If a function cannot return a value or throw an exception without blocking, it can return a promise instead. A promise is an object that represents the return value or the thrown exception that the function may eventually provide. A promise can also be used as a proxy for a remote object to overcome latency.

On the first pass, promises can mitigate the “Pyramid of Doom”: the situation where code marches to the right faster than it marches forward.

step1(function (value1) {
    step2(value1, function(value2) {
        step3(value2, function(value3) {
            step4(value3, function(value4) {
                // Do something with value4
            });
        });
    });
});

With a promise library, you can flatten the pyramid.

Q.fcall(step1)
.then(step2)
.then(step3)
.then(step4)
.then(function (value4) {
    // Do something with value4
}, function (error) {
    // Handle any error from step1 through step4
})
.end();

With this approach, you also get implicit error propagation, just like try, catch, and finally. An error in step1 will flow all the way to step5, where it’s caught and handled.

The callback approach is called an “inversion of control”. A function that accepts a callback instead of a return value is saying, “Don’t call me, I’ll call you.”. Promises un-invert the inversion, cleanly separating the input arguments from control flow arguments. This simplifies the use and creation of API’s, particularly variadic, rest and spread arguments.

Getting Started

The Q module can be loaded as:

• a <script> tag (creating a Q global variable): only ~2 KB minified and gzipped!
• a NodeJS and CommonJS module available from NPM as the q package
• a RequireJS module

Q can exchange promises with jQuery and Dojo and the following libraries are based on Q.

• q-fs file system
• q-http http client and server
• q-comm remote objects
• jaque promising HTTP server, JSGI middleware

Many other projects in NPM use Q internally or provide Q promises.

Please join the Q-Continuum mailing list.

Tutorial

Promises have a then method, which you can use to get the eventual return value (fulfillment) or thrown exception (rejection).

foo()
.then(function (value) {
}, function (reason) {
})

If foo returns a promise that gets fulfilled later with a return value, the first function (the value handler) will be called with the value. However, if the foo function gets rejected later by a thrown exception, the second function (the error handler) will be called with the error.

Note that resolution of a promise is always asynchronous: that is, the value or error handler will always be called in the next turn of the event loop (i.e. process.nextTick in Node). This gives you a nice guarantee when mentally tracing the flow of your code, namely that then will always return before either handler is executed.

Propagation

The then method returns a promise, which in this example, I’m assigning to bar.

var bar = foo()
.then(function (value) {
}, function (reason) {
})

The bar variable becomes a new promise for the return value of either handler. Since a function can only either return a value or throw an exception, only one handler will ever be called and it will be responsible for resolving bar.

• If you return a value in a handler, bar will get fulfilled.

• If you throw an exception in a handler bar will get rejected.

• If you return a promise in a handler, bar will “become” that promise. Being able to become a new promise is useful for managing delays, combining results, or recovering from errors.

If the foo() promise gets rejected and you omit the error handler, the error will go to bar:

var bar = foo()
.then(function (value) {
})

If the foo() promise gets fulfilled and you omit the value handler, the value will go to bar:

var bar = foo()
.then(null, function (error) {
})

Q promises provide a fail shorthand for then when you are only interested in handling the error:

var bar = foo()
.fail(function (error) {
})

They also have a fin function that is like a finally clause. The final handler gets called, with no arguments, when the promise returned by foo() either returns a value or throws an error. The value returned or error thrown by foo() passes directly to bar.

var bar = foo()
.fin(function () {
    // close files, database connections, stop servers, conclude tests
})

• If the handler returns a value, the value is ignored
• If the handler throws an error, the error passes to bar
• If the handler returns a promise, bar gets postponed. The eventual value or error has the same effect as an immediate return value or thrown error: a value would be ignored, an error would be forwarded.

