Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
767 lines (625 sloc) 39 KB

How we built TodoMVC - Example


TodoMVC is a project which offers a wide verity of implementations to a well defined application (called TodoMVC) using popular JavaScript MV* frameworks. We have implemented TodoMVC using Backbone and Marionette as our core MV* frameworks, Handlebars as our templating engine, and RequireJS as our AMD framework. Before continuing, you should get familiarized with the basic concepts of the aforementioned frameworks and the flow of the TodoMVC application. In this chapter we present and explain the various decisions we took while developing this flavor of TodoMVC application. Please feel free to leave your comments and contributions in the issues section. Note that for the sake of this example we added a welcome screen which is not part of the specification. The welcome screen takes you to our implementation of TodoMVC which is aligned with the spec.

Code Structure

Before deciding on the structure of the application we must think of the build and deployment process. We consider several possibilities:

  • Put all the code in a big (actually HUGE) JavaScript file (or concatenate several files into one file).

  • Use Marionette built in modules.

  • Use CommonJS style modules.

  • Use AMD style modules.

Clearly the first option is not a good idea for even a small sized application, even when concatenating several separate files into one big JavaScript file in order to enable easier development. There are various issues with this process starting from awkward debugging (where you debug a >7K lines file that keeps changing under your feet) and ending with dependencies on the order of the concatenation which are not always easy to overcome. The main benefit of this method is that it creates a single file which is faster to download than many small files, especially on high latency connections. Marionette modules are a good solution but they do not offer a full solution for a larger scale application. Although in the case of TodoMVC Marionette modules are more than enough, we want to simulate building a large scale application. The choice between CommonJS and AMD is harder because both concepts are somewhat similar and would fit our needs (For example Node.js uses CommonJS style modules). Our decision is to go with AMD style modules due to the following reasons:

  • There is a complete solution for AMD style modules in the form of the RequireJS framework.

  • There is a full support of AMD modules by Marionette.

  • There is a plug-in that supports handlebars templates for RequireJS.

  • There is an optimizer which comes with RequireJS that allows compilation of all the code into a single JavaScript file for production mode.

Choosing AMD style modules using RequireJS as our modules system allows us to use a simple directory structure which is aligned with the TodoMVC specifications.

Directory Structure

The basic directory structure for our application is:

|- lib
|- js
|  |- common
|  |- welcome
|  |  |-views
|  |  |  |- templates
|  |  |- controller.js
|  |  |- router.js
|  |- todo-list
|  |  |-views
|  |  |  |- templates
|  |  |- models
|  |  |- controller.js
|  |  |- router.js
|  |- main.js
|  |- app.js
|- css
|- assets
|- index.html

(For brevity we omit some of the internal files).

In the root folder we have only one file which is index.html. Since we are using RequireJS, the content of the index.html file is quite simple. In it we have the stylesheet and the base script of TodoMVC (downloaded from the TodoMVC project page) , and some basic layouting of our application (a section and a footer). The most important element is the last script tag

<script data-main="js/main" src="lib/require.js"></script>

It tells RequireJS to load a file named main.js from the js directory using the RequireJS script file require.js which exists in the lib directory. More details on the structure of main.js in the next section.

The lib directory contains all the third party libraries we are using. At this point we described Backbone, Underscore and jQuery which are dependencies of Backbone, RequireJS, and Marionette.

The js directory contains all the mini-apps (sometimes called module-apps) directories and all the JavaScript which is common in all the mini-apps in the common directory. For example, todo-list is a mini-app and it contains folders for the two main Backbone constructs: models and views (we regard Backbone collections as models but it is possible to create a separate directory for the collection files). We discuss the folder structure of a single mini-app in detail in one of the next sections.

The assets directory is a requirement of the TodoMVC specification and is not important for this discussion. It contains the resources needed by the specification to support the common look and feel of the TodoMVC application.


