deferred

Modular and fast Promises implementation

Deferred

Modular and fast Promises implementation for JavaScript

Implementation originally inspired by Kris Kowal's Q

Deferred is complete, one of the fastest and natural promise implementation in JavaScript, with Deferred you can write clear maintainable code that takes maximum out of asynchronicity, in fact due to multi-dimensional nature of promises (chaining and nesting) you're forced to program declaratively.

With Deferred you also can: Process collections of deferred calls. Handle Node.js asynchronous functions. Limit concurrency of scheduled tasks. Emit progress events or stream results partially on the go.

In the end you may debug your flow by tracking unresolved promises or gathering usage statistics.

For good insight into promise/deferred concept and in general asynchronous programming see also slides from meetjs summit presentation: Asynchronous JavaScript

If you need help with deferred, please ask on dedicated mailing list: deferred-js@googlegroups.com

Comparision with callback style

Example of JavaScript files concatenation:

Plain Node.js, callbacks approach

var fs = require('fs');
 
var readdir = fs.readdir;
var readFile = fs.readFile;
var writeFile = fs.writeFile;
 
// Read all filenames in given path 
readdir(__dirname, function (err, files) {
  var result, waiting;
  if (err) {
    // if we're unable to get file listing throw error 
    throw err;
  }
 
  // Filter *.js files and generated lib.js 
  files = files.filter(function (file) {
    return (file.slice(-3) === '.js') && (file !== 'lib.js');
  });
 
  // Read content of each file 
  waiting = 0;
  result = [];
  files.forEach(function (file, index) {
    ++waiting;
    readFile(file, function (err, content) {
      if (err) {
        // We were not able to read file content, throw error 
        throw err;
      }
      result[index] = content;
 
      if (!--waiting) {
        // Got content of all files 
        // Concatenate into one string and write into lib.js 
        writeFile(__dirname + '/lib.js', result.join("\n"), function (err) {
          if (err) {
            // We cannot write lib.js file, throw error 
            throw err;
          }
        });
      }
    });
  });
});

Implementation with promises:

var promisify = require('deferred').promisify;
var fs = require('fs');
 
// Convert node.js async functions, into ones that return a promise 
var readdir = promisify(fs.readdir);
var readFile = promisify(fs.readFile, 1); // Restrict arity to 1 + callback 
var writeFile = promisify(fs.writeFile);
 
writeFile(__dirname + '/lib.js',
  // Read all filenames in given path 
  readdir(__dirname)
 
  // Filter *.js files and generated lib.js 
  .invoke('filter', function (file) {
    return (file.slice(-3) === '.js') && (file !== 'lib.js');
  })
 
  // Read content of all files 
  .map(readFile)
 
  // Concatenate files content into one string 
  .invoke('join', '\n')
 
).done(); // If there was any error on the way throw it

Examples

See examples folder for a demonstration of promises usage in some other real world cases.

Installation

NPM

In your project path:

$ npm install deferred

Browser

Browser bundle can be easily created with help of modules-webmake. Assuming that you have latest Node.js and Git installed, following will work in command shell of any system (Linux/MacOS/Windows):

$ npm install -g webmake
$ git clone git://github.com/medikoo/deferred.git
$ cd deferred
$ npm install
$ cd ..
$ webmake --name=deferred deferred/index.js deferred.js

Last command bundles deferred with all it's functionalities, but you may need just a subset, you can have that by addressing specific modules directly, e.g. with following you will build just core functionality with map extension:

$ webmake --name=deferred --include=deferred/ext/promise/map.js deferred/deferred.js deferred.js

If you work with AMD modules, use amd option, so generated bundle is one:

$ webmake --amd deferred/index.js deferred.js

Mind that deferred relies on some ECMAScript5 features, so for older browsers you need to load as well es5-shim

Deferred/Promise concept

Deferred

For work that doesn't return immediately (asynchronous) you may create deferred object. Deferred holds both resolve and promise objects. Observers interested in value are attached to promise object, with resolve we resolve promise with an actual value. In common usage promise is returned to the world and resolve is kept internally

