> this can also resolve immediately, taking the second space on the buffer
    console.log('after put 2'); // also fires immediately
    await ch.put(3); //=> buffer is full! this will block until another process takes a value from the Channel
    console.log('after put 3'); // fires after the unblock!
}

async function takes(ch) {
    console.log(await ch.take()); //=> resolves to 1, clears a space on the buffer and allows the blocked ch.put(3) to also resolve
    console.log(await ch.take()); //=> resolves to 2
    console.log(await ch.take()); //=> resolves to 3
}

// execute the puts right away
puts(channel);

// use a helper method to wait for 1 second
// to help show the effects of blocking
await timeout(1000);

// after 1 second, start executing takes
takes(channel);

Non-blocking puts

A common use for a Channel requires data to be input from a non async context, or without waiting for the put to resolve.

In this scenario, do not await the result of Channel#put().

let ch = new Channel();

// non-blocking puts, don't use `await`
ch.put(1);
ch.put(2);
ch.put(3);

console.log(await ch.take()); //=> 1
console.log(await ch.take()); //=> 2
console.log(await ch.take()); //=> 3

Transforming

When constructing a Channel, you can pass in a callback to transform values as they are taken.

let ch = new Channel(x => x * 2);

ch.put(1);
ch.put(2);
ch.put(3);

console.log(await ch.take()); //=> 2
console.log(await ch.take()); //=> 4
console.log(await ch.take()); //=> 6

If values should be dropped from the Channel, simply return undefined from the transform callback.

let ch = new Channel(x => {
    if (x > 2)
        return x;
});

ch.put(1);
ch.put(1);
ch.put(3);
ch.put(4);

console.log(await ch.take()); //=> 3
console.log(await ch.take()); //=> 4

If a transform needs to expand a single value into multiple values, use the push parameter with the transform.

Note that when using this callback style, all values must be sent through push. Any value returned from the transform callback will be ignored when the provided transformer has more than one parameter defined.

let ch = new Channel((x, push) => {
    push(x);
    push(x + 1);
});

ch.put(1);
ch.put(3);

console.log(await ch.take()); //=> 1
console.log(await ch.take()); //=> 2
console.log(await ch.take()); //=> 3
console.log(await ch.take()); //=> 4

If the transform needs to work asynchronously, there are a few ways to accomplish this.

The first is to use an async callback.

let ch = new Channel(async x => {
    await timeout(100);
    return x;
});

ch.put(1);
ch.put(2);

console.log(await ch.take()); //=> 1
console.log(await ch.take()); //=> 2

The second way to use an asynchronous transform is by passing in an async callback with a parameter length of 2. Similar to the non-async callback with a parameter length of 2, all values must be sent through push, and returned values will be ignored.

let ch = new Channel(async(x, push) => {
    await timeout(100);
    push(x);
    await timeout(100);
    push(x + 1);
});

ch.put(1);
ch.put(3);

console.log(await ch.take()); //=> 1
console.log(await ch.take()); //=> 2
console.log(await ch.take()); //=> 3
console.log(await ch.take()); //=> 4

The final way to use an asynchronous transform is with a three-parameter callback. To signify that the transform has completed, execute the third argument.

let ch = new Channel((x, push, done) => {
    push(x);
    setTimeout(() => {
        push(x + 1);
        done();
    }, 100);
});

ch.put(1);
ch.put(3);

console.log(await ch.take()); //=> 1
console.log(await ch.take()); //=> 2
console.log(await ch.take()); //=> 3
console.log(await ch.take()); //=> 4

One final note: Using a transform does not prevent you from simultaneously using a buffer. To use a transform with a buffered Channel, pass in the buffer size as the first argument, and the transform as the second.

let ch = new Channel(2, x => x + 1);

// note that puts will be resolved immediately, since we have space on the buffer
await ch.put(1);
await ch.put(3);

console.log(await ch.take()); //=> 2
console.log(await ch.take()); //=> 4

Channel#pipe()

Similarly to Streams, Channels can be piped from one to another.

let ch1 = new Channel();
let ch2 = new Channel();
let ch3 = new Channel();

ch1.pipe(ch2).pipe(ch3);
/*
    +---+
    |ch1|
    +---+
      |
      V
    +---+
    |ch2|
    +---+
      |
      V
    +---+
    |ch3|
    +---+
*/

ch1.put(1);
ch1.put(2);
ch1.put(3);

console.log(await ch3.take()); //=>  1
console.log(await ch3.take()); //=>  2
console.log(await ch3.take()); //=>  3

A Channel can be piped to multiple destinations. In this case, downstream Channels will receive every value from upstream.

let ch1 = new Channel();
let ch2 = new Channel();
let ch3 = new Channel();

ch1.pipe(ch2, ch3); // or `ch1.pipe(ch2); ch1.pipe(ch3);`
/*
        +---+
      +-|ch1|-+
      | +---+ |
      |       |
      V       V
    +---+   +---+
    |ch2|   |ch3|
    +---+   +---+
*/

ch1.put(1);
ch1.put(2);
ch1.put(3);

// 1 is taken from ch1 and put on ch2 and ch2

console.log(await ch2.take()); //=>  1
console.log(await ch3.take()); //=>  1

// 2 is taken from ch1 and put on ch2 and ch3

console.log(await ch2.take()); //=>  2
console.log(await ch3.take()); //=>  2

// 3 is taken from ch1 and put on ch2 and ch3

console.log(await ch2.take()); //=>  3
console.log(await ch3.take()); //=>  3

Typically, it is recommended to pass a Channel into pipe, but if you need a shortcut for creating a Channel from a transform callback, you may also pass in a callback directly.

let ch1 = new Channel(x => x + 1);

// note that this returns a reference to the last `Channel` from `#pipe()`
let ch3 = ch1.pipe(x => x + 2).pipe(x => x + 3); 

ch1.put(1);
ch1.put(2);
ch1.put(3);

console.log(await ch3.take()); //=> 7
console.log(await ch3.take()); //=> 8
console.log(await ch3.take()); //=> 9

Also take note that if one downstream Channel is blocked from a currently unresolved Channel#put() (buffered or non-buffered), then the entire pipe will be blocked. In the example above, an attempt to take all 3 values from ch2 before taking any values from ch3 would have resulted in deadlock.

Finally, any piped Channel will also execute transforms.

let ch1 = new Channel(x => x + 2);
let ch2 = new Channel(x => x.toString());
let ch3 = new Channel(x => ({ x: x }));

ch1.pipe(ch2).pipe(ch3);

ch1.put(1);
ch1.put(2);
ch1.put(3);

console.log(await ch3.take()); //=> { x: '3' }
console.log(await ch3.take()); //=> { x: '4' }
console.log(await ch3.take()); //=> { x: '5' }

Channel.pipeline()

Channel.pipeline() is a helper method for creating piped channels from any number of callbacks or Channels. Inputs can be provided either as separate arguments, or contained in an array as the first argument. If an input is provided as a callback, it will be turned into a Channel using that callback as the transform.

Channel.pipeline() will return an array containing the first and the last Channel in the pipeline.

let [ ch1, ch3 ] = Channel.pipeline(
    x => x + 2,
    new Channel(x => x.toString()),
    async x => ({ x })
);

ch1.put(1);
ch1.put(2);
ch1.put(3);

console.log(await ch3.take()); //=> { x: '3' }
console.log(await ch3.take()); //=> { x: '4' }
console.log(await ch3.take()); //=> { x: '5' }

Channel#unpipe()

If a Channel should be taken out of an existing pipe, use Channel#unpipe().

let ch1 = new Channel();
let ch2 = new Channel();
let ch3 = new Channel();

ch1.pipe(ch2).pipe(ch3);

ch1.put(1);
console.log(await ch3.take()); //=> 1

// now take ch2 out of the pipe
ch1.unpipe(ch2);

ch1.put(2);
console.log(await ch1.take()); //=> 2, note that we took from ch1

// note that ch2 is still piping to ch3
ch2.put(3);
console.log(await ch3.take()); //=> 3

Channel#merge()

Channel.merge() is a helper method for piping multiple Channels into a single, new Channel.

let ch1 = new Channel();
let ch2 = new Channel();
let ch3 = ch1.merge(ch2); // or, `ch3 = Channel.merge(ch1, ch2)`

ch1.put(1);
ch2.put(2);

console.log(await ch3.take()); //=> 1
console.log(await ch3.take()); //=> 2

Channel#close()

A Channel has 3 states: open, closed, and ended. An open Channel can be written to, a closed Channel will not accept any new values but may be non-empty, and an ended Channel is both closed and empty.

To signify that a Channel should be done accepting new values, execute Channel#close(). Data can still be taken from the channel after that point, but no more values can be added.

let ch1 = new Channel();

ch1.put(1);
ch1.put(2);

ch1.close();

ch1.put(3); // resolves immediately with value of Channel.DONE

console.log(await ch1.take()); //=> 1
console.log(await ch1.take()); //=> 2
console.log(await ch1.take()); //=> Channel.DONE

If Channels are piped together, and you want the entire pipeline to close when possible, simply pass true as an argument to Channel#close().

let ch1 = new Channel();
let ch2 = new Channel();
ch1.pipe(ch2);

ch1.put(1);
ch1.put(2);

ch1.close(true);

console.log(await ch2.take()); //=> 1
console.log(await ch2.take()); //=> 2
console.log(await ch2.take()); //=> Channel.DONE

Channel#done()

In order to wait for a channel to be ended (closed and empty), await the resolution of done.

let ch = new Channel();

ch.put(1);
ch.put(2);
ch.close();

let arr = [];
(async() => {
    await timeout(1000);
    arr.push(await ch.take());
    await timeout(1000);
    arr.push(await ch.take());
})();

await ch.done(); // will not resolve until the async IIFE takes both values from the channel
console.log(arr); //=> [ 1, 2 ]

Channel#tail()

While manually appending values to a Channel can be accomplished, it often becomes significantly more difficult when items such as pipes and asynchronous transforms are in play.

For simplicity, Channel#tail() is provided as an alternative method for providing values to Channel#take() only after the Channel is closed and all existing Channel#put()s have been resolved.

let ch = new Channel();

ch.put(1);
ch.tail(4);
ch.put(2);
ch.put(3);
ch.close();

console.log(await ch.take()); //=> 1
console.log(await ch.take()); //=> 2
console.log(await ch.take()); //=> 3
console.log(await ch.take()); //=> 4

Note that when a Channel has a transform, any values provided through Channel#tail() will also use that transform.

let ch = new Channel(x => x + 2);

ch.put(1);
ch.tail(4);
ch.put(2);
ch.put(3);
ch.close();

console.log(await ch.take()); //=> 3
console.log(await ch.take()); //=> 4
console.log(await ch.take()); //=> 5
console.log(await ch.take()); //=> 6

Channel#consume()

If you would like to execute a callback as soon as values can be taken from the Channel, you may add a consumer by using Channel#consume().

let ch = new Channel();
ch.consume(x => {
    console.log(x);
});

await ch.put(1);
await ch.put(2);
await ch.put(3);
await ch.put(4);

// console logs
//=> 1
//=> 2
//=> 3
//=> 4

Channel#consume() can also be handled asynchronously, and will not attempt to queue up another Channel#take() until the consumer callback has completed running.

let ch = new Channel();
let arr = [];
ch.consume(async x => {
    await timeout(1000);
    arr.push(x);
    console.log(x);
});

await ch.put(1);
await ch.put(2);
await ch.put(3);
await ch.put(4);

// console logs, once a second
//=> 1
//=> 2
//=> 3
//=> 4

Channel#produce()

Similar to Channel#consume(), Channel#produce() will put returned values onto the Channel as soon as space becomes available.

let ch = new Channel();
let counter = 0;
ch.produce(() => ++counter);

console.log(await ch.take()); //=> 1
console.log(await ch.take()); //=> 2
console.log(await ch.take()); //=> 3
console.log(await ch.take()); //=> 4

As with Channel#consume(), Channel#produce() can also work asynchronously.

let ch = new Channel();
let counter = 0;
ch.produce(async() => {
    await timeout(1000);
    return ++counter;
});

console.log(await ch.take()); //=> 1, after 1 second
console.log(await ch.take()); //=> 2, after 2 seconds
console.log(await ch.take()); //=> 3, after 3 seconds
console.log(await ch.take()); //=> 4, after 4 seconds

Channel.from()

If you have an iterable item which you would like to convert into a Channel, use Channel.from() to construct a Channel from that iterable.

let arr = [ 1, 2, 3 ];
let ch = Channel.from(arr);

console.log(await ch.take()); //=> 1
console.log(await ch.take()); //=> 2
console.log(await ch.take()); //=> 3

Note that in this case, a buffer is created with the size of the iterable, all values are placed directly onto the buffer, and the Channel is marked as closed, which will include any attached downstream pipes.

If the channel or any downstream pipes should remain open to continue receiving puts, pass in a true as the second argument.

let arr = [ 1, 2, 3 ];
let ch = Channel.from(arr, true);
ch.put(4);
ch.close();

console.log(await ch.take()); //=> 1
console.log(await ch.take()); //=> 2
console.log(await ch.take()); //=> 3
console.log(await ch.take()); //=> 4

License

All code released under the MIT license.