As we mention above, main.js is the first file that is loaded by the browser when our application starts. This file contains the configuration for RequireJS in a requirejs.config({}) call. The first things we define are the baseUrl , the shims, and the paths of the libs our application is using. The baseUrl property determines the base path for all the non-relative paths used to require modules. The shims property defines the order of loading for all the non-AMD modules we are loading. In our application we load jQuery, Underscore, Backbone, and Marionette from the lib directory. RequireJS makes sure that all the shims are loaded in the old fashioned script tag style in your final HTML page and that those script tags appear in a correct order (based on the dependencies in the code).

After the call to requirejs.config() we delegate our call to the loader. This step is not mandatory but we wanted to keep main.js clean of any logic. The call to the loader simply calls loader.start():

 require(['js/loader', function (loader) {

The loading process and loader.js

Before we look at the code of loader.js we describe the outline of the loading process. RequireJS needs to figure out the loading order of the various modules. Clearly the Marionette Application must be loaded first and it must not depend on any other module. This allows us to pass the App object to any other module that uses the addInitializer functionality before the loader starts the Application. We must also load and start all the controllers and all the routers before Backbone.history.start() is called. Once the App is loaded it can be passed down to all subsequently loaded modules (specifically controllers) and then started using App.start(). The full loading order is therefore (--> means "loads" or "depends on"):

    main.js       --> loader.js
    loader.js     --> app.js
    loader.js     --> controller.js 
    loader.js     --> router.js
    controller.js --> app.js (dosn't load anything since loader.js already loaded the app)
    controller.js --> some_view.js
    controller.js --> some_model.js
    some_view.js  --> template

mini-app internal code structure

Each mini-app represents a closed functionality of the application. The external API of the mini-app is the Marionette.Controller which is responsible to creating and loading all the models and views. The controller serves as an internal and external events vent for all the events within the miniapp. The mini-app directory contains the views and the models directories. The views directory contains all the modules which return the constructor function for any kind of Marionette view (e.g. Marionette.CompositeView). Inside the views directory we have the templates directory which contains all the templates consumed by the views of the mini-app. The models directory contains all the modules which return either Backbone.Model or Backbone.Collection constructor functions. It is also possible to separate each construct to its own folder but since they are so tightly coupled together we decided to put everything in one folder. (Note: A constructor function is what you get when you call Backbone.XXX.extend({}).)

Application Flow

Loading your mini-apps

We start the discussion of this section after the loader loads all the controllers of all the mini-apps. At this point the loader calls the controller.start() function on each controller which, in turn, returns a jQuery promise. The loader waits until all the promises are resolved and then starts the application. We dive into the load flow of the todo-list controller as this flow should be similar for all controllers. In the controller start function we do two main things:

  • The collections are created and loaded via the fetch function.

  • The topmost views are loaded via the async require statement.

  • The router of the mini-app is loaded via the async require statement.

After the async loads are done:

  • An instance of each view is created and shown in the appropriate Marionette.Region.

  • We register on the promise of the fetch function returned from the collection so that when it has been resolved we update the todoList.

todosCollection = new TodoItemCollection(),
todoPromise = todosCollection.fetch();
todoPromise.done(function () {
    _this.vent.trigger("todosUpdated", { collection: todosCollection });

An important thing to notice is that we don't wait for the collection to finish loading before we show the views. We want an empty UI (or some "loading..." screen) to be visible to the user and when the collection is loaded it will fill the elements in the UI.

Views and Sub-Views

In the previous section the controller loaded the topmost views (usually only the main Marionette.Layout is loaded in this stage). Each view of type Marionette.Layout, Marionette.CompositeView, and Marionette.CollectionView (i.e. the parent view) is responsible for loading all the sub-views (sub-views are the views that are embedded within the parent view) through the define statement of the module where the parent view is defined. In our example the controller loads the MainLayoutView into the main region of the page called section. MainLayoutView, as the name suggests, extends Marionette.Layout and therefore it loads all the embedded views: main_header_view, main_content_view, and main_footer_view.