Let's create delay function decorator:

var deferred = require('deferred');
 
var delay = function (fn, timeout) {
  return function () {
    var def = deferred(), self = this, args = arguments;
 
    setTimeout(function () {
      var value;
      try {
        value = fn.apply(self, args));
      } catch (e) {
        def.reject(e);
        return;
      }
      def.resolve(value);
    }, timeout);
 
    return def.promise;
  };
};
 
var delayedAdd = delay(function (a, b) {
  return a + b;
}, 100);
 
var resultPromise = delayedAdd(2, 3);
 
console.log(deferred.isPromise(resultPromise)); // true 
 
resultPromise(function (value) {
  // Invoked after 100 milliseconds 
  console.log(value); // 5 
});

Promise

Promise is an object that represents eventual value which may already be available or is expected to be available in a future. Promise may succeed (fulfillment) or fail (rejection). Promise can be resolved only once.
In deferred (and most of the other promise implementations) you may listen for the value by passing observers to then function:

promise.then(onsuccess, onfail);

In deferred promise is really a then function, so you may use promise function directly:

promise === promise.then; // true 
promise(onsuccess, onfail);

If you want to keep clear visible distinction between promises and other object I encourage you to always use promise.then notation.

Both callbacks onsuccess and onfail are optional. They will be called only once and only either onsuccess or onfail will be called.

Chaining

Promises by nature can be chained. promise function returns another promise which is resolved with a value returned by a callback function:

delayedAdd(2, 3)(function (result) {
  return result * result
})(function (result) {
  console.log(result); // 25 
});

It's not just functions that promise function can take, it can be other promises or any other JavaScript value (with exception of null or undefined which will be treated as no value). Going that way you may override result of a promise chain with specific value.

Nesting

Promises can be nested. If a promise resolves with another promise, it's not really resolved. It's resolved only when final promise is resolved with a real value:

var def = deferred();
def.resolve(delayedAdd(2, 3)); // Resolve promise with another promise 
def.promise(function (result) {
  console.log(5); // 5; 
});

Error handling

Errors in promises are handled with separate control flow, that's one of the reasons why code written with promises is more readable and maintainable than when using callbacks approach.

A promise resolved with an error (rejected), propagates its error to all promises that depend on this promise (e.g. promises initiated by adding observers).
If observer function crashes with error or returns error, its promise is rejected with the error.

To handle error, pass dedicated callback as second argument to promise function:

delayedAdd(2, 3)(function (result) {
  throw new Error('Error!')
})(function () {
  // never called 
}, function (e) {
  // handle error; 
});

Ending chain

To expose the errors that are not handled, end promise chain with .done(), then error that broke the chain will be thrown:

delayedAdd(2, 3)
(function (result) {
  throw new Error('Error!')
})(function (result) {
  // never executed 
})
.done(); // throws error!

It's important to end your promise chains with done otherwise eventual ignored errors will not be exposed.

Signature of done function is same as for then (or promise itself)

done is aliased with end function, however end will be removed with introduction of v0.7 release.

promise(function (value) {
  // process 
}).done(function (result) {
  // process result 
}, function (err) {
  // handle error 
});

And as with then either callback can be provided. If callback for error was omitted, eventual error will be thrown.

Creating resolved promises

You may create initially resolved promises.

var promise = deferred(1);
 
promise(function (result) {
  console.log(result); // 1; 
});

Working with asynchronous functions

promisify(fn[, length])

There is a known convention (coined by Node.js) for working with asynchronous calls. An asynchronous function receives a callback argument which handles both eventual error and expected value:

var fs = require('fs');
 
fs.readFile(__filename, 'utf-8', function (err, content) {
  if (err) {
    // handle error; 
    return;
  }
  // process content 
});

It's not convenient to work with both promises and callback style functions. When you decide to build your flow with promises don't mix both concepts, just promisify asynchronous functions so they return promises instead.

var deferred = require('deferred')
  , fs = require('fs')
 
  , readFile = deferred.promisify(fs.readFile);
 
readFile(__filename, 'utf-8')(function (content) {
  // process content 
}, function (err) {
  // handle error 
});

promisify accepts also second argument, through which we may specify length of arguments that function takes (not counting callback argument), it may be handy if there's a chance that unexpected arguments will be passed to function (e.g. Array's forEach or map)

promisify also takes care of input arguments. It makes sure that all arguments that are to be passed to asynchronous function are first resolved.

callAsync(fn, context, ...args)

If for some reason you need to turn asynchronous functions into ones that return promises, inline in algorithm, then callAsync is for you.

Still mind that promisify is much better (cleaner) choice if it's possible to prepare reusable wrapper upfront.

var callAsync = require('deferred').callAsync;
 
callAsync(someAsyncFn, context, arg1, arg2).done(function (result) {
  // process result 
});

invokeAsync(obj, fnName | fn, ...args)

If you need to turn asynchronous methods to ones that return promises, and you prefer not to augment its class prototypes, invokeAsync addresses that use case.

var invokeAsync = require('deferred').invokeAsync
 
invokeAsync(db, 'find', 'books', { title: "Some title" }).done(function (book) {
  // process result 
});

Grouping promises

When we're interested in results of more than one promise object we may group them into one promise with deferred function:

deferred(delayedAdd(2, 3), delayedAdd(3, 5), delayedAdd(1, 7))(function (result) {
  console.log(result); // [5, 8, 8] 
});

Processing collections

Map

It's analogous to Array's map, with that difference that it returns promise (of an array) that would be resolved when promises for all items are resolved. Any error that would occur will reject the promise and resolve it with same error.

In following example we take content of each file found in an array:

var readFile = deferred.promisify(fs.readFile);
 
deferred.map(filenames, function (filename) {
  return readFile(filename, 'utf-8');
})(function (result) {
  // result is an array of file's contents 
});

map is also available directly on a promise object, so we may invoke it directly on promise of a collection.

Let's try again previous example but this time instead of relying on already existing filenames, we take list of files from current directory:

var readdir = deferred.promisify(fs.readdir);
var readFile = deferred.promisify(fs.readFile);
 
readdir(__dirname).map(function (filename) {
  return readFile(filename, 'utf-8');
})(function (result) {
  // result is an array of file's contents 
});

This function is available also as an extension on promise object.

See limiting concurrency section for info on how to limit maximum number of concurrent calls in map

Reduce

It's same as Array's reduce with that difference that it calls callback only after previous accumulated value is resolved, this way we may accumulate results of collection of promises or invoke some asynchronous tasks one after another.

deferred.reduce([delayedAdd(2, 3), delayedAdd(3, 5), delayedAdd(1, 7)], function (a, b) {
  return delayedAdd(a, b);
})
(function (result) {
  console.log(result); // 21 
});

This function is available also as an extension on promise object.

Some

Promise aware Array's some. Process collection one after another and stop when first item matches your criteria

deferred.some([filename1, filename2, filename3], function (a) {
  return readFile(filename, 'utf8', function (data) {
    if (data.indexOf('needle')) {
      // Got it! Stop further processing 
      return true;
    }
  });
});

This function is available also as an extension on promise object.

Limiting concurrency

There are cases when we don't want to run too many tasks simultaneously. Like common case in Node.js when we don't want to open too many file descriptors.

Handle that with deferred.gate, it wraps functions that return promises. It doesn't do anything to promise objects, it just limits creation of them by blocking calls to function it wraps.

var fn = deferred.gate(function async() {
  var def = deferred();
  // .. 
  return def.promise;
}, 10);

If there are already 10 concurrent tasks running async function invocation will be postponed into the queue and released when first of the running tasks will finish its job.

Additionally with third argument, we may limit number of postponed calls, so if there's more than n of them rest is discarded. In below example, queue holds maximum 3 postponed calls, rest will be discarded.

var fn = deferred.gate(function async() { .. }, 10, 3);

In following example we'll limit concurrent readFile calls when using deferred.map:

// Open maximum 100 file descriptors at once 
deferred.map(filenames, deferred.gate(function (filename) {
  return readFile(filename, 'utf-8');
}, 100))(function (result) {
  // result is an array of file's contents 
});

Progress and other events

Promise objects are also an event emitters. Deferred implementation is backed by cross-environment event-emitter solution

Simple Ajax file uploader example:

var ajaxFileUploader = function (url, data) {
  var def = deferred();
  var xhr = new XMLHttpRequest();
 
  xhr.open('POST', url, true);
  xhr.onload = def.resolve;
  xhr.onerror = function () {
    def.resolve(new Error("Could not upload files"));
  };
  xhr.upload.onprogress = function (e) {
    def.promise.emit('progress', e);
  };
  xhr.send(data);
  return def.promise;
};
 
var upload = ajaxFileUploader(formData);
upload.on('progress', function () {
  // process progress events 
});
upload.done(function (e) {
  // All files uploaded! 
});

Streaming data partially

Another use case would be to provide obtained data partially on the go (stream like). Imagine recursive directory reader that scans whole file system and provides filenames as it approaches them:

var reader = readdirDeep(rootPath); // reader promise is returned 
reader.on('data', function (someFilenames) {
  // Called many times during scan with obtained names 
});
reader.done(function (allFilenames) {
  // File-system scan finished! 
});

