Moving to a JavaScript Application Architecture

Preface

From 2003 to 2011 I became a pretty decent ActionScript developer. Starting with single frame movies then to fully structured MVC applications.

With my team, I’d spent a lot of time looking at two powerful frameworks   Cairngorm and RobotLegs. I really liked RobotLegs, and being my first introduction to dependency injection – was really intrigued.

I ended up writing my own framework based on RobotLegs called Nori. It was pretty slick – mediators, contexts, mapping, Signals, views and tons of interfaces (any DI system has). I only used it for a few projects before a career change and a transition to consulting and design / prototyping.

Fast forward a few years and I’m trying to break into coding again with JavaScript. I’d done a fair bit of JS dev in 2000-3, but really hadn’t paid much attention to it since then.

Holy shit!

I have a lot to learn.

Writing my own …

Over the past two years, I’ve spent (too much) time looking for a starting point – Angular, React, ExtJS, Ember, Dojo, etc. I didn’t have a project so I just cycled around in this framework limbo.

Then I found a project. Time to get realz …

My team needed a gallery to showcase our work and provide simple case studies to prospective clients. I’d build something similar to this before, in haXe, that was based on Microsoft’s Pivot experiment. I decided to recreate it but with more of a Pinterest influence.

The front end would be a fluid, responsive JavaScript application that also worked on our primary browser, IE9. It would also scale to mobile phones and be accessible though our corporate GOOD app.

The back end would use SharePoint lists. SharePoint is a decent backend for database challenged teams in big companies and it also comes with a “free” content maintenance system.

I couldn’t waste time researching frameworks anymore, and I had a strong desire to actually learn the language, so I decided to writing my own. My basic task list is/was:

  1. How the hell so I set this up and what build system do I use?
  2. How do I architect this thing? How can I best create class/subclasses and implement the command pattern?
  3. How do I create a global event system (pub/sub) system with command mapping?
  4. What’s cutting edge? FRP, Reactive programming and RxJS!

The project started in November as a part-time/between-other-work project. I’m wrapping up the front end now with the back end still to go. My purpose for this post, is to document my thought process and hopefully solicit some feedback on my approach.

Here’s a break down of what I’ve learned and done over the past few months. I know that it’s nearly useless without screen shots (source code on GitHub) but I’m working on that.

The project source is available on GitHub.

Project setup and build options

No respecting front-end dev writes HTML or CSS anymore. Attending Meet Ups around Charlotte, I saw how people were using Sass and Jade tempting. I have give the credit to the talented Tessa Harmon for introducing these to me at a Charlotte Front-End Developers meeting.

Both of these preprocessors resonated with me as being a much more efficient way to produce CSS and HTML – variables and mixins, oh my! The biggest challenge is working with these in a typically Windows based corporate environment. Lucky, I was able to get a Mac at work. I’ve been developing prototypes with Jade and Sass for a few years at this point with the excellent CodeKit app. On the Windows side, Prepros looks like a great alternative.

I’ve found the indented syntax style of Sass to be the best fit for me. Not only does it go very well with the style of Jade, it’s saves a ton of typing. The fewer braces and semicolons you need to worry about, the better.

I’d seen Grunt but hadn’t used it. Reading articles. like this one from Chris Coyier, really simplified the concepts. It was pretty easy to get it all up and running and enabled a much simpler build process than I was able to achieve with CodeKit. So I have SASS + Jade processing, CSS linting, CSS minification, JSLinting, JS concatenation + minification and live reload all working well.

Being able to work with JavaScript in a modular way, split across many files, is a critical workflow feature. I was used to splitting up my classes in to single files and I was able to preserve that with this process thanks to these tools.

Other libraries

I went back and forth on jQuery. On one hand, the efficiencies gained by utilizing it’s selector and DOM manipulation engines cannot be underestimated, but on the other – I wanted to learn how to do this on my own. In the end, I decided to use it, it just saves too much time.

Having to support IE9 meant that CSS animations were not possible. While developing in Flash, I relied heavily on the awesome GreenSock animation platform and with a JS version available, it was a no-brainer.

I’d struggled a bit with the view styling. Do animations count as “style” (CSS) or functionality (code)? I didn’t have a choice in this situation, but now I’m firmly on the side of code – the control and possibilities presented by JS animation libraries over CSS is too much to ignore.

To handle HTML templating, I compared Mustache.js and Underscore.js. Mustache provided a nice single purpose solution, but Underscore did all I needed it to plus provided a ton of additional functionality that I might need later on. So I went with it.

What’s next

  • Clean up my CSS classes by implementing the BEM system
  • Optimize my Grunt file.
  • Learn more about git and GitHub

OOP / classes, subclasses / structure / commands

I came from a strongly classed development environment. So naturally I wanted to recreate that in JS. But JS doesn’t do classes but it does do objects. I couldn’t use ES6 because of IE9.

Finding way to create namespaces was key and paired with the revealing module pattern, made for a great solution. I implemented the approach Kenneth Truyers documented for my application.

I do have a few global objects, for utility classes, but generally my application structure is:

All of the major parts of the app (bold, above) are constructed like this:

APP.createNameSpace('APP.AppController');
APP.AppController = function () {

// private var
// private methods

return {
// public api
};
}();

These are effectively singletons.

Commands

Both the Cairngorm and RobotLegs frameworks handled the heavy lifting of controller duties in commands. The separation of core functionality outside of the controller made extending the app very simple. Command classes are single responsibility, small and have access to all of the actors in the application.

My commands have a single entry point, an execute() method, that takes one object parameter for passing in data. In my framework, commands are only invoked in response to an event. For every command there is a corresponding event that is published from somewhere in the application. This is discussed in the Events section below.

In my current implementation, they’re synchronous, but I can see needing asynchronous in the future as I’m loading data from sources.

Pattern for subclasses – commands and subviews

To implement commands and sub views, I had to seek an alternate class/object creation pattern. Creating multiple instances of objects with the module pattern isn’t entirely straightforward. Although there are several methods of extending new objects with the properties of an existing object (jQuery’s extend(), etc.), the private methods of the closure aren’t readily accessible in the subclasses. Composition would be been a way to handle this, but I choose a different approach.

I discovered Eric Elliot’s Fluent JS talk on prototypal OO. His StampIt pattern was just what I was after, but I wanted a more accessible approach. I stumbled on an article by Jake Lucas which described a method just like Eric’s, but broke it down so that it was easier for a noob, like myself, to understand.

Using commands as an example, I have this as the abstract command class

APP.AppController.AbstractCommand = {
  state: {},

  methods: {

    app: APP,
    appController: APP.AppController,
    appModel: APP.AppModel,
    appView: APP.AppView,
    eventDispatcher: APP.EventDispatcher,

    execute: function(data) {
      console.log('Abstract command executing with data: '+data);
    }
  },

  closures: []
};

And then I create a new command instance by using a factory method to create a new object and then override the execute method of the new command

APP.createNameSpace('APP.AppController.ItemSelectCommand');
APP.AppController.ItemSelectCommand = APP.AppController.createCommand(APP.AppController.AbstractCommand);
APP.AppController.ItemSelectCommand.execute = function(data) {
  
// new implementation

};

A few disadvantages for this approach:

  1. Alternate coding style compared to the modules
  2. Lots of boilerplate to create the objects

What’s next

  • I’m not entirely satisfied with my approach on commands and subviews. I want to look at other ways to subclass modules.
  • Look at ES6 transpilers so that I can leverage these features now.

Event system (pub/sub)

A publish/subscribe system is key for maintaining loose coupling between components of the application. For my app, I wanted to create a global system similar to the EventDispatcher in AS3.

When I last coded AS3, I was heavily using Robert Penner’s Signals implementation over native events. His approach won me over especially consider the fragility of the “magic strings” naming of event types. They were easier to work with and implement than native events. There is a JS port of it that I initially used, but later decided that it wasn’t worth the additional overhead.

Looking to simplify my approach, I found a great example from Michaël Duerinckx on his blog. It demonstrated exactly what I needed and was straight forward. I created an object APP.Events.Events that stored all of my event strings as constants.

Generally, I’m using events to invoke an instance of a command class – there is 1:1 relationship – the APP.Events.ITEM_SELECT event has a corresponding APP.Controller.Command.ItemSelectCommand associated with it. This mapping is handled by an EventCommandMap class.

I had implemented a command mapping solution in my Nori AS3 framework that was based on Joel Hook’s Signal Command Map for RobotLegs and Omar Gonzalez’s for PureMVC. I greatly simplified my original approach and with some helpful hints from a friend, had it working with my EventDispatcher.

In the controller, commands are mapped like this:

mapCommand(APP.Events.ITEM_SELECT, APP.AppController.ItemSelectCommand);

And trigged like this (‘id’ being wrapped in an object and passed to the command’s execute() method):

_eventDispatcher.publish(APP.Events.ITEM_SELECT, id);

The command map handles determining what command class should be instantiated and executed.

What’s next

FRP!

Future State: FRP, Reactive programming and RxJS

Functional programming is the current hotness in JavaScript. Honestly FRP is still kind of a dark art to me, but it’s clearing up thanks to excellent articles explaining the concepts, libraries like RxJS from Microsoft and tutorials from Netflix.

While I’m not ready to architect a reactive application, I am using Array methods forEach, map, filter, etc. pretty heavily. There are some performance issues with these, but it’s completely worth it for code readability. I’m only dealing with up to hundreds of objects so I’m not likely to see these in my current implementation. But it does make filtering my data more sane:

function filterProperties() {
  _filterProperties.forEach(function(filter) {
    var props = [];
    _data.forEach(function(item) {
      if(item.hasOwnProperty(filter.filter)) {
        var itemPropVal = item[filter.filter];
        if(typeof itemPropVal === 'string') {
          props.push(itemPropVal);
        } else if(itemPropVal instanceof Array) {
          props = props.concat(itemPropVal);
        }
      }
    });
    filter.data = ArrayUtils.unique(props).sort();
    filter.menuData = getDataFormattedForMenu(filter.data);
  });
}

I’ve implemented RxJS for some events – primarily execution environment: browser, mouse, touch, etc. The pure simplicity of this approach is really cool. Just cool.

_itemOverStream = Rx.Observable.fromEvent(_containerEl[0], 'mouseover')
  .filter(filterForMouseEventsOnItems)
  .map(getMouseEventTargetID)
  .subscribe(function (id) {
    selectItemByID(id);
  });

I need to delve deeper in to this, replacing my events and turning my classes into RxJS emitters.

Wrap up

So that’s been my learning path for the past few months. I’ve omitted a few dead-ends and research for brevity (ha!). I still have the back end of this to work though, but I’ve a similar setup at another company so it shouldn’t be that difficult.

I’m going to write a few more posts on various aspects of this – my view components and looking at AMD/CommonJs or WebPack.

If you made it this far, please comment! I’m seeking advice on my approach and rationale.

Thanks for reading.