Each view of type Marionette.Layout, Marionette.CompositeView, and Marionette.ItemView is responsible for loading its template. We will discuss views in details in the next section but for now we can see that the template is loaded trough the define statement of the parent view much like the sub-views.

    define(['marionette', 'hbs!./templates/main_layout', './main_header_view', 
            './main_content_view', './main_footer_view', './../controller'],
        function (Marionette, layoutTemplate, MainHeaderView, 
                  MainContentView, MainFooterView, controller) {

After the parent view is loaded it is rendered by Backbone using the render function. Marionette allows you to perform some logic right after the view is rendered by calling onRender function (similar to what you would expect in any .NET control). In the onRender function the el of your view is ready and you are able to show/render all the sub-views. In the case of our MainLayoutView we create and show three sub-views into the internal regions:

    onRender: function () { MainHeaderView()); MainContentView({ collection: this.options.todosCollection })); MainFooterView());

We can see that the TodoItemCollection that was instantiated in the controller is passed on to the appropriate sub-view. The MainContentView is of type Marionette.CompositeView and it is responsible for maintaining the correct state of the collection within our application. This is the place to mentions the two possible approaches of instantiating and passing your collections and models. The first approach is the one we take in this application. The collection is instantiated by the controller and passed on only to the view which is responsible for all the collection operations (i.e. MainContentView). To convey various events of the collection we use the controller as the EventBus through the vent property which is an extension of Backbone.Events object.

     var Controller = Marionette.Controller.extend({
        vent: _.extend({}, Backbone.Events)

This approach is more of the OO style because the collection remains 'isolated' from all other objects except the one which encapsulates it. The same approach works for models encapsulated in Marionette.ItemView or Marionette.CompositeView. The second approach is to instantiate the collection (or model) in the controller but pass it to any view or module that needs to track the events of that collection. In this approach the collection itself is the EventBus (Backbone.Collection already extends Backbone.Events so the collection instance itself is the vent). The benefit of this approach is that every module has an easy access to the collection and it does not need to route the logic through a common event bus. For our project we took the first approach.

Template Loading

One of the key aspects in a Backbone application is managing templates for the views. For our implementation of TodoMVC we selected Handlebars as our templating framework. The nice thing about this selection is that there is a plugin which allows RequireJS to load and compile the templates as if they were regular script files. The plugin is developed by Alex Slexton and is simply named require-handlebars-plugin. From now on we refer to the plugin by the friendly name hbs. To use the plugin we added

    hbs: { disableI18n: true }

to the requirejs.config() call so that the plugin internal i18n interpreter isn't loaded. We also added the plugin main file hbs.js and all of its dependencies (i18nprecompile.js and json2.js) to the lib directory and into the path attribute of requirejs.config(). Now we are ready to load the templates from our views. To load a template we created our template files with the extension .hbs within the templates directory. Let's revisit the module definition of the MainLayoutView. The second parameter of the define statement ('hbs!./templates/main_layout') tells the hbs plugin to load the file main_layout.hbs in the relative directory templates. The ! notation tells RequireJS to use the plugin for loading the file. In the appropriate variable of the define callback (i.e. layoutTemplate) we get the compiled Handlebars template. This is very good because the template property of a Marionette view expects you to set it to a compiled template. Voila! Your template is now loaded just as needed and cached by the RequireJS internal caching mechanism.

Application Logic - Deep Dive

In this section we will deep dive into the implementation of the various views, collections, and models of our application. We will explain the main logic and the considerations we took to make the logic as it is implemented.


The following image depicts the outline of our application in terms of views/templates: Views Layout

The application (Marionette.Application) splits the viewable area into two regions. The footer region contains the footer_view view (Marionette.ItemView) which displays three lines of text. This view is not important to the discussion. The main section region which contains the main_layout (Marionette.Layout) view. main_layout in itself splits the main area into three sub-views.

main_header_view (Marionette.ItemView) is responsible for the input line that adds a new 'todo'. Note that the main_header_view view is the only visible view when there are no todos in the list. This is enforced by the logic within main_layout_view according to the specifications of the TodoMVC application. The function updateSubViews is responsible for showing and hiding the footer and the content according to the state of the todos collection. _isDataVisible is an internal property which keeps track of the current visibility of these two regions.

            updateSubViews: function (data) {
                if (data.collection.length === 0 && this._isDataVisible) {
                    this._isDataVisible = false;

                if (!this._isDataVisible && data.collection.length > 0) {
                    this._isDataVisible = true;

main_content_view (Marionette.CompositeView) is responsible for rendering the entire todos list. The model part of the composite view (the part that doesn't contain the collection) contains only the toggle button which located to the left of the main_header_view input box. As we mention above the main_content_view is responsible for performing all the operations on the collection and triggering the appropriate events on the EventBus. We decided to trigger the events from the view and not directly from the collection to allow some decoupling between the collection and the controller. In fact the guideline we used is that the collections/models are not dependant on anything (except for inner dependencies required by Marionette). The collection part of main_content_view contains a list which renders todo_item_view for each element in the collection creating the list in the UI.

main_footer_view (Marionette.ItemView) is responsible for displaying the footer of the list. The footer contains information regarding the completion state of the todos in the list and allows you to filter the list based on the completion state or remove all the completed todos. What is important about this view is that it is strongly coupled (in terms of data) with the todos collection which is controlled by the main_content_view. Therefore, the updates in this view are done through events on the EventBus. In fact, we have only one event, todosUpdated, which is sent when the todos collection is updated. The event passes just enough information to allow the main_footer_view to update all the controls on the UI without allowing the main_footer_view perform operations on the collection itself.

todo_item_view (Marionette.ItemView) is a single line in the list of todos. This view is instantiated by main_content_view when it renders its collection. This view holds the TodoItem model and handles all the CRUD operations of a single todo item.


As we state above, main_header_view is responsible for entering text for a new todo. Note that it is not responsible for creating a new todo, this view doesn't even "know" what a todo it. It's job is simple, wait for the user to input some text in the text box. If the return key is pressed trigger an event that the user submitted some text on the EventBus and clear the input. The triggered event is todoTextReady.

       inputKeypress: function(e) {
         var ENTER_KEY = 13,
            todoText = this.ui.input.val().trim();

         if (e.which === ENTER_KEY && todoText) {
            controller.vent.trigger('todoTextReady', todoText);


Before we jump into the details of main_content_view lets review the inner workings of todo_item_view. This view is responsible for all the operations done on a single todo line. The user can perform one of three operations on this view:

  • A click on the 'V' (left side) completes/uncompletes the todo. The click event handler delegates to a function called finishClicked which sets the isFinished model property and triggers the finishChanged event on the collection, passing the changed model. We don't make use of the automatically triggered change:isFinished event to be able to aggregate all the changed models into one event in the case where more than one model's isFinished property has changed.
        finishClicked: function (e) {
            var finishState =;
            this.model.set('isFinished', finishState);
            this.model.collection.trigger('finishChanged', [this.model], this.model.collection);
  • A click on the 'X' (right side) removes the todo from the list. This is quite straight forward as we call the model.destroy() function and let Backbone do all the magic.
        deleteClicked: function () {
  • A double-click on the todo text enters the edit mode where the label showing the todo text is replaced by a text box which allows you to enter new text (even if this todo is completed). To show the text box we simple change some pre-defined css classes. The interesting part is the handling of the focusout event on the text box which, essentially, submits the edited text. In the event handler we verify that the given text is not empty and decide whether we want to save or delete it. The TodoMVC specifications dictate that we delete the todo if the edited text is empty.
        editFocusout: function () {
            var todoText = this.ui.input.val().trim();
            if (todoText) {
                this.model.set('todoText', todoText);
            } else {

