The purpose of this little app is to demonstrate one way of using Rendr to build a web app that runs on both the client and the server.
First, make sure to have Node >= 0.8.0 installed on your system. Also, make sure to have grunt-cli installed globally.
$ npm install -g grunt-cli
If you see an error on startup that looks like this, then you may need to un-install a global copy of grunt:
$ npm uninstall -g grunt
Clone this repo to a local directory and run npm install to install dependencies:
$ git clone [email protected]:airbnb/rendr-app-template.git
$ cd rendr-app-template
$ npm install
Then, use grunt server to start up the web server. Grunt will recompile and restart the server when files change.
$ grunt server
Running "bgShell:runNode" (bgShell) task
Running "handlebars:compile" (handlebars) task
File "app/templates/compiledTemplates.js" created.
Running "rendr_stitch:compile" (rendr_stitch) task
4 Apr 09:58:02 - [nodemon] v0.7.2
4 Apr 09:58:02 - [nodemon] watching: /Users/spike1/code/rendr-app-template
4 Apr 09:58:02 - [nodemon] starting `node index.js`
4 Apr 09:58:02 - [nodemon] reading ignore list
File "public/mergedAssets.js" created.
Running "stylus:compile" (stylus) task
File public/styles.css created.
server pid 87338 listening on port 3030 in development mode
Running "watch" task
Waiting...
Now, pull up the app in your web browser. It defaults to port 3030.
$ open http://localhost:3030
You can choose a different port by passing the PORT environment variable:
$ PORT=80 grunt server
GitHub rate limits unauthenticated requests to its public API to 60 requests per hour per IP. This should be enough for just playing with the sample app, but if you pull it down and start developing off it you may run up against the rate limit.
If this happens to you, you can supply your GitHub creds for HTTP Basic Auth using the BASIC_AUTH environment variable. Be very, very careful with this! It means you will be typing your GitHub credentials in plain text, which will be saved to your Bash history and may be intercepted by other programs. If you do this, immediately change your password before and afterwards. This should only be necessary if you're developing on the app and need to keep refreshing the page.
$ BASIC_AUTH=githubusername:githubpassword grunt server
You've been warned. Your best bet may be to alter the project to read from your favorite RESTful API.
It's worthwhile to read the first blog post, which has some background on Rendr and its raison d'être.
This basic Rendr app looks like a hybrid between a standard client-side MVC Backbone.js app and an Express app, with a little Rails convention thrown in.
Check out the directory structure:
|- app/
|--- collections/
|--- controllers/
|--- models/
|--- templates/
|--- views/
|--- app.js
|--- router.js
|--- routes.js
|- assets/
|- config/
|- public/
|- server/
Note: I want to stress that this is just one way to build an app using Rendr. I hope it can evolve to support a number of different app configurations, with the shared premise that the components should be able to run on either side of the wire. For example, the full-on client-side MVC model isn't appropriate for all types of apps. Sometimes it's more appropriate to load HTML fragments over the wire, also known as PJAX. Rendr apps should be able to support this as well.
Node.js uses the CommonJS module pattern, and using a tool called Stitch, we can emulate it in the browser. This looks familiar in Node.js:
var User = require('app/models/user');Using Stitch, we can use the same require() function in the browser. This allows us to focus on application logic, not packaging modules separately for client and server.
In Node.js, you can also use require() to load submodules within NPM models. For example, we could load Rendr's base view in order to extend it to create a view for our app.
var BaseView = require('rendr/shared/base/view');Because of a trick in the way we do Stitch packaging, this module path works in the browser as well.
// app/routes.js
module.exports = function(match) {
match('', 'home#index');
match('repos', 'repos#index');
match('repos/:owner/:name', 'repos#show');
match('users' , 'users#index');
match('users/:login', 'users#show');
};A controller is a simple JavaScript object, where each property is a controller action. Keep in mind that controllers are executed on both the client and the server. Thus, they are an abstraction whose sole responsibility is to specify which data is needed to render the view, and which view to render.
On the server, controllers are executed in response to a request to the Express server, and are used to render the initial page of HTML. On the client, controllers are executed in response to pushState events as the user navigates the app.
Here is a very simple controller:
// app/controllers/home_controller.js
module.exports = {
index: function(params, callback) {
callback(null, 'home_index_view');
}
};Every action gets called with two arguments: params and callback. The params object contains both route params and query string params. callback is called to kick off view rendering. It has this signature:
function(err, viewName, viewData) {}Following the Node.js convention, the first argument to the callback is err. We'll pass null here because we're not fetching any data, but if we were, that's how we'd communicate a fetching error.
This is a string identifier of a view, used by the router to find the view class, i.e.:
require('app/views/' + viewName);An object to pass to the view constructor. This is how we pass data to the view.
All our index action above really does is specify a view class. This is the simple case -- no data fetching, just synchronous view rendering.
It gets more interesting when we decide to fetch some data. Check out the repos_controller below:
// app/controllers/repos_controller.js
module.exports = {
// ...
show: function(params, callback) {
var spec = {
model: {model: 'Repo', params: params}
};
this.app.fetch(spec, function(err, result) {
callback(err, 'repos_show_view', result);
});
}
};You see here that we call this.app.fetch() to fetch our Repo model. Our controller actions are executed in the context of the router, so we have a few properties and methods available, one of which is this.app. This is the instance of our application's App context, which is a sublcass of rendr/base/app, which itself is a subclass of Backbone.Model. You'll see that we inject app into every model, view, collection, and controller; this is how we maintain app context throughout our app.
You see here that we call callback with the err that comes from this.app.fetch(), the view class name, and the result of the fetch. result in this case is an object with a single model property, which is our instance of the Repo model.
this.app.fetch() does a few nice things for us; it fetches models or collections in parallel, handles errors, does caching, and most importantly, provides a way to boostrap the data fetched on the server in a way that is accessible by the client-side on first render.
A Rendr view is a subclass of Backbone.View with some additional methods added to support client-server rendering, plus methods that make it easier to manage the view lifecycle.
Creating your own view should look familiar if you've used Backbone:
// app/views/home_index_view.js
var BaseView = require('./base_view');
module.exports = BaseView.extend({
className: 'home_index_view',
events: {
'click p': 'handleClick',
},
handleClick: function() {…}
});
module.exports.id = 'HomeIndexView';You can add className, tagName, events, and all of the other Backbone.View properties you know and love.
We set the property id on the view constructor to aid in the view hydration process. More on that later.
Our views, just like all of the code in the app/ directory, are executed in both the client and the server, but of course certain behaviors are only relevant in the client. The events hash is ignored by the server, as well as any DOM-related event handlers.
Notice there's no render() method or template property specified in the view. The philosophy here is that sensible defaults and convention over configuration should allow you to skip all the typical boilerplate when creating views. The render() method should be the same for all your views; all it does is mash up the template with some data to generate HTML, and insert that HTML into the DOM element.
Now, because we're not using a DOM to render our views, we must make sure that the view returns all its HTML as a string. On the server, view.getHtml() is called, which returns the view's outer HTML, including wrapper element. This is then handed to Express, which wraps the page with a layout and sends the full HTML page to the client. Behind the scenes, view.getHtml() calls view.getInnerHtml() for the inner HTML of the view, not including wrapping element, and then constructs the wrapping element based on the tagName, className, etc. properties, and the key-value pairs of HTML attributes returned by view.getAttributes(), which allows you to pass custom attributes to the outer element.
On the client, view.render() is called, which updates the view's DOM element with the HTML returned from view.getInnerHtml(). By default, Backbone will create the wrapper DOM element on its own. We make sure to also set any custom HTML attributes in view.getAttributes() on the element.
A common need is to run some initialization code that touches the DOM after render, for things like jQuery sliders, special event handling, etc. Rather than overriding the render() method, use postRender(). The postRender() method is executed for every view once after rending, including after initial pageload.
// app/views/home_index_view.js
var BaseView = require('./base_view');
module.exports = BaseView.extend({
className: 'home_index_view',
postRender: function() {
this.$('.slider').slider();
}
});
module.exports.id = 'HomeIndexView';If you have a need to customize the way your views generate HTML, there are a few specific methods you can override.
By default, getTemplateName() returns the underscored version of the view constructor's id property; so in our case, home_index_view. It will also look for options.template_name, which is useful for initialing views to use a certain template. The view will look in app/templates for the value returned by this function.
If getTemplateName() isn't enough, you can override getTemplate() to return a function that takes a single data argument and returns HTML:
function(data) {
...
return html;
}This HTML is used to populate the view's inner HTML; that is, not including the wrapper element, which you can specify on the view itself using tagName, className, and id.
If you're building some sort of composite view that doesn't utilize a simple template, override getInnerHtml(). This is useful for tabbed views, collection views, etc.
You probably shouldn't ever need to override this; by default it just combines the HTML returned by getInnerHtml() and the HTML attributes returned by getAttributes() to produce an outer HTML string.
Rendr provides a Handlebars helper {{view}} that allows you to declaratively nest your views, creating a view hierarchy that you can traverse in your JavaScript. Check out app/templates/users_show_view.hbs and app/views/users_show_view.js for an example:
<!-- app/templates/users_show_view.hbs -->
...
<div class="span6">
{{view "user_repos_view" collection=repos}}
</div>
<div class="span6">
...
</div>You see that we use the {{view}} helper with an argument that indicates which view to be rendered. We can pass data into the view using Handlebars' hash arguments. Anything you pass as hash arguments will be pass to the subview's constructor and be accessible as this.options within the subview. There are a few special options you can pass to a view: model or collection can be used to directly pass a model or collection instance to a subview. The options model_name + model_id or collection_name + collection_params can be used in conjunction with lazy="true" to lazily fetch models or collections; more on that later.
Now, from within the users_show_view view, we can access any child views using the this.childViews array. A good way to debug and get a feel for this in the browser is to drill down into the global App property, which is your instance of BaseApp. From App you can access other parts of your application. App.router is your instance of ClientRouter, and it has a number of properties that you can inspect. One of these is App.router.currentView, which will always point to the current main view for a page. For example, if you are viewing wycats' page in our app, http://localhost:3030/users/wycats, currentView will be an instance of users_show_view:
App.router.currentView
=> child {render: function, cid: "view434", model: child, options: Object, $el: p.fn.p.init[1]…}
From there, we can find our child user_repos_view view:
App.router.currentView.childViews
=> [child]
App.router.currentView.childViews[0]
=> child {render: function, cid: "view436", options: Object, $el: p.fn.p.init[1], el: div.user_repos_view…}
Check out its collection property, which is the instance of Repos which we fetched in the controller and passed down in the {{view}} helper:
App.router.currentView.childViews[0].collection
=> child {options: Object, app: child, params: Object, meta: Object, length: 30…}
You can nest subviews ad infinitum. Our user_repos_view has an empty childViews array now, but we could add some subviews if we found it useful for organizing our codebase, or keeping things DRY.
App.router.currentView.childViews[0].childViews
=> []
Views also have a parentView property, which will be non-null unless they are a top-level view.
App.router.currentView.childViews[0].parentView === App.router.currentView
=> true
App.router.currentView.parentView
=> null
So far, our users#show action pulls down both a User model and a Repos collection for that model. If we were to navigate from users#index to users#show, we already have that user model cached in memory (because we fetched it in order to render the list), but we have to make a roundtrip to the server to fetch the Repos, which aren't part of the User attributes. This means that instead of immediately rendering the users_show_view view, we wait for the Repos API call to finish. But what if instead we want to lazy-load the Repos so we can render that view immediately for a better user experience?
We can achieve this by lazy-loading models or collections in our subviews. Check out the users#show_lazy action, which demonstrates this approach:
// app/controllers/users_controller.js
module.exports = {
// ...
show_lazy: function(params, callback) {
var spec = {
model: {model: 'User', params: params}
};
this.app.fetch(spec, function(err, result) {
if (err) return callback(err);
// Extend the hash of options we pass to the view's constructor
// to include the `template_name` option, which will be used
// to look up the template file. This is a convenience so we
// don't have to create a separate view class.
_.extend(result, {
template_name: 'users_show_lazy_view'
});
callback(err, 'users_show_view', result);
});
}
}The first thing to notice is that in our fetch spec, we only specify the User model, leaving out the Repos collection. Then, we tell the view to use a different template than the default. We do this by passing in a template_name property to the view's options, which is passed to its constructor. We extend the result object to have this; the third argument to our callback is an object that's passed to the view's constructor. We could have also created a separate view class in JavaScript for this, to match our new template.
Here's the users_show_lazy_view template, abbreviated:
<!-- app/templates/users_show_lazy_view.hbs -->
...
<div class="span6">
{{view "user_repos_view" collection_name="Repos" param_name="login" param_value=login lazy="true"}}
</div>
<div class="span6">
...
</div>So, the only difference to our original users_show_view template is that instead of passing collection=repos to our user_repos_view subview, we pass collection_name="Repos" param_name="login" param_value=login lazy="true". When fetching collections, we specify params, which are used to fetch and cache the models for that collection. We quote all of these arguments except for param_value=login; quoted arguments are passed in as string literals, and unquoted arguments are references to variables that are available in the current Handlebars scope. login is one of the attributes of a User model, which gets passed into the template. The lazy="true" tells the view that it needs to fetch (or find a cached version of) the specified model or collection.
We can see this at play in our app if we add a route in our app/routes.js file that routes users_lazy/:login to users#show_lazy, and change our app/templates/users_index_view.hbs to link to /users_lazy/{{login}}.
Now, if we click from the list of users on users#index, you'll see the page gets rendered immediately, and the repos are rendered once the API call finishes. If you click back and forward in your browser, you see it's cached.
So far, Rendr just supports Handlebars templates, but it should be possible to make this interchangeable. For now, place your templates in app/templates with a name that matches the underscorized view's identifier and file extension of .hbs. So, the view with an identifier of HomeIndexView will look for a template at app/templates/home_index_view.hbs.
In this example we use Grunt to manage asset compilation. We compile JavaScripts using Stitch and stylesheets using Stylus. Check out Gruntfile.js in the root directory of this repo for details.
MIT