A brief and incomplete history of JavaScript bundlers

Ever since I read Malte Ubl’s proposal for a JavaScript bundle syntax, I’ve been fascinated by the question: does JavaScript need a “bundle” standard?

Unfortunately that question will have to wait for another post, because it’s much more complicated than what I can cover here. But to at least make the first tentative stabs at answering it, I’d like to explore some more basic questions:

  • What is a JavaScript bundler?
  • What purpose do bundlers serve in the modern webdev stack?

To try to answer these questions, I’d like to offer my historical perspective on what are (arguably) the two most important bundlers of the last five years: Browserify and Webpack.

A bundle of bamboo

A bundle of bamboo, via Wikipedia

What is a bundle?

Conceptually, a JavaScript bundle is very simple: it’s a collection of multiple scripts, combined into a single file. The original bundler was called +=, i.e. concatenation, and for a long time it was all anyone really needed. The whole point was to avoid the 6-connections-per-origin limit and the built-in overhead of HTTP/1.1 connections by simply jamming all your JavaScript into a single file. Easy-peasy.

Disregarding some interesting but ultimately niche bundlers such as GWT, RequireJS, and Closure Compiler, concatenation was still the most common bundler until very recently. Even fairly modern scaffolding tools like Yeoman were still recommending concatenation as the default bundler well into 2013, using lightweight tools such as usemin.

It was only really when Browserify hit the scene in 2013 did non-concatenation bundlers start to go mainstream.

The rise of Browserify

Interestingly, Browserify wasn’t originally designed to solve the problem of bundling. Instead, it was designed to solve the problem of Node developers who wanted to reuse their code in the browser. (It’s right there in the name: “browser-ify” your Node code!)

Screenshot of Browserify homepage from 2013

Screenshot of the original Browserify homepage from January 2013 (via the Internet Archive)

Before Browserify, if you were writing a JavaScript module that was designed to work in both Node or the browser, you’d have to do something like this:

var MyModule = 'hello world';

if (typeof module !== 'undefined' && module.exports) {
  module.exports = MyModule;
} else {
  (typeof self !== 'undefined' ? self : window).MyModule = MyModule;
}

This works fine for single files, but if you’re accustomed to Node conventions, it becomes aggravating that you can’t do something like this:

var otherModule = require('./otherModule');

Or even:

var otherPackage = require('other-package');

By 2014, npm had already grown to over 50,000 modules, so the idea of reusing those modules within browser code was a compelling proposition. The problem Browserify solved was thus twofold:

  1. Make the CommonJS module system work for the browser (by crawling the dependency tree, reading files, and building a single bundle file).
  2. Make Node built-ins and conventions (process, Buffer, crypto, etc.) work in the browser, by implementing polyfills and shims for them.

This second point is an often-overlooked benefit of the cowpath that Browserify paved. At the time Browserify debuted, many of those 50,000 modules weren’t written with any consideration for how they might run in the browser, and Node-isms like process.nextTick() and setImmediate() ran rampant. For Browserify to “just work,” it had to solve the compatibility problem.

What this involved was a lot of effort to reimplement nearly all of Node’s standard library for the browser, tackling the inevitable issues of cross-browser compatibility along the way. This resulted in some extremely battle-tested libraries such as events, process, buffer, inherits, and crypto, among others.

If you want to understand the ridiculous amount of work that had to go into building all this infrastructure, I recommend taking a look at Calvin Metcalf’s series on implementing crypto for the browser. Or, if you’re too faint of heart, you can instead read about how he helped fix process.nextTick() to work with Sinon or avoid bugs in oldIE’s timer system. (Calvin is truly one of the unsung heroes of JavaScript. Look in your bundle, and odds are you will find his code in there somewhere!)

All of these libraries – buffer, crypto, process, etc. – are still in wide use today via Browserify, as well as other bundlers like Webpack and Rollup. They are the magic behind why new Buffer() and process.nextTick() “just work,” and are a big part of Browserify’s success story.

Enter Webpack

While Browserify was picking up steam, and more and more browser-ready modules were starting to be published to npm, Webpack rose to prominence in 2015, buoyed by the popularity of React and the endorsement of Pete Hunt.

Webpack and Browserify are often seen today as solutions to the same problem, but Webpack’s initial focus was a bit different from Browserify’s. Whereas Browserify’s goal was to make Node modules run in the browser, Webpack’s goal was to create a dependency graph for all of the assets in a website – not just JavaScript, but also CSS, images, SVGs, and even HTML.