Note that after we change the model property todoText we re-render the view. Rendering the view triggers the onRender function where we manipulate the UI to comply with the model properties. This is the correct place to manipulate the UI since the view may be re-rendered from various execution paths (e.g. when the user clicks the toggle all completed button) and it must always be rendered correctly.

         onRender: function () {
            var finishState = this.model.get('isFinished');


            if (controller.displayMode === && this.model.get('isFinished')) {

            if (controller.displayMode === controller.displayModes.completed && !this.model.get('isFinished')) {

            if (finishState) {
            } else {
            this.ui.finishedCheckbox.prop('checked', finishState);


main_content_view is where all the collection operations happen. The only UI element which is controlled by the main_content_view is the toggle-all button (on the left of the header text box) which toggles the completion state of all the todos in the list. The most important thing to note here is the regular separation between the operation and the UI change. For example, to add a new model to the collection we listen to the todoTextReady event on the EventBus. When the event is triggered we add a new model to the collection with the provided text (we don't make any updates yet). This triggers the add event on the collection. We listen to the add collection event and when it is triggered we trigger the todosUpdated event on the EventBus. We listen to the todosUpdated event on the EventBus and when that is triggered we update the check state of the toggle-all button according to the TodoMVC specifications.

            initialize: function () {
                controller.vent.on('todoTextReady', this.addTodo, this);
                controller.vent.on('todosUpdated', this.todosUpdated, this);

            addTodo: function (todoText) {
                this.collection.push({ todoText: todoText });

            collectionEvents: {
                'add': 'todoAdded',

            todosUpdated: function (data) {
                var hasUnfinished = data.collection.some(function (item) {
                    return !item.get('isFinished');

                this.ui.toggleAll.prop('checked', !hasUnfinished);

A similar flow happens when a model is removed from the collection or when a collection is reset.


The main_footer_view performs three main functions:

  • Displaying the number of incomplete todos in the todos list. To this end the view listens to the todosUpdated event on the EventBus. When this event is triggered the event argument is an array of all the models (although this can be optimized to contain a separate list of the changed models) from which the view can easily calculate the number of incomplete items by reading the isFinished property of each model. The view keeps two variables to track the current state: unfinishedItemsCount the number of incomplete todos and finishedItemsCount the number of complete todos. When the view is rendered the template expects these two properties to be passed in to the template rendering function. Fortunately for us, Marionette handles this case by allowing the view to override a function called SerializeData. The return value of this function is passed to the template rendering function allowing us to easily pass the two stored values. We could use a specialized model to store the two values for consistency but we felt that we rather not create transient models and therefore decided to store the two properties directly on the view.
        serializeData: function () {
            return {
                unfinishedItemsCount: this.unfinishedItemsCount,
                finishedItemsCount: this.finishedItemsCount,
                itemString: this.unfinishedItemsCount === 1 ? 'item' : 'items'

        updateData: function (data) {
            var count = 0;
            data.collection.each(function (todo) {
                if (!todo.get('isFinished')) {
                    count += 1;

            this.unfinishedItemsCount = count;
            this.finishedItemsCount = data.collection.length - count;

  • Allowing the user to filter between three views. The first view called "All" shows all the todos, the second called "Active" shows only the incomplete todos, and finally the third view called "Completed" shows only the completed todos which were not removed from the list. Since TodoMVC is a single page application we are changing the filtering through internal routing using Backbone.Router or more specifically Marionette.AppRouter which extends Backbone.Router and adds the controller connection to it. We capture three routes as defined by the TodoMVC specifications and route them to the appropriate functions in the controller.
    var MainRouter = Marionette.AppRouter.extend({
        appRoutes: {
            '': 'displayModeAll',
            'active': 'displayModeActive',
            'completed': 'displayModeCompleted'
        controller: controller

The tree functions implemented on the controller basically do the same thing: set the current display mode on the controller and trigger the displayModeChanged event on the EventBus.

