Promise-oriented coroutines for node.js


Keywords
fibers, coroutine, async, promise
License
MIT
Install
npm install f-promise@2.1.0

Documentation

f-promise

Promise-oriented coroutines for node.js.

npm install f-promise

API

The f-promise API consists in 2 calls: wait and run.

  • result = wait(promise): waits on a promise and returns its result (or throws if the promise is rejected).
  • promise = run(fn): runs a function as a coroutine and returns a promise for the function's result.

Constraint: wait may only be called from a coroutine (a function which is executed by run, directly or indirectly, through one of its callers).

Simple example

import { wait, run } from 'f-promise';
import * as fs from 'mz/fs';
import { join } from 'path';

function diskUsage(dir) {
    return wait(fs.readdir(dir)).reduce((size, name) => {
        const sub = join(dir, name);
        const stat = wait(fs.stat(sub));
        if (stat.isDirectory()) return size + diskUsage(sub);
        else if (stat.isFile()) return size + stat.size;
        else return size;
    }, 0);
}

function printDiskUsage(dir) {
    console.log(`${dir}: ${diskUsage(dir)}`);
}

run(() => printDiskUsage(process.cwd()))
    .then(() => {}, err => { console.error(err); });

Note: this is not a very efficient implementation because the logic is completely serialized.

Why f-promise?

To understand the benefits of f-promise, let us compare the example above with the ES7 async/await equivalent:

import * as fs from 'mz/fs';
import { join } from 'path';

async function diskUsage(dir) {
    var size = 0;
    for (var name of await fs.readdir(dir)) {
        const sub = join(dir, name);
        const stat = await fs.stat(sub);
        if (stat.isDirectory()) size += await diskUsage(sub);
        else if (stat.isFile()) size += stat.size;
    }
    return size;
}

async function printDiskUsage(dir) {
    console.log(`${dir}: ${await diskUsage(dir)}`);
}

printDiskUsage(process.cwd())
    .then(() => {}, err => { console.error(err); });

Two observations:

  • Async is contagious: printDiskUsage must be marked as async because it needs to await on diskUsage. This is not dramatic in this simple example but in a large code base this translates into a proliferation of async/await keywords throughout the code.
  • ES7 async/await does not play well with array methods (forEach, map, reduce, ...) because you cannot use await inside the callbacks of these methods. You have to write the loop differently, with for ... of ... or Promise.all.

f-promise solves these problems:

  • Functions that wait on async operations are not marked with async; they are normal JavaScript functions. async/await keywords don't invade the code.
  • wait plays well with array methods, and with other APIs that expect synchronous callbacks.

Coroutines have other advantages, like providing complete meaningful stacktraces without any overhead.

TypeScript support

TypeScript is fully supported.

Callbacks support

You can also use f-promise with callback APIs. So you don't absolutely need wrappers like mz/fs, you can directly call node's fs API:

import { wait } from 'f-promise';

// promise style
import * as mzfs from 'mz/fs';
const readdir = path => wait(mzfs.readdir(path));

// callback style
import * as fs from 'fs';
const readdir = path => wait(cb => fs.readdir(path, cb));

Control Flow utilities

These goodies solve some common problems and offer an easy upgrade path from streamline.js (which bundled a similar API).

funnel

  • fun = fpromise.funnel(max)
    limits the number of concurrent executions of a given code block.

The funnel function is typically used with the following pattern:

import { funnel } from 'f-promise';

// somewhere
var myFunnel = funnel(10); // create a funnel that only allows 10 concurrent executions.

// elsewhere
myFunnel(() => { /* code with at most 10 concurrent executions */ });

The funnel function can also be used to implement critical sections. Just set funnel's max parameter to 1.

If max is set to 0, a default number of parallel executions is allowed. This default number can be read and set via funnel.defaultSize.
If max is negative, the funnel does not limit the level of parallelism.

The funnel can be closed with fun.close().
When a funnel is closed, the operations that are still in the funnel will continue but their callbacks won't be called, and no other operation will enter the funnel.

handshake and queue

  • hs = fpromise.handshake()
    allocates a simple semaphore that can be used to do simple handshakes between two tasks.
    The returned handshake object has two methods:
    hs.wait(): waits until hs is notified.
    hs.notify(): notifies hs (without waiting for an acknowledgement) Note: wait calls are not queued. An exception is thrown if wait is called while another wait is pending.
  • q = fpromise.queue(options)
    allocates a queue which may be used to send data asynchronously between two tasks.
    The max option can be set to control the maximum queue length.
    When max has been reached q.put(data) discards data and returns false. The returned queue has the following methods:
    data = q.read(): dequeues an item from the queue. Waits if no element is available.
    q.write(data): queues an item. Waits if the queue is full.
    ok = q.put(data): queues an item synchronously. Returns true if the queue accepted it, false otherwise.
    q.end(): ends the queue. This is the synchronous equivalent of q.write(undefined)
    data = q.peek(): returns the first item, without dequeuing it. Returns undefined if the queue is empty.
    array = q.contents(): returns a copy of the queue's contents.
    q.adjust(fn[, thisObj]): adjusts the contents of the queue by calling newContents = fn(oldContents).
    q.length: number of items currently in the queue.

CLS (Continuation Local Storage)

  • cx = fpromise.context()
    returns the current context.

  • fn = fpromise.withContext(fn, cx)
    wraps a function so that it executes with context cx (or a wrapper around current context if cx is falsy). The previous context will be restored when the function returns (or throws).
    returns the wrapped function.

Miscellaneous

  • results = fpromise.map(collection, fn)
    creates as many coroutines with fn as items in collection and wait for them to finish to return result array.

  • fpromise.sleep(ms)
    suspends current coroutine for ms milliseconds.

  • ok = fpromise.canWait()
    returns whether wait calls are allowed (whether we are called from a run).

  • wrapped = fpromise.eventHandler(handler)
    wraps handler so that it can call wait.
    the wrapped handler will execute on the current fiber if canWait() is true. otherwise it will be run on a new fiber (without waiting for its completion)

Error stack traces

Three policies available for error stack trace handling:

  • fast: stack traces are not changed. Call history might be difficult to read; cost less.
  • whole: stack traces due to async tasks errors in wait() are concatenate with the current coroutine stack. This allow to have a complete history call (including f-promise traces).
  • default: stack traces are like whole policy, but clean up to remove f-promise noise.

The policy can be set with FPROMISE_STACK_TRACES environment variable. Any value other than fast and whole are consider as default policy.

Related projects

License

MIT.

Credits

f-promise is just a thin layer. All the hard work is done by the fibers library.

Gotchas

The absence of async/await markers in code that calls asynchronous APIs is unusual in JavaScript (and considered harmful by some). But this is the norm in other languages. Basically f-promise enables goroutines in JavaScript.