Promises are a tool to help write asynchronous code in a more readable style that looks more like synchronous code.
In short, promises allow you to interact with a value that may or may not be available yet.
Promises let wrap, and even chain, asynchronous operations using a consistent API, avoiding writing nested anonymous callbacks (the "pyramid of doom"). And they let you handle any errors that happen during those operations.
The `Y.Promise` class is compatible with the Promises/A+ specifiation.
As mentioned above, promises allow you to interact with a value that may or may not be available yet. In synchronous code, values are assigned to variables and immediately available to use, but if you need to use or assign a value that depends on an asynchronous operation to get, the rest of your code needs to be wrapped in a callback that is executed when that asynchronous operations completes.
Callbacks work, but they don't maintain any state, the APIs responsible for the callbacks are likely to differ, and they might not handle errors. It's also quite easy to find yourself building up multi-step transactions by nesting anonymous callbacks multiple levels deep.
Promises address this by providing an object that can be referred to immediately and any time in the future that represents the value produced by the asynchronous operation. Here's how you use them:
Promises operate using two methods: the `Y.Promise` constructor, and the promise instance's `then()` method. `\ // Create a promise for a value var promise = new Y.Promise(function (resolve, reject) { var promisedValue; // ...do some work to assign promisedValue, most likely asynchronously // When the work is done, fulfill the promise with the resolve function, // which was passed in the arguments. resolve(promisedValue); // Or if something went wrong, reject the promise with the reject function, // also passed in the arguments. reject(reasonForFailure); }); // Do something with the promised value using the then() method. then() takes // two functions as arguments. promise.then(onFulfilled, onRejected); promise.then( // aka onFulfilled function (promisedValue) { alert("Here's that value I promised I'd get for you: " + promisedValue); }, // aka onRejected function (reason) { alert("Oh no! I broke my promise. Here's why: " + reason); }); `\
The `Y.Promise` constructor takes as its argument a function we'll call the "executor function". This function is responsible for saying when the promised value is ready, or notifying that something went wrong.
The executor function receives two customized functions as its arguments, commonly called `resolve` and `reject`. If the work in the executor function to get the promised value completes successfully, pass the value to the `resolve()` method. If something went wrong, pass the reason—commonly an `Error`—to the `reject()` method.
// dataPromise represents the data parsed from the IO response,
// or the error that occurred fetching it
var dataPromise = new Y.Promise(function (resolve, reject) {
Y.io('getdata.php', {
on: {
success: function (id, response) {
// The IO completed, so the promise can be resolved
try {
resolve(Y.JSON.parse(response.responseText));
} catch (e) {
// any failure to produce the value is a rejection
reject(e);
}
},
failure: function (id, response) {
// The IO failed
reject(new Error("getdata.php request failed: " + response));
}
}
});
});
Promises can be in one of three states:
- `pending` - the promised value is not ready yet (default)
- `fulfilled` - the value is ready
- `rejected` - something went wrong, the value can't be produced
"Resolving" a promise moves a `pending` promise to either `fulfilled` or `rejected`, though the term is often used interchangeably with "fulfill" (it's good to have a positive outlook). Once a promise is fulfilled or rejected, it can't be transitioned to another state.
There are two ways promises get resolved. The first is using the `resolve()` function passed to the executor function in the `Y.Promise` constructor. We'll talk about the second way when we discuss promise chaining.
Since the promised value probably isn't ready when you create the promise, you can't synchronously consume the value. Schedule the code that will use the promised value to execute with the promise's `then()` method.
`then()` takes two callbacks as arguments, that we call `onFulfilled` and `onRejected`. As you might have guessed, the `onFulfilled` callback is executed if the promise resolves to a value, and the `onRejected` callback is executed if it is rejected.
Only one of the callbacks will be executed, and only once. Both callbacks are optional, though in practice you'll always pass at least one to `then()`.
var stuff;
var promise = new Y.Promise(getStuff);
// When getStuff says the promise is fulfilled, update the stuff variable.
// No onRejected callback is passed, so if there was an error, do nothing.
promise.then(function (stuffValue) {
stuff = stuffValue;
});
// Stuff isn't populated yet because the promise hasn't been fulfilled
console.log("Stuff value is " + stuff); // => "Stuff value is undefined"
It's important to note that even if the `getStuff` function above resolved the promise immediately, callbacks scheduled with `then` will always be called asynchronously. So the example code above will always log "Stuff value is undefined", regardless of whether `getStuff` operates synchronously or asynchronously.
