promise
Rationale
Most of the JS APIs are asynchronous complementing it's non-blocking nature. While this has a good reason and many advantages, it comes with a price. Instead of structuring our programs into logical black boxes:
function blackbox(a, b) {
var c = assemble(a);
return combine(b, c);
}
We're forced into continuation passing style, involving lot's of machinery:
function sphagetti(a, b, callback) {
assemble(a, function continueWith(error, c) {
if (error) callback(error);
else combine(b, c, callback);
});
}
This style also makes doing things in sequence hard:
widget.on('click', function onClick() {
promptUserForTwitterHandle(function continueWith(error, handle) {
if (error) return ui.displayError(error);
twitter.getTweetsFor(handle, funtion continueWith(error, tweets) {
if (error) return ui.displayError(error);
ui.showTweets(tweets);
});
});
});
Doing things in parallel is even harder:
var tweets, answers, checkins;
twitter.getTweetsFor(user, function continueWith(result) {
tweets = result;
somethingFinished();
});
stackOverflow.getAnswersFor(question, function continueWith(result) {
answers = result;
somethingFinished();
});
fourSquare.getCheckinsBy(user, function continueWith(result) {
checkins=result;
somethingFinished();
});
var finished = 0;
functions somethingFinished() {
if (++finished === 3)
ui.show(tweets, answers, checkins);
}
This also makes error handling quite of an adventure.
Promises
Consider another approach, where instead of continuation passing via callback
,
function returns an object, that represents eventual result, either successful
or failed. This object is a promise, both figuratively and by name, to
eventually resolve. We can call a function on the promise to observe
either its fulfillment or rejection. If the promise is rejected and the
rejection is not explicitly observed, any derived promises will be implicitly
rejected for the same reason.
In the Add-on SDK we follow
CommonJS Promises/A specification
and model a promise as an object with a then
method, which can be used to get
the eventual return (fulfillment) value or thrown exception (rejection):
foo().then(function success(value) {
// ...
}, function failure(reason) {
// ...
});
If foo
returns a promise that gets fulfilled with the value
, success
callback (the value handler) will be called with that value
. However,
if the returned promise gets rejected, the failure
callback (the error
handler) will be called with the reason
of an error.
Propagation
The then
method of a promise returns a new promise that is resolved with the
return value of either handler. Since function can either return value or throw
an exception, only one handler will be ever called.
var bar = foo().then(function success(value) {
// compute something from a value...
}, function failure(reason) {
// handle an error...
});
In this example bar
is a promise and it's fulfilled by one of two handlers
that are responsible for:
- If handler returns a value,
bar
will be resolved with it. - If handler throws an exception,
bar
will be rejected with it. - If handler returns a promise,
bar
will "become" that promise. To be more precise it will be resolved with a resolution value of the returned promise, which will appear and feel as if it was that returned promise.
If the foo()
promise gets rejected and you omit the error handler, the
error will propagate to bar
(bar
will be rejected with that error):
var bar = foo().then(function success(value) {
// compute something out of the value...
});
If the foo()
promise gets fulfilled and you omit the value handler, the
value will propagate to bar
(bar
will be fulfilled with that value):
var bar = foo().then(null, function failure(error) {
// handle error...
});
Chaining
There are two ways to chain promise operations. You can chain them using either inside or outside handlers.
Flat chaining
You can use then
for chaining intermediate operations on promises
(var data = readAsync().then(parse).then(extract)
). You can chain multiple
then
functions, because then
returns a promise resolved to a return value
of an operation and errors propagate through the promise chains. In general
good rule of thumb is to prefer then
based flat chaining. It makes code
easier to read and make changes later:
var data = readAsync(url). // read content of url asynchronously
then(parse). // parse content from the url
then(extractQuery). // extract SQL query
then(readDBAsync); // exectue extracted query against DB
Nested chaining
Flat chaining is not always an option though, as in some cases you may want to capture an intermediate values of the chain:
var result = readAsync(url).then(function(source) {
var json = parse(source)
return readDBAsync(extractQuery(json)).then(function(data) {
return writeAsync(json.url, data);
});
});
In general, nesting is useful for computing values from more then one promise:
function eventualAdd(a, b) {
return a.then(function (a) {
return b.then(function (b) {
return a + b;
});
});
}
var c = eventualAdd(aAsync(), bAsync());
Error handling
One sometimes-unintuitive aspect of promises is that if you throw an exception in the value handler, it will not be be caught by the error handler.
readAsync(url).then(function (value) {
throw new Error("Can't bar.");
}, function (error) {
// We only get here if `readAsync` fails.
});
To see why this is, consider the parallel between promises and try
/catch
.
We are try
-ing to execute readAsync()
: the error handler represents a
catch
for readAsync()
, while the value handler represents code that happens
after the try
/catch
block. That code then needs its own try
/catch
block to handle errors there.
In terms of promises, this means chaining your error handler:
readAsync(url).
then(parse).
then(null, function handleParseError(error) {
// handle here both `readAsync` and `parse` errors.
