I should have written this post a while ago because something I love is not getting the traction it deserves and writing this earlier may have helped in some small way to change that.

Earlier this year we spent a lot of time trying to understand the best way to package and deliver client-side dependencies. It’s a problem that afflicts all modern web development regardless of the stack you use. Most of the solutions we tried don’t address the real problem, which is about delivering large monolithic libraries and frameworks to the client because some small part of them is needed. Like jQuery, for example. Even underscore, as much as I love it. You might use a few features from each. And then it all adds up. Even uglified and gzipped, it’s not uncommon for a page to be accompanied by up to a megabyte of JavaScript. That’s no good. Not even with pervasive broadband. Especially not on mobile devices over a flaky connection.

Some of these, like bootstrap, allow you to customise the download to include just the bits you want. This is great. But a bit of a faff. And it seems like the wrong solution. I don’t know many people that actually do it.

As an industry we’re moving away from all that. We’re learning from the age-old UNIX way that Eric Raymond so brilliantly described in The Art of UNIX Programming; small, sharp tools, each only doing one thing but doing it well. Modern polyglot architectures are assembled from concise and highly focussed modules of functionality. Software is all about abstracting complexity because our brains cannot be everywhere at once. We all know that if we focus on one job and do it well, we can be sure it works properly and we won’t have to build that same thing again. This is the most efficient way to exploit reuse in software engineering.

But small modules have to be composed. And their dependencies managed. We need something that allows us to pluck a module out of the ether and just use it. We want to depend on it without worrying about what it depends on.

npm is one of the best dependency managers I’ve used. I love how it allows your app to reference a directed acyclic graph of dependencies that is managed for you by the beautiful simplicity of ‘require’ (commonjs modules). In node.js, this works brilliantly well, allowing each module to reference specific versions of its dependencies so that overall there may be lots of different versions of a module in the graph. Even multiple copies of the same version. It allows each module to evolve independently on its own track. And it doesn’t matter how many different versions or copies of a library you’ve got in your app when it’s running on a server. Disk space and memory are cheap. And the stability and flexibility it promotes is well worth the price.

But on the client it’s a different story. You wouldn’t want to download several versions of a library in your page just because different modules were developed independently and some haven’t been updated to use the latest version of something. And the bigger the modules are the worse this would become. Fortunately, the smaller they are, the easier they are to update and the less they, themselves, depend on in the first place. It’s simple to keep a small module up to date. And by small, I’m talking maybe 10 lines of code. Maybe a few hundred, but definitely not more than that.

Enter Component by the prolific (and switched on) TJ Holowaychuk. Not perfect, but until we get Web Components, it’s the best client-side module manager out there. Why? Because it promotes tiny modules. They can be just a bit of functionality, or little bits of UI (widgets if you like). If you use Component, you’re encouraged to use, and/or write, small modules. Like a string trimmer, for example; only 13 loc. Or a tiny, express-like, client-side router in under 1200 bytes.  There are thousands of them. This is a Hacker News button, built with Component:

The Hacker News button, built with Component, showing dependencies
The Hacker News button, built with Component, showing dependencies

The registry

The great thing about Component is that it fetches the files specified in the component.json from Github, following the pattern “organisation/account” (you can specify other locations). This is great. The namespacing stops the bun-fight for cool names because the organisation is included in the unique identifier.

The other major benefit of this is that you can fork a component, modify it and point your app at your own repo if you’re not getting any of your pull requests integrated.

App structure

But it’s not really about 3rd party modules. In my head it’s more about how you structure the code that drives your page.

Component allows you to write completely self-contained features and plug them together. Your components will be DRY and follow the SRP. Each component can have scripts (e.g. JavaScript, or CoffeeScript), styles (e.g. CSS, or Less, or Sass), templates (e.g. compiled from HTML, or Jade), data (e.g. JSON), images, fonts and other files, as well as their own dependencies (other components). All this is specified in the component.json file, which points to everything the component needs, and informs the build step so that everything is packaged up correctly. It can be a little laborious to specify everything in the component.json, but it’s worth it. When you install a component, the component.json specifies exactly which files (in the Github repo) should be downloaded (unlike Bower, for example, where the whole repo has to be fetched) - check out how fast “component install” is.

The self-contained nature of components means that you don’t have a separate scripts folder with every script for the page in it, and a styles folder with all the CSS. Instead, everything is grouped by function, so everything the component needs is contained in the component’s folder. At build time, you can use Grunt to run component build which transpiles the CoffeeScript to JavaScript, the Less to CSS, the Jade to JavaScript functions, and packages the assets. The dependencies are analysed and all the JavaScript ends up in the right order in one file, all the CSS in another. These and the other assets are copied to the build directory, uglified/compressed ready for delivery to the client.

Getting started

The best docs are in the wiki on the Github repo. The FAQ is especially germane. And TJ’s original blog post is great reading, including the rather brilliant discussion about AMD vs Common JS modules. AMD was invented for asynchronous loading. But when you think about it, you’re gonna package all your script up in one compressed HTTP response anyway; there’s still too much overhead associated with multiple requests, even with HTTP keepalive (it’s not so bad with Spdy). The perceived benefits of loading asynchronously, as required, are not yet fully realisable, so we may as well go for the simple require and module.exports pattern we know and love from node.js.

If you’re using CoffeeScript, Jade and JSON in your components, you can use a Gruntfile that looks like this (which contains a workaround for the fact that the coffee compilation step changes the filename extensions from .coffee to .js):

We’ve tried a bunch of different tools to solve the problem of easily and efficiently distributing your app to the browser. All of them have flaws. We used to loveJam.js and Bower. But we got into a jam with jam, because updates were getting jammed due to unresponsive maintainers (sorry, couldn’t resist that). Bower was great, but too heavy. Browserify is too tightly coupled with node.js and npm. None of them make simple, self contained, focused modules as straightforward and elegant as Component. Nice one, TJ!