Chaining

There are two ways to chain promises. You can chain promises either inside or outside handlers. The next two examples are equivalent.

return foo()
.then(function (fooValue) {
    return bar(fooValue)
    .then(function (barValue) {
        // if we get here without an error,
        // the value returned here
        // or the exception thrown here
        // resolves the promise returned
        // by the first line
    })
})

return foo()
.then(function (fooValue) {
    return bar(fooValue);
})
.then(function (barValue) {
    // if we get here without an error,
    // the value returned here
    // or the exception thrown here
    // resolves the promise returned
    // by the first line
})

The only difference is nesting. It’s useful to nest handlers if you need to capture both fooValue and barValue in the last handler.

function eventualAdd(a, b) {
    return a.then(function (a) {
        return b.then(function (b) {
            return a + b;
        });
    });
}

Combination

You can turn an array of promises into a promise for the whole, fulfilled array using all.

return Q.all([
    eventualAdd(2, 2),
    eventualAdd(10, 20)
])

If you have a promise for an array, you can use spread as a replacement for then. The spread function “spreads” the values over the arguments of the value handler. The error handler will get called at the first sign of failure. That is, whichever of the recived promises fails first gets handled by the error handler.

function eventualAdd(a, b) {
    return Q.all([a, b])
    .spread(function (a, b) {
        return a + b;
    })
}

But spread calls all initially, so you can skip it in chains.

return foo()
.then(function (name, location) {
    return [name, FS.read(location, "utf-8")];
    // FS.read returns a promise, so this array
    // mixes values and promises
})
.spread(function (name, text) {
})

And you can use Q.spread directly on an array of promises.

function eventualAdd(a, b) {
    return Q.spread([a, b], function (a, b) {
        return a + b;
    })
}

The all function returns a promise for an array of values. If one of the given promise fails, the whole returned promise fails, not waiting for the rest of the batch. If you want to wait for all of the promises to either be fulfilled or rejected, you can use allResolved.

Q.allResolved(promises)
.then(function (promises) {
    promises.forEach(function (promise) {
        if (promise.isFulfilled()) {
            var value = promise.valueOf();
        } else {
        }
    })
})

Sequences

If you have a number of promise-producing functions that need to be run sequentially, you can of course do so manually:

return foo(initialVal).then(bar).then(baz).then(quux);

However, if you want to run a dynamically constructed sequence of functions, you'll want something like this:

var funcs = [foo, bar, baz, quux];

var result = Q.resolve(initialVal);
funcs.forEach(function (f) {
    result = result.then(f);
});
return result;

You can make this slightly more compact using reduce:

return funcs.reduce(function (soFar, f) {
    return soFar.then(f);
}, Q.resolve(initialVal));

Handling Errors

One sometimes-unintuive aspect of promises is that if you throw an exception in the value handler, it will not be be caught by the error handler.

foo()
.then(function (value) {
    throw new Error("Can't bar.");
}, function (error) {
    // We only get here if "foo" fails
})

To see why this is, consider the parallel between promises and try/catch. We are try-ing to execute foo(): the error handler represents a catch for foo(), while the value handler represents code that happens after the try/catch block. That code then needs its own try/catch block.

In terms of promises, this means chaining your error handler:

foo()
.then(function (value) {
    throw new Error("Can't bar.");
})
.fail(function (error) {
    // We get here with either foo's error or bar's error
})

The End

When you get to the end of a chain of promises, you should either return the last promise or end the chain. Since handlers catch errors, it’s an unfortunate pattern that the exceptions can go unobserved.

So, either return it,

return foo()
.then(function () {
    return "bar";
})

Or, end it.

foo()
.then(function () {
    return "bar";
})
.end()

Ending a promise chain makes sure that, if an error doesn’t get handled before the end, it will get rethrown and reported.

This is a stopgap. We are exploring ways to make unhandled errors visible without any explicit handling.

The Beginning

Everything above assumes you get a promise from somewhere else. This is the common case. Every once in a while, you will need to create a promise from scratch.