            displayModes: {
                all: 'All',
                active: 'Active',
                completed: 'Completed'

            displayModeAll: function () {
                this.displayMode = this.displayModes.all;
                this.vent.trigger('displayModeChanged', this.displayMode);

            displayModeActive: function () {
                this.displayMode =;
                this.vent.trigger('displayModeChanged', this.displayMode);

            displayModeCompleted: function () {
                this.displayMode = this.displayModes.completed;
                this.vent.trigger('displayModeChanged', this.displayMode);

The event is captured by the main_footer_view itself to update the selected caption of the appropriate view and by the todo_item_view to update the visibility of each item based on the current display mode.

        onRender: function () {
            var finishState = this.model.get('isFinished');


            if (controller.displayMode === && finishState) {

            if (controller.displayMode === controller.displayModes.completed && !finishState) {

Note how every view retains its responsibility and handles its own assets based on events on the EventBus. This is a key element in good separation of concerns and modular application design.

  • Allowing the user to remove all the completed todos with a single click. When there is one or more complete todo in the todos list, the main_footer_view displays a button which displays the number of complete todos. Clicking the button removes the todos from the list. At this point you should be able to figure out the implementation of this functionality. Clicking the "Clear completed" button does not affect the todos list directly because it is not the responsibility of the main_footer_view to do so. Instead, it triggers a clearCompleted event on the EventBus. The main_content_view is listening to this event and removes all the todos from the todos collection. Since the collection has changed, main_content_view triggers the todosUpdated event on the EventBus. The main_footer_view listens to the todosUpdated event and retenders itself according to the todos collection state, in this case hiding the "Clear completed" button.

Integrating with Backbone.localStorage

The specifications of TodoMVC require us to retain the todos between browser refreshes. This means that we must store the todos somewhere. Fortunately Backbone architecture allows extensions and mixins. We use the Backbone.localStorage extension to store the todos in the browser local storage. First we add the definition for Backbone.localStorage to the shims and the paths properties in main.js. Once we do that, the plugin overrides a key function in Backbone called Backbone.sync. This function is responsible for the saving and loading of Backbone models. The new Backbone.sync function checks if the model (or collection) has a property called localStorage. If this property doesn't exist then it the original Backbone.sync function is called. If this property exists then the implementation of the plugin is called. The implementation of the plugin saves and loads each model to and from the local storage of the browser. Adding the plugin and setting the property makes our entire implementation work with the local storage without additional effort.

    var TodoItemCollection = Backbone.Collection.extend({
        model: TodoItem,
        localStorage: new Backbone.LocalStorage("TodoItemCollection")

We pass a key name to the Backbone.LocalStorage constructor function. It uses that key to store our data in a partition of the local storage with that name.

Optimizing with r.js

RequireJS provides an optimization tool called r.js which optimizes the code for production use. We give a small taste of the tool's capabilities by compiling all of the application code into one file called main-built.js. r.js is capable of doing various operation to optimize your code, see full documentation here. To use r.js you need to install it using Node.js package manager (a.k.a. npm). Make sure you have Node.js installed on your system, open a command line window in the project directory and write npm install. After the installation is over you should have a directory called node_modules in the project directory. Do not commit or change files from that directory in your repository. To run r.js you need a configuration file that sets all the relevant options. We provide such a file for our project called The configuration file resembles the main.js file of our project but has some optimization related properties defined. Please refer to the r.js documentation for full information on all the properties or use this annotated source.

To build main-built.js run the following command from the project directory:

node node_modules\requirejs\bin\r.js -o

The optimizer finds all the dependencies and the loading order and creates a single js file that contains a concatenated version of all the js files that would load on your page. You can browse to index2.html to see that the resulting page behaves the same as the original page but with a significant performance boost on high latency connections.


