Asynchronous Routing

Ember's approach to handling asynchronous logic in the router makes heavy use of the concept of Promises. In short, promises are objects that represent an eventual value. A promise can either fulfill (successfully resolve the value) or reject (fail to resolve the value). The way to retrieve this eventual value, or handle the cases when the promise rejects, is via the promise's then method, which accepts two optional callbacks, one for fulfillment and one for rejection. If the promise fulfills, the fulfillment handler gets called with the fulfilled value as its sole argument, and if the promise rejects, the rejection handler gets called with a reason for the rejection as its sole argument. For example:

var promise = fetchTheAnswer();

promise.then(fulfill, reject);

function fulfill(answer) {
  console.log("The answer is " + answer);
}

function reject(reason) {
  console.log("Couldn't get the answer! Reason: " + reason);
}

Much of the power of promises comes from the fact that they can be chained together to perform sequential asynchronous operations:

// Note: jQuery AJAX methods return promises
var usernamesPromise = Ember.$.getJSON('/usernames.json');

usernamesPromise.then(fetchPhotosOfUsers)
                .then(applyInstagramFilters)
                .then(uploadTrendyPhotoAlbum)
                .then(displaySuccessMessage, handleErrors);

In the above example, if any of the methods fetchPhotosOfUsers, applyInstagramFilters, or uploadTrendyPhotoAlbum returns a promise that rejects, handleErrors will be called with the reason for the failure. In this manner, promises approximate an asynchronous form of try-catch statements that prevent the rightward flow of nested callback after nested callback and facilitate a saner approach to managing complex asynchronous logic in your applications.

This guide doesn't intend to fully delve into all the different ways promises can be used, but if you'd like a more thorough introduction, take a look at the readme for RSVP, the promise library that Ember uses.

When transitioning between routes, the Ember router collects all of the models (via the model hook) that will be passed to the route's controllers at the end of the transition. If the model hook (or the related beforeModel or afterModel hooks) return normal (non-promise) objects or arrays, the transition will complete immediately. But if the model hook (or the related beforeModel or afterModel hooks) returns a promise (or if a promise was provided as an argument to transitionTo), the transition will pause until that promise fulfills or rejects.

The router considers any object with a then method defined on it to be a promise.

If the promise fulfills, the transition will pick up where it left off and begin resolving the next (child) route's model, pausing if it too is a promise, and so on, until all destination route models have been resolved. The values passed to the setupController hook for each route will be the fulfilled values from the promises.

A basic example:

export default Ember.Route.extend({
  model: function() {
    return new Ember.RSVP.Promise(function(resolve) {
      Ember.run.later(function() {
        resolve({ msg: "Hold Your Horses" });
      }, 3000);
    });
  },

  setupController: function(controller, model) {
    console.log(model.msg); // "Hold Your Horses"
  }
});

When transitioning into route:tardy, the model hook will be called and return a promise that won't resolve until 3 seconds later, during which time the router will be paused in mid-transition. When the promise eventually fulfills, the router will continue transitioning and eventually call route:tardy's setupController hook with the resolved object.

This pause-on-promise behavior is extremely valuable for when you need to guarantee that a route's data has fully loaded before displaying a new template.

We've covered the case when a model promise fulfills, but what if it rejects?

By default, if a model promise rejects during a transition, the transition is aborted, no new destination route templates are rendered, and an error is logged to the console.

You can configure this error-handling logic via the error handler on the route's actions hash. When a promise rejects, an error event will be fired on that route and bubble up to route:application's default error handler unless it is handled by a custom error handler along the way, e.g.:

export default Ember.Route.extend({
  model: function() {
    return Ember.RSVP.reject("FAIL");
  },

  actions: {
    error: function(reason) {
      alert(reason); // "FAIL"

      // Can transition to another route here, e.g.
      // this.transitionTo('index');

      // Uncomment the line below to bubble this error event:
      // return true;
    }
  }
});

In the above example, the error event would stop right at route:good-for-nothing's error handler and not continue to bubble. To make the event continue bubbling up to route:application, you can return true from the error handler.

Rejected model promises halt transitions, but because promises are chainable, you can catch promise rejects within the model hook itself and convert them into fulfills that won't halt the transition.

export default Ember.Route.extend({
  model: function() {
    return iHopeThisWorks().then(null, function() {
      // Promise rejected, fulfill with some default value to
      // use as the route's model and continue on with the transition
      return { msg: "Recovered from rejected promise" };
    });
  }
});

The model hook covers many use cases for pause-on-promise transitions, but sometimes you'll need the help of the related hooks beforeModel and afterModel. The most common reason for this is that if you're transitioning into a route with a dynamic URL segment via {{link-to}} or transitionTo (as opposed to a transition caused by a URL change), the model for the route you're transitioning into will have already been specified (e.g. {{#link-to 'article' article}} or this.transitionTo('article', article)), in which case the model hook won't get called. In these cases, you'll need to make use of either the beforeModel or afterModel hook to house any logic while the router is still gathering all of the route's models to perform a transition.

Easily the more useful of the two, the beforeModel hook is called before the router attempts to resolve the model for the given route. In other words, it is called before the model hook gets called, or, if model doesn't get called, it is called before the router attempts to resolve any model promises passed in for that route.

Like model, returning a promise from beforeModel will pause the transition until it resolves, or will fire an error if it rejects.

The following is a far-from-exhaustive list of use cases in which beforeModel is very handy:

• Deciding whether to redirect to another route before performing a potentially wasteful server query in model
• Ensuring that the user has an authentication token before proceeding onward to model
• Loading application code required by this route

export default Ember.Route.extend({
  beforeModel: function() {
    if (!this.controllerFor('auth').get('isLoggedIn')) {
      this.transitionTo('login');
    }
  }
});

See the API Docs for beforeModel

The afterModel hook is called after a route's model (which might be a promise) is resolved, and follows the same pause-on-promise semantics as model and beforeModel. It is passed the already-resolved model and can therefore perform any additional logic that depends on the fully resolved value of a model.

export default Ember.Route.extend({
  model: function() {
    // `this.store.find('article')` returns a promise-like object
    // (it has a `then` method that can be used like a promise)
    return this.store.find('article');
  },
  afterModel: function(articles) {
    if (articles.get('length') === 1) {
      this.transitionTo('article.show', articles.get('firstObject'));
    }
  }
});

You might be wondering why we can't just put the afterModel logic into the fulfill handler of the promise returned from model; the reason, as mentioned above, is that transitions initiated via {{link-to}} or transitionTo likely already provided the model for this route, so model wouldn't be called in these cases.

See the API Docs for afterModel

• Embercasts: Client-side Authentication Part 2
• RC6 Blog Post describing these new features