To limit the runtime impact of `then` callbacks always being executed asynchronously, they are scheduled using `Y.soon()`, which will attempt to avoid any minimum delay that some browsers impose on `setTimeout`.
Here's where things start getting fun. When you call `promise.then(...)`, a new promise is returned. The new promise will resolve when either of the original promise's `onFulfilled` or `onRejected` functions returns a value or throws an error. This allows you to schedule several operations using chained `then()` calls.
// Verbose form
startSpinner();
// Create the initial promise
var userDataLoaded = new Y.Promise(function (resolve, reject) {
Y.io('getUserData.php', {
data: 'id=1234',
on: {
success: function (id, response) {
try {
resolve(Y.JSON.parse(response.responseText));
} catch (e) {
reject(e);
}
},
failure: function (id, response) {
reject(new Error(response));
}
}
});
});
// after the user data is loaded, render stuff or show the loading error
var uiUpdated = userDataLoaded.then(renderTemplates, showError);
// after the UI is updated, stop the spinner
uiUpdated.then(stopSpinner);
// Concise form (more common)
// Note Y.Promise can be called without 'new'
Promise(function (resolve, reject) { Y.io(...); })
.then(renderTemplates, showError) // returns another promise
.then(stopSpinner); // returns another promise
A chained promise is resolved by the return value of the previous promise's callbacks. Or, if an error is thrown, the chained promise is rejected.
Note that functions will return `undefined` if no explicit `return` statement is included. That will result in the promise being fulfilled with a value of `undefined`. Sometimes that's ok, but it's often helpful to pass along _some_ data.
function renderTemplates(userData) {
// Update the UI
Y.one('#userForm').setHTML(Y.Lang.sub(MyApp.userFormTemplate, userData));
// return a value to resolve the chained promise (aka uiUpdated) and pass
// to the uiUpdated's then(onFulfilled) callback, stopSpinner
return true;
}
function stopSpinner(updated) {
// updated will receive the return value of the previous promise's callback
// In this case, the boolean true.
var face = updated ? happyFace : sadFace;
spinnerNode.replace(face).hide(true);
}
// Using the original promise from the example above
userDataLoaded
.then(renderTemplates, showError)
.then(stopSpinner);
When a promise is rejected, the `onRejected` callback (the second argument to `then()`) is executed. Like `onFulfilled`, it is called with whatever is passed to the executor function's `reject()` function.
The `onRejected` callback can then rethrow the error to propagate the failed state, or recover from the failure by returning a value. Again, without an explicit `throw` or `return`, the callback will return `undefined` _which will mark the failure as recovered_, but with a resolved value of `undefined`. This may not be what you want!
function showError(reason) {
Y.one('#userForm').hide(true);
Y.one('#message .details').setHTML(reason.message || reason);
Y.one('#message').show();
// Choosing not to rethrow the error, but consider it recovered from for
// the sake of this transaction. Returning false as resolved value to send
// to stopSpinner.
return false;
}
userDataLoaded
.then(renderTemplates, showError)
.then(stopSpinner);
Because `showError` returned a value, and didn't rethrow the `reason`, the promise wrapping `renderTemplates` and `showError` was resolved to a "fulfilled" state with a value of `false`. Since the promise was fulfilled, not rejected, that promise's `onFulfilled` callback (`stopSpinner`) is called with the value `false`.
Because thrown errors are caught by the `Y.Promise` internals and used as a signal to reject a promise, it's possible to write promise chains that fail silently. This can be hard to debug.
To avoid this, it's highly recommended to always include an `onRejected` callback at the end of your promise chains. The reason you only need to put one at the end is discussed below.
Both `onFulfilled` and `onRejected` callbacks are optional, though in practice, you will always pass at least one. When a callback isn't provided for a `then()` call in a promise chain, that promise is automatically fulfilled with the value returned from the prior `onFulfilled` callback or rejected with the reason thrown from the prior `onRejected` callback.
getHandleFromServerA()
.then(null, getHandleFromServerB)
.then(getUserData)
.then(renderTemplates, showError);
// Same code, commented
// Try to get a DB handle from Server A...
getHandleFromServerA()
// if that fails, try Server B, otherwise, pass through the Server A handle
.then(null, getHandleFromServerB)
// if either server provided a handle, get user data.
