Stabilizing Ember Data

– By Tom Dale, Yehuda Katz

Yesterday, we gave you an update on our progress making Ember.js easier to use. One thing we didn't discuss was our plan for Ember Data.

It's no secret that, while many developers are building awesome apps with Ember.js, Ember Data still causes lots of frustration due to bugs and a changing, complex API. Documentation about it is also mixed in with Ember.js documentation, making it difficult for new developers to understand what is stable and what is not.

To be clear, Ember Data is not a dependency of Ember.js. Discourse, for example, uses its own wrapper around $.ajax.

Even though Ember Data is not a dependency of Ember.js, loading data from the server is an extremely important part of most web applications, and it's a problem that every Ember.js application will have to deal with.

Our long-term goal is as follows: we don't think most web developers should have to write any custom XHR code for loading data. Strong conventions on the client and strong conventions on the server should allow them to communicate automatically.

We know we're not there yet.

In order for this to work, there are many necessary features that must be rock-solid across all sorts of different persistence layers—local storage, relational databases, and key-value stores, to name a few. To top it off, the asynchronous environment of the browser (with an often unreliable internet connection) adds significant complexity, and means we have more to do besides porting solutions to these problems that have been pioneered on the server.

Getting all of these features working well together is a challenging problem, and we have not been able to deliver everything we thought we could in a reasonable amount of time.

Additionally, many developers are writing web applications that need to consume an existing JSON API that evolved organically and is not consistently named or structured. The API we have provided so far is suboptimal for this task.

To make the experience of writing an Ember Data application less frustrating, we're doing two things:

  1. We're identifying a subset of features that already work reliably.
  2. We're introducing a new, simpler API for working with remote data that makes fewer assumptions. You still get to use the model API in Ember Data, but can "bring your own $.ajax" to load and store records.

Right now, what gets documented is somewhat ad-hoc. Going forward, we will heavily document the stable features. Finding documentation about an Ember Data feature on the emberjs.com website will be your indication that we consider it stable and safe to use.

Over time, after we have put new features through their paces and written extensive documentation, the set of stable features will grow. In other words, we are refocusing on a small core of practical features, which we will slowly iterate towards our long-term, ambitious goal.

The Basic Adapter

The Ember Data adapter layer, which is responsible for finding and saving records, is currently designed to help with building reusable adapters, like the ones people have written for Parse or CouchDB. It is not well-suited for delegating out to $.ajax to work with an API that is not 100% consistent.

To make it easier to use Ember Data with any kind of JSON data, we are introducing the Basic Adapter, which delegates to a sync object on your model.

Let's look at an example of using the Basic Adapter with the Twitter API.

First, note that the syntax for defining a model hasn't changed (historically, this has been one of the most stable parts of Ember Data):

var attr = DS.attr, hasMany = DS.hasMany, belongsTo = DS.belongsTo;

App.User = DS.Model.extend({
  defaultProfileImage: attr('boolean'),
  description: attr('string'),
  screenName: attr('string'),
  isVerified: attr('boolean'),
  createdAt: attr('date'),

  tweets: hasMany('App.Tweet')
});

App.Tweet = DS.Model.extend({
  coordinates: attr('point'),
  createdAt: attr('date'),
  isFavorited: attr('boolean'),
  retweetCount: attr('number'),
  text: attr('string'),
  isTruncated: attr('boolean'),

  replyTo: belongsTo('App.User'),
  user: belongsTo('App.User')
});

Finding records also hasn't changed:

// use the Promise API
App.User.find(userId).then(function(user) {
  return user.get('tweets');
}).then(function(tweets) {
  // do something with `tweets`
});

When you request the User and then its tweets, Ember Data will make calls to your models' sync object.

App.User.sync = {
  find: function(id, process) {
    $.getJSON("/users/show", { screen_name: id }).then(function(user) {        
      process(user)                 
        .primaryKey('screen_name')
        .camelizeKeys()
        .applyTransforms('twitter')
        .load();
    });
  },

  findTweets: function(user, name, process) {
    var screenName = user.get('id');

    $.getJSON("/statuses/user_timeline", { screen_name: screenName }).then(function(timeline) {
      process(timeline)
        .primaryKey('id_str')
        .camelizeKeys()
        .applyTransforms('twitter')
        .munge(function(json) {
          json.isTruncated = json.truncated;
          json.replyTo = json.inReplyToScreenName;
        })
        .load();
    });
  }
};

As you can see, in each of the hooks on the sync object, the last argument passed in is a function called process. You use this function to load JSON data returned from the XHR into the store. It also includes several conveniences for common transformations, like camelizing property names and transforming values like dates.

Note that you are not required to use these conveniences. You can write whatever imperative code you'd like to transform the JSON returned from the server into the form that Ember Data is expecting. Here is the above example re-written without using the chained conveniences.

findTweets: function(user, name, process) {
  var screenName = user.get('id');

  $.getJSON("/statuses/user_timeline", { screen_name: screenName }).then(function(timeline) {
    var tweets = timeline.map(function(json) {
      // Map primary key
      json.id = json.id_str;

      // Camelize property names
      for (var prop in json) {
        var value = json[prop];
        delete json[prop];
        json[camelize(prop)] = value;
      }

      // Convert string-formatted date to object
      json.createdAt = Date.parse(json.createdAt);

      // Convert hash to JavaScript object
      json.coordinates = new Twitter.Point(json.coordinates);

      // Rename properties
      json.isTruncated = json.truncated;
      json.replyTo = json.inReplyToScreenName;

      return json;
    });

    process(tweets).load();
  });
}

Timeline

We have been working on this new API part-time for the past few weeks. You can see our progress on Ember Data's master branch, by looking at the tests or the implementation.

Over the next few weeks, we will be writing documentation that will be available in the Ember.js Guides. Once a few people have had the opportunity to use the Basic Adapter and sanity-check our work, we will start cutting beta releases of Ember Data. We think that this will be a lot easier for new developers than "make a build from master."

Our thanks go out to John McDowall, who has been tracking our progress on Basic Adapter and writing documentation to go with it.

Finally, we'd like to give a big thanks to Addepar for financially supporting us while we do this work. They are big users of and contributors to Ember Data, and we couldn't do it without them.