Setting the environment

Solid Testing suite is a crucial component of building a robust JavaScript application. This is especially important due to JavaScript dynamic nature and lack of compiler. There are many alternatives to create and run tests. Mocha and [Jasmine]a( are the two most popular unit tests frameworks. Both are quite similar in functionality and use rSpec BDD syntax (describe and it) with a large and active community. We have implemented our tests suite using Mocha; it runs over node.js, therefore very easy to install and update using node.js' npm package manager.

It is a common practice to devide that basic tests into two layers: unit tests and functional tests. Unit test tests specific part of the code in isolation (e.g. non-trivial function). Functional test tests the functional behavior of the application, usually from the prespective of the end-user. In this sample application we have only implemented the functional tests.

The tests are running in the Zombie.js browser which is a headless browser. It means that it doesn't have a GUI and it is only useful for testing. The advantages are that the tests execute fast and can run in environment that doesn't have GUI.(such as CI server). The disadvantages are that it might be harder to debug the tests, and that it does not support cross-browser testing. It is important to note that currently, Zombie.js is currently not supported in Windows.

To start using Mocha, you must install Node.js. Next, install the required node.js packages by typing:

    npm install mocha zombie chai

chai is a BDD assertion library for node and it allows you to easily declare the conditions (sometimes called expectations) that can make your test pass or fail (verification points). If one of the declated expectations fails, then the entire mocha test fails. You can add a file called mocha.opts and place it in the tests directory. Add the following line to the mocha.opts file to set the reporter output for the tests runner.

--reporter spec

Add the following line to the mocha.opts file to tell Mocha to do a recursive search for test directories and execute the code in them.


The content of the mocha.opts file should be:

		--reporter spec

Writing the tests

The tests are implemented in the /test/functional.js file. In BDD style tests, you choose a module as the test target. Then, you describe the module's functionality using common language. For example, we describe our TodoMVC application as follows:

  • it allows to add a new todo item
  • it shows how many items left to be done
  • can mark an item as completed

In the test directory we add the following skeleton implementation.

		describe("Todo application", function(){
			it("allows to add a new todo item");
			it("shows how many items left to be done")
			it("can mark an item as completed")

To run the tests, type mocha in the console. At this point, all the tests are presented as "pending" since we did not provide an implementation. The implementation is done by manipulating the Zombie.js browser object. Mocha provides a beforeEach() function which is executed before each test. In this function we create a new browser instance; we do not reuse the browser to prevent interactions between different tests. We then use the visit method of the browser to navigate to a specific URL.

A very important pitfall in JavaScript acceptance testing is the wait issue. In a traditional web page, the page DOM is constructed from an HTML file. In this case, when the code is in the $(window).ready() function, all the DOM is guaranteed to be ready. However, in a single page JavaScript application, the DOM is built dynamically. This poses a difficulty for testing tools as they cannot know when the page was fully loaded. If the testing tool tries to execute code that interacts with the DOM before it is fully loaded, the tests fails. Consider our TodoMVC application; when the application page is first loaded is doesn't contain any UI elements. Only then, RequireJS dynamically loads the main_header_view which allows adding new todos. If we consider our first test (adding new todo item); it will probably fail, since the text box in which the user enters the text for the todo item doesn't exist yet. To handle this case we have to explicitly wait for some condition to become satisfied. In the tests, we check that an item with id #new-todo exists. We use it as indication that the JavaScript code executed and created the relevant DOM elements.

		function waitForPageLoadEnd(callback) {
			that.browser.wait(function() {
				return that.browser.querySelector("#new-todo");
			}, callback);

A Mocha test function takes a done function as an argument. It is the responsibility of the user to call this function when the test finishes its execution. This is required due to Mocha's async nature. The done function notifies the Mocha framework that all the relevant code was executed and that it can now report the test as complete. If the done function is not invoked in a timely manner, Mocha reports the test as failed due to timeout.

		it("allows to add a new todo item", function(done) {
			addItem("first item", function() {
				expect(that.browser.text("")).to.eql("first item");