// otherwise, there was an error, so pass it along the chain
.then(getUserData)
// render the user data if everything worked.
// if there was an error getting a DB handle or getting user data show it
.then(renderTemplates, showError);
It's not uncommon to see promise chains with only `onFulfilled` callbacks, then an `onRejected` callback at the very end.
As mentioned above, the return value from either `onFulfilled` or `onRejected` fulfills the promise with that value. There is one exception.
If you return a promise instead of a regular value (call it `returnedPromise`), the original promise will wait for `returnedPromise` to resolve, and adopt its state when it does. So if `returnedPromise` is fulfilled, the original promise is fulfilled with the same value, and if `returnedPromise` is rejected, the original promise is rejected with the same reason.
Y.Promise(function (resolve, reject) {
Y.io('getDataUrl.php', {
on: {
success: function (id, response) {
resolve(response.responseText);
},
failure: function () {
reject(new Error("Can't reach the server"));
}
}
});
})
// Chain another async operation by returning a promise.
// Don't worry, we'll wait for you.
.then(function (url) {
return new Y.Promise(function (resolve, reject) {
// Do another async operation
Y.jsonp(data.url, {
on: {
success: resolve,
failure: reject
}
});
});
})
// Called after both async operations have completed. The data response
// from the JSONP call is passed to renderTemplates
.then(renderTemplates)
// Then wait for 2 seconds before continuing the chain
.then(function () {
return new Y.Promise(function (resolve) {
setTimeout(resolve, 2000);
});
})
.then(hideMessage, showError);
Similarly, you can pass a promise to the `resolve()` function passed to the `Y.Promise` executor function.
Caution: Do not pass a promise to `reject()` or `throw` a promise from a callback. You're definitely doing something wrong if you find yourself doing that.
If you're unsure if a variable has a value or a promise, or you want an API to support both value or promise inputs, use `Y.when(value)` to wrap non-promise values in promises. Wrapped non-promise values will be immediately fulfilled with the wrapped value. Passing a promise to `Y.when` will return the promise.
// Accept either a regular object or a promise to save
MyDatabase.prototype.save = function (key, data) {
// Ensure we are dealing with a promise and call then() to get its value
// return the promise chained off this then() call
return Y.when(data).then(function (data) {
// Store the data somehow, for instance in localStorage
localStorage.set(key, data);
});
};
Promise chaining works great to serialize synchronous and asynchronous operations, but often several asynchronous operations can be performed simulataneously. This is where `Y.batch()` comes in.
`Y.batch()` takes any number of promises as arguments, and returns a new promise that will resolve when all the batched promises have resolved. The resolved value will be an array of values from the individual promises, in the order they were passed to `Y.batch()`.
If any one of the batched promises should be rejected, the batch promise is immediately rejected with that reason, so failures can be dealt with sooner rather than later.
Y.batch(
getUserAccountInfo(userId),
getUserPosts(userId, { page: 1, postsPerPage: 5 }),
getUserRank(userId)
)
.then(function (data) {
var account = data[0],
posts = data[1],
rank = data[2];
...
}, handleError);
Events are used to create a relationship between two objects, and better represent an open communication channel. Promises represent single values, and chains encapsulate transactions.
It's not uncommon to have event subscribers launch a promise chain, or to have events fired from within operations inside a promise chain. They are complementary tools.
`Y.AsyncQueue` is a tool for splitting up long synchronous operations into asynchronous chunks to avoid blocking UI updates unnecessarily. It doesn't (as yet) support asynchronous steps. It also supports conditional looping and various other things that promises don't, out of the box.
`Y.Parallel` is similar to `Y.batch` in that it provides a mechanism to execute a callback when several independent asynchronous operations have completed. However, it doesn't handle errors or guarantee asynchronous callback execution. It is also transactional, but the batch of operations is bound to a specific callback, where `Y.batch()` returns a promise that represents the aggregated values of those operations. The promise can be used by multiple consumers if necessary.
There are a lot of opportunities inside YUI to move transactional APIs to consume and/or return promises rather than use callbacks or one-time events. While there are no set plans for which APIs will be changed or in what priority order, you can expect to see promises showing up across the library in the near future.
Great job!