Promise extensions

Promise objects are equipped with some useful extensions. All extension are optional but are loaded by default when deferred is loaded via require('deferred') import. When preparing client-side file (with help of e.g. modules-webmake) you are free to decide, which extensions you want to take (see source of lib/index.js on how to do that)

aside

Third brother of then and done. Has same signature but neither extends chain nor ends it, instead splits it by returning promise on which it was invoked. Useful when we want to return promise, but on a side (in parallel) do something else with obtained value:

var x = deferred({ foo: 'bar' });
var promise = deferred({ foo: 'bar' });
 
var y = promise.aside(function (value) {
  console.log(value === x); // true 
});
console.log(y === promise); // true

catch

Same as then but accepts only onFail callback.

 
var def = deferred(), promise2;
 
promise2 = def.promise.catch(function () {
  return 'Never mind';
});
 
def.reject(new Error("Error"));
 
promise2.done(function (value) {
  console.log(value); // Prints "Never mind" 
});

cb

Convert back to callback style. Useful if you want to process regular callback at the end of promise chain. Simple use case would be regular asynchronous function built internally with promises. cb also makes sure that your callback is not called immediately but in next tick earliest.

With cb we may build hybrid functions that do both, handle asynchronous callback and return promise:

var asyncFunction = function (x, y, callback)  {
  return someAsyncProcessingThatReturnsPromise(x, y).cb(callback);
});

finally

Invokes given callback when promise is either fulfilled or rejected

 
var prepare = function () { ... }
  , cleanup = function () { ... }
 
prepare();
promise = asyncFn();
promise.finally(cleanup);

get

To directly get to object property use get

var promise = deferred({ foo: 'bar' });
 
promise(function (obj) {
  console.log(obj.foo); // 'bar'; 
})
 
promise.get('foo')(function (value) {
  console.log(value); // 'bar' 
});

You can get to nested properties as well:

var promise = deferred({ foo: { bar: 317 });
 
promise(function (obj) {
  console.log(obj.foo.bar); // 317; 
})
 
promise.get('foo', 'bar')(function (value) {
  console.log(value); // 317 
});

invoke & invokeAsync

Schedule function call on returned object

var promise = deferred({ foo: function (arg) { return arg*arg; } });
 
promise.invoke('foo', 3)(function (result) {
  console.log(result); // 9 
});
 
// For asynchronous functions use invokeAsync 
var promise = deferred({ foo: function (arg, callback) {
  setTimeout(function () {
    callback(null, arg*arg);
  }, 100);
} });
 
promise.invokeAsync('foo', 3)(function (result) {
  console.log(result); // 9 
});

spread

If promise expected value is a list that you want to spread into function arguments then use spread

var promise = deferred([2, 3]);
 
promise.spread(function (a, b) {
  console.log(a + b); // 5 
});

Debugging

Monitoring unresolved promises

In properly constructed flow, there should be no promises that are never resolved. If you want to be sure that it's not the case, or you suspect there are such issues, check whether deferred's monitor has something to say

deferred.monitor();

By default monitor will log error for every promise that was not resolved in 5 seconds. You can customize that timeout, and handle errors with your own listener:

deferred.monitor(10000, function (err) {
  // Called for each promise not resolved in 10 seconds time 
});

This extension affects performance and it's best not to use it in production environment

Usage statistics

Being able to see how many promises were initialized (and where) in our flow can be helpful to track application issues, it's also good way to confirm that constructed flow works as intended.

deferred.profile(); // Start collecting statistics 
 
//... 
 
var stats = deferred.profileEnd(); // End profiling 
console.log(stats.log); // See readable output

Example log output:

------------------------------------------------------------
Deferred/Promise usage statistics:
 