You can create a promise from a value using Q.call. This returns a promise for 10.

return Q.fcall(function () {
    return 10;
});

You can also use call to get a promise for an exception.

return Q.fcall(function () {
    throw new Error("Can't do it");
})

As the name implies, call can call functions, or even promised functions. This uses the eventualAdd function above to add two numbers. The second argument is the this object to pass into the function.

return Q.fcall(eventualAdd, null, 2, 2);

When nothing else will do the job, you can use defer, which is where all promises ultimately come from.

var deferred = Q.defer();
FS.readFile("foo.txt", "utf-8", function (error, text) {
    if (error) {
        deferred.reject(new Error(error));
    } else {
        deferred.resolve(text);
    }
});
return deferred.promise;

Note that a deferred can be resolved with a value or a promise. The reject function is a shorthand for resolving with a rejected promise.

var rejection = Q.fcall(function () {
    throw new Error("Can't do it");
});
deferred.resolve(rejection);

This is a simplified implementation of Q.delay.

function delay(ms) {
    var deferred = Q.defer();
    setTimeout(deferred.resolve, ms);
    return deferred.promise;
}

This is a simplified implementation of Q.timeout

function timeout(promise, ms) {
    var deferred = Q.defer();
    Q.when(promise, deferred.resolve);
    Q.when(delay(ms), function () {
        deferred.reject(new Error("Timed out"));
    });
    return deferred.promise;
}

The Middle

If you are using a function that may return a promise, but just might return a value if it doesn’t need to defer, you can use the “static” methods of the Q library.

The when function is the static equivalent for then.

return Q.when(valueOrPromise, function (value) {
}, function (error) {
});

All of the other methods on a promise have static analogs with the same name.

The following are equivalent:

return Q.all([a, b]);

return Q.fcall(function () {
    return [a, b];
})
.all();

When working with promises provided by other libraries, you should convert it to a Q promise. Not all promise libraries make the same guarantees as Q and certainly don’t provide all of the same methods. Most libraries only provide a partially functional then method. This thankfully is all we need to turn them into vibrant Q promises.

return Q.when($.ajax(...))
.then(function () {
})

If there is any chance that the promise you receive is not a Q promise as provided by your library, you should wrap it using a Q function. You can even use Q.invoke as a shorthand.

return Q.invoke($, 'ajax', ...)
.then(function () {
})

Over the Wire

A promise can serve as a proxy for another object, even a remote object. There are methods that allow you to optimistically manipulate properties or call functions. All of these interactions return promises, so they can be chained.

direct manipulation         using a promise as a proxy
--------------------------  -------------------------------
value.foo                   promise.get("foo")
value.foo = value           promise.put("foo", value)
delete value.foo            promise.del("foo")
value.foo(...args)          promise.post("foo", [args])
value.foo(...args)          promise.invoke("foo", ...args)
value(...args)              promise.fapply([args])
value(...args)              promise.fcall(...args)

If the promise is a proxy for a remote object, you can shave round-trips by using these functions instead of then. To take advantage of promises for remote objects, check out Q-Comm.

Even in the case of non-remote objects, these methods can be used as shorthand for particularly-simple value handlers. For example, you can replace

return Q.fcall(function () {
    return [{ foo: "bar" }, { foo: "baz" }];
})
.then(function (value) {
    return value[0].foo;
})

with

return Q.fcall(function () {
    return [{ foo: "bar" }, { foo: "baz" }];
})
.get(0)
.get("foo")

Adapting Node

There is a makeNodeResolver method on deferreds that is handy for the NodeJS callback pattern.

var deferred = Q.defer();
FS.readFile("foo.txt", "utf-8", deferred.makeNodeResolver());
return deferred.promise;

And there are Q.ncall and Q.ninvoke for even shorter expression.

return Q.ncall(FS.readFile, FS, "foo.txt", "utf-8");

return Q.ninvoke(FS, 'readFile', "foo.txt", "utf-8");

There is also a Q.nbind function that that creates a reusable wrapper.

var readFile = Q.nbind(FS.readFile, FS)
return readFile("foo.txt", "utf-8");

Note that, since promises are always resolved in the next turn of the event loop, working with streams can be tricky. The essential problem is that, since Node does not buffer input, it is necessary to attach your "data" event listeners immediately, before this next turn comes around. There are a variety of solutions to this problem, and even some hope that in future versions of Node it will be ameliorated.

Reference

A method-by-method Q API reference is available on the wiki.

---

Copyright 2009-2012 Kristopher Michael Kowal MIT License (enclosed)