});
Consuming promises
In general, whole purpose of promises is to avoid a callback spaghetti in the
code. As a matter of fact it would be great if we could convert any synchronous
functions to asynchronous by making it aware of promises. Module exports
promised
function to do exactly that:
const { promised } = require('api-utils/promise');
function sum(x, y) { return x + y }
var sumAsync = promised(sum);
var c = sum(a, b);
var cAsync = asyncSum(aAsync(), bAsinc());
promised
takes normal function and composes new promise aware version of it
that may take both normal values and promises as arguments and returns promise
that will resolve to value that would have being returned by an original
function if it was called with fulfillment values of given arguments.
This technique is so powerful that it can replace most of the promise utility functions provided by other promise libraries. For example grouping promises to observe single resolution of all of them is as simple as this:
var group = promised(Array);
var abc = group(aAsync, bAsync, cAsync).then(function(items) {
return items[0] + items[1] + items[2];
});
Making promises
Everything above assumes you get a promise from somewhere else. This
is the common case, but every once in a while, you will need to create a
promise from scratch. Add-on SDK's promise
module provides API for doing
that.
defer
Module exports defer
function, which is where all promises ultimately
come from. Lets see implementation of readAsync
that we used in lot's
of examples above:
const { defer } = require('api-utils/promise');
function readAsync(url) {
var deferred = defer();
let xhr = new XMLHttpRequest();
xhr.open("GET", url, true);
xhr.onload = function() {
deferred.resolve(xhr.responseText);
}
xhr.onerror = function(event) {
deferred.reject(event);
}
xhr.send();
return deferred.promise;
}
So defer
returns an object that contains promise
and two resolve
, reject
functions that can be used to resolve / reject that promise
. Note: that
promise can be rejected or resolved and only once. All subsequent attempts will be
ignored.
Another simple example may be delay
function that returns promise which
is fulfilled with a given value
in a given ms
, kind of promise based
alternative to setTimeout
:
function delay(ms, value) {
let { promise, resolve } = defer();
setTimeout(resolve, ms, value);
return promise;
}
delay(10, 'Hello world').then(console.log);
// After 10ms => 'Helo world'
Advanced usage
If general defer
and promised
should be enough to doing almost anything
you may think of with promises, but once you start using promises extensively
you may discover some missing pieces and this section of documentation may
help you to discover them.
Doing things concurrently
So far we have being playing with promises that do things sequentially, but there are bunch of cases where one would need to do things concurrently. In the following example we implement functions that takes multiple promises and returns one that resolves to first on being fulfilled:
function race() {
let { promise, resolve } = defer();
Array.slice(arguments).forEach(function(promise) {
promise.then(resolve);
});
return promise;
}
var asyncAorB = race(readAsync(urlA), readAsync(urlB));
Note: that this implementation forgives failures and would fail if all promises fail to resolve.
There are cases when promise may or may not be fulfilled in a reasonable time. In such cases it's useful to put a timer on such tasks:
function timeout(promise, ms) {
let deferred = defer();
promise.then(deferred.resolve, deferred.reject);
delay(ms, 'timeout').then(deferred.reject);
return deferred.promise;
}
var tweets = readAsync(url);
timeout(tweets, 20).then(function(data) {
ui.display(data);
}, function() {
alert('Network is being too slow, try again later');
});
Alternative promise APIs
There may be a cases where you will want to provide more than just then
method on your promises. In fact some other promise frameworks do that.
Such use cases are also supported. Earlier described defer
may be passed
optional prototype
argument, in order to make returned promise and all
the subsequent promises decedents of that prototype
:
let { promise, resolve } = defer({
get: function get(name) {
return this.then(function(value) {
return value[name];
})
}
});
promise.get('foo').get('bar').then(console.log);
resolve({ foo: { bar: 'taram !!' } });
// => 'taram !!'
Also promised
function maybe passed second optional prototype
argument to
achieve same effect.
Treat all values as promises
Module provides a simple function for wrapping values into promises:
const { resolve } = require('api-utils/promise');
var a = resolve(5).then(function(value) {
return value + 2
});
a.then(console.log); // => 7
Also resolve
not only takes values, but also promises. If you pass it
a promise it will return new identical one:
const { resolve } = require('api-utils/promise');
resolve(resolve(resolve(3))).then(console.log); // => 3
If this construct may look strange at first, but it becomes quite handy when writing functions that deal with both promises and values. In such cases it's usually easier to wrap value into promise than branch on value type:
function or(a, b) {
var second = resolve(b).then(function(bValue) { return !!bValue });
return resolve(a).then(function(aValue) {
return !!aValue || second;
}, function() {
return second;
})
}
Note: We could not use promised
function here, as they reject returned
promise if any of the given arguments is rejected.
If you need to customize your promises even further you may pass resolve
a
second optional prototype
argument that will have same effect as with defer
.
Treat errors as promises
Now that we can create all kinds of eventual values, it's useful to have a
way to create eventual errors. Module exports reject
exactly for that.
It takes anything as an argument and returns a promise that is rejected with
it.
const { reject } = require('api-utils/promise');
var boom = reject(Error('boom!'));
future(function() {
return Math.random() < 0.5 ? boom : value
})
As with rest of the APIs error may be given second optional prototype
argument to customize resulting promise to your needs.