104540 Total promises initialized
104540 Initialized as Unresolved
     0 Initialized as Resolved
 
Unresolved promises were initialized at:
 22590 at Object.module.exports.factory (/Users/medikoo/Projects/_packages/next/lib/fs/_memoize-watcher.js:21:10)
 11553 at Object.IsIgnored.init (/Users/medikoo/Projects/_packages/next/lib/fs/is-ignored.js:140:18)
 11553 at module.exports.factory (/Users/medikoo/Projects/_packages/next/lib/fs/_memoize-watcher.js:21:10)
  7854 at Object.Readdir.filterIgnored (/Users/medikoo/Projects/_packages/next/lib/fs/readdir.js:434:23)
  4619 at Object.module.exports.factory (/Users/medikoo/Projects/_packages/next/lib/fs/_memoize-watcher.js:21:10)
  3927 at Object.Readdir.filterByType (/Users/medikoo/Projects/_packages/next/lib/fs/readdir.js:222:15)
  3927 at Object.Readdir.filterByType (/Users/medikoo/Projects/_packages/next/lib/fs/readdir.js:236:15)
  3927 at Object.self (/Users/medikoo/Projects/_packages/next/lib/fs/readdir.js:164:12)
  3927 at Object.Readdir.readdir (/Users/medikoo/Projects/_packages/next/lib/fs/readdir.js:540:9)
  3927 at Object.self (/Users/medikoo/Projects/_packages/next/lib/fs/readdir.js:164:21)
  3729 at directory (/Users/medikoo/Projects/_packages/next/lib/fs/_watch-alt.js:95:2)
  2820 at Readdir.filterIgnored.promise.root (/Users/medikoo/Projects/_packages/next/lib/fs/readdir.js:517:9)
  2163 at basic (/Users/medikoo/Projects/_packages/next/lib/fs/is-gitrepo-root.js:14:8)
  2159 at buildMap (/Users/medikoo/Projects/_packages/next/lib/fs/is-ignored.js:117:22)
  2159 at Object.FindRoot (/Users/medikoo/Projects/_packages/next/lib/fs/find-root.js:18:15)
  1107 at Readdir.filterIgnored.promise.root (/Users/medikoo/Projects/_packages/next/lib/fs/readdir.js:527:11)
   697 at Object.Map.addPaths (/Users/medikoo/Projects/_packages/next/lib/fs/_get-conf-file-map.js:107:19)
   697 at readFile (/Users/medikoo/Projects/_packages/next/lib/fs/read-file.js:18:8)
   697 at Object.readRulesWatch (/Users/medikoo/Projects/_packages/next/lib/fs/_get-conf-file-map.js:45:12)
   247 at module.exports (/Users/medikoo/Projects/_packages/next/lib/fs/_watch.js:90:2)
   247 at module.exports (/Users/medikoo/Projects/_packages/next/lib/fs/_watch.js:90:13)
     1 at Object.Readdir.init (/Users/medikoo/Projects/_packa

Using profiler significantly affects performance don't use it in production environment.

Performance

Promises just by being rich objects introduce overhead over regular callbacks. If we do a lot asynchronous operations that are fast, performance of promise implementation that we rely on becomes a significant factor.

benchmark folder contains two tests. Tests reflect real use case I had in which performance of promise implementation appeared to be crucial.

Base of a test is lstat (fastest asynchronous call in node.js API), It's called 10000 times in parallel and sequentially.

Note for benchmark purists: This test does real I/O, but there's no way to produce shim which will provide more reliable results (shim based on setImmediate adds more randomness to result than we got from calling real lstat)

Example output taken under Node v0.10.20 on 2008 MBP.

Promise overhead (calling one after another) x10000:
 
 1:   439ms  Base (plain Node.js lstat call)
 2:   461ms  Kew: Dedicated wrapper
 3:   609ms  Bluebird: Dedicated wrapper
 4:   614ms  Bluebird: Promisify (generic wrapper)
 5:   642ms  Deferred: Dedicated wrapper
 6:   720ms  Deferred: Promisify (generic wrapper)
 7:   792ms  When: Dedicated wrapper
 8:  1068ms  Q: Dedicated wrapper
 9:  1611ms  Q: nbind (generic wrapper)
 
Promise overhead (concurrent calls) x10000:
 
 1:   279ms  Base (plain Node.js lstat call)
 2:   293ms  Bluebird: Promisify (generic wrapper)
 3:   294ms  Bluebird: Dedicated wrapper
 4:   329ms  Kew: Dedicated wrapper
 5:   406ms  When: Dedicated wrapper
 6:   430ms  Deferred: Dedicated wrapper
 7:   598ms  Deferred: Promisify (generic wrapper)
 8:   683ms  Deferred: Map + Promisify
 9:  1610ms  Q: Dedicated wrapper
10:  3645ms  Q: nbind (generic wrapper)

Tests

Covered by over 300 hundred unit tests

$ npm test

Tests

Stats

Keywords

Dependencies (4)