The Webpack view of the world, with multiple types of assets all treated as part of the dependency graph

The Webpack view of the world, via “What is Webpack?”

In contrast to Browserify, which was almost dogmatic in its insistence on Node compatibility, Webpack was cheerful to break Node conventions and introduce code like this:

require('./styles.css');

Or even:

var svg = require('svg-url?limit=1024!./file.svg');

Webpack did this for a few different reasons:

  1. Once all of a website’s assets can be expressed as a dependency graph, it becomes easy to define “components” (collections of HTML, CSS, JavaScript, images, etc.) as standalone modules, which can be easily reused and even published to npm.
  2. Using a JavaScript-based module system for assets means that Hot Module Replacement is easy and natural, e.g. a stylesheet can automatically update itself by injection and replacement into the DOM via script.
  3. Ultimately, all of this is configurable using loaders, meaning you can get the benefits of an integrated module system without having to ship a gigantic JavaScript bundle to your users. (Although how well this works in practice is debatable.).

Because Browserify was originally the only game in town, though, Webpack had to undergo its own series of compatibility fixes, so that existing Browserify-targeting modules could work well with Webpack. This wasn’t always easy, as a JavaScript package maintainer of the time might have told you.

Out of this push for greater Webpack-Browserify compatibility grew ad-hoc standards like the node-browser-resolve algorithm, which defines what the "browser" field in package.json is supposed to do. (This field is an extension of npm’s own package.json definition, which specifies how modules should be swapped out when building in “browser mode” vs “Node mode.”)

Closing thoughts

Today, Browserify and Webpack have largely converged in functionality, although Browserify still tends to be preferred by old-school Node developers, whereas Webpack is the tool of choice for frontend web developers. Up-and-comers such as Rollup, splittable, and fuse-box (among many others) are also making the frontend bundler landscape increasingly diverse and interesting.

So that’s my view of the great bundler wars of 2013-2017! Hopefully in a future blog post I’ll be able to cover whether or not bundlers like Browserify and Webpack demonstrate the need for a “standard” to unite them all.

Feel free to weigh in on Twitter or on Mastodon.

6 responses to this post.

  1. […] Ever since I read Malte Ubl’s proposal for a JavaScript bundle syntax, I’ve been fascinated by the question: does JavaScript need a “bundle” standard? Unfortunately that que… Read more […]

    Reply

  2. I predict that Closure compiler will become a more popular choice going forward. Adding a transpilation step like TypeScript -> JavaScript before feeding the Javascript to the Closure compiler will make the barrier to entry a little smaller. During this step we can add Closure annotations etc that will make code more compatible with Closure.

    Bundlers will likely continue to improve. I predict CommonJS based approaches will fall out of favor though since it’s harder to statically analyze commonJS than ES2015.

    Earlier today I made a few comparisons between different bundlers. Sharing in case you are interested in the numbers. (SystemJS-Builder, Rollup, Webpack and Closure compiler).

    http://www.syntaxsuccess.com/viewarticle/angular-production-builds

    Reply

  3. Posted by ezekielcyrus on May 28, 2017 at 6:51 PM

    Require.js, r.js, almond.js definitely helped paved the way. Code splitting comes to mind.

    Reply

  4. I’m a bit surprised you didn’t mention “packer” from Dean Edwards (http://dean.edwards.name/packer/) which exists since 2004. It was one of the recommended tools to enhance web performance (needed very strict JS code). The code was kind of binary compressed and had to be uncompressed on client-side before being given to eval().

    There was also:

    JSMin (http://www.crockford.com/javascript/jsmin.html), Douglas Crockford which promotion was granted by the excellent other tool he wrote (JSLint), along with its JSON standard & “The good parts” book;
    dojo compressor aka shrinksafe (http://shrinksafe.dojotoolkit.org/), by one of the firsts “framework” providers, which also inspired the AMD pattern & requirejs;
    and YUI Compressor (http://yui.github.io/yuicompressor/), which had the advantage to be developed by the Yahoo performance team which provided a very browser extension at that time (YSlow) and the first really serious JSDoc tool (YUI Doc)

    Reply

  5. […] ※本記事は、A brief and incomplete history of JavaScript bundlersを翻訳・再構成したものです。 […]

    Reply

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: