Meet the New Stack, Same as the Old Stack

2013-02-04 00:00:00 +0000 by Alex R. Young

Five years ago, if you asked any client-side developer which library or framework to use the most likely answer would have been jQuery. Since then, client-side development has become far more complex. A friendly wrapper for DOM programming and a browser compatibility layer isn't enough to help us write modern applications.

Starting a project today involves selecting a package manager, module system, build system, templating language, data binding library, a sync layer, a widget library, and a test framework. And it doesn't end there: many of us also regularly use a dynamic CSS language, and mobile SDKs.

I wouldn't say this is a negative trend, but with large companies backing or offering solutions to these problems, making an informed decision on each of these technologies is difficult. And soon the major players will offer a complete set of tools and libraries to satisfy each requirement: we'll be back to something that looks like the original monolithic frameworks.

Until recently, starting a client-side web application might have looked like this:

There are other popular choices for each of these bullets, of course: Zurb Foundation is a popular front-end framework, and I've used Mocha instead of QUnit since Mocha appeared. I also like Knockout for data binding, because the two-way declarative bindings are easy to get the hang of.

These libraries are not interchangeable once a project has been started -- Bootstrap uses different CSS classes to jQuery UI, for example. The major difficulty is keeping libraries up to date, particularly if they have a lot of dependencies.

And that's when you need a package manager. Using a package manager can make the choices even more fine grained, because managing each library and its dependencies is easier. Switching to something like Component is one option, which can lead to a totally new stack:

Bootstrap and Zurb Foundation can be provided as components, there are projects on GitHub to do this. I've tried to design projects 100% around components without these larger libraries, and it was a huge amount of work. It may get easier with time, or once the right balance of functionality is found. I've noticed there are some "meta packages" that exist to group commonly used dependencies together.

You'll notice I haven't mentioned AngularJS yet. The reason for that is AngularJS is now compatible with Closure Library, which makes it possible to use an almost 100% Google-powered development stack:

While Closure Library is more like the "last generation" monolithic frameworks, each module can be loaded separately, so you don't need to use the whole thing. You could make a project with Closure Library, ClosureBuilder, Backbone.js, and Boostrap if you wanted. You could also go the other way: deploy a Go/Python/Java app to App Engine that's built on Closure Library and AngularJS. Google effectively provides the entire stack, including server-side development, data storage, user authentication, billing, and client-side development.

Recently we've also seen a huge amount of open source projects coming out of Twitter. A pure Twitter stack looks like this:

Using components through Flight could satisfy the other dependencies as well. It wouldn't be difficult to use Flight as a test runner with a suitable assertion library, although the authors use Jasmine and PhantomJS at the moment.

Let's not forget, however, the incredible features provided by Dojo. Compare Google and Twitter's client-side stacks to this:

Dojo has the entire client-side stack covered, and also includes many more features that I haven't mentioned here. YUI is comparable:

On the surface, it seems like Twitter's developers aim to create something more like Component, where each piece of the client-side puzzle can be picked à la carte. Closure Library is more like the older, monolithic model used by YUI, where a custom module system is used, and modules are harder to reuse without the base framework.

The question is, can these larger companies support modern, "componentised" client-side development, or is it easier to offer a consolidated monolithic framework? Projects like AngularJS and Flight suggest to me that developers within Twitter and Google want to promote that approach, while how that fits into the wider organisation remains to be seen. Will we see AngularJS alongside Closure Library in the Python, Java, and Go App Engine documentation, or will it remain a useful library that exists outside of Google's core developer offering?

An interesting third option is Yeoman. This is another project from Google that provides a selection of smaller libraries to kick-start development:

Yeoman generates application templates by helping you select various options, and then gives you a build system so you can easily generate something deployable. It doesn't enforce decisions like template language or CoffeeScript, but provides a harness for installing packages, and building and testing your application.


Client-side development is changing. It's no-longer enough to learn a small amount of jQuery -- an entire client-side stack is becoming the norm. The libraries that plugged the gaps in jQuery and its clones now have serious competition from tech giants.