browserify-handbook

how to build modular applications with browserify

introduction

This document covers how to use browserify to build modular applications.

browserify is a tool for compiling node-flavored commonjs modules for the browser.

You can use browserify to organize your code and use third-party libraries even if you don't use node itself in any other capacity except for bundling and installing packages with npm.

The module system that browserify uses is the same as node, so packages published to npm that were originally intended for use in node but not browsers will work just fine in the browser too.

Increasingly, people are publishing modules to npm which are intentionally designed to work in both node and in the browser using browserify and many packages on npm are intended for use in just the browser. npm is for all javascript, front or backend alike.

table of contents

node packaged manuscript

You can install this handbook with npm, appropriately enough. Just do:

npm install -g browserify-handbook

Now you will have a browserify-handbook command that will open this readme file in your $PAGER. Otherwise, you may continue reading this document as you are presently doing.

node packaged modules

Before we can dive too deeply into how to use browserify and how it works, it is important to first understand how the node-flavored version of the commonjs module system works.

In node, there is a require() function for loading code from other files.

If you install a module with npm:

npm install uniq

Then in a file nums.js we can require('uniq'):

var uniq = require('uniq');
var nums = [ 5, 2, 1, 3, 2, 5, 4, 2, 0, 1 ];
console.log(uniq(nums));

The output of this program when run with node is:

$ node nums.js
[ 0, 1, 2, 3, 4, 5 ]

You can require relative files by requiring a string that starts with a .. For example, to load a file foo.js from main.js, in main.js you can do:

var foo = require('./foo.js');
console.log(foo(4));

If foo.js was in the parent directory, you could use ../foo.js instead:

var foo = require('../foo.js');
console.log(foo(4));

or likewise for any other kind of relative path. Relative paths are always resolved with respect to the invoking file's location.

Note that require() returned a function and we assigned that return value to a variable called uniq. We could have picked any other name and it would have worked the same. require() returns the exports of the module name that you specify.

How require() works is unlike many other module systems where imports are akin to statements that expose themselves as globals or file-local lexicals with names declared in the module itself outside of your control. Under the node style of code import with require(), someone reading your program can easily tell where each piece of functionality came from. This approach scales much better as the number of modules in an application grows.

To export a single thing from a file so that other files may import it, assign over the value at module.exports:

module.exports = function (n) {
    return n * 111
};

Now when some module main.js loads your foo.js, the return value of require('./foo.js') will be the exported function:

var foo = require('./foo.js');
console.log(foo(5));

This program will print:

555

You can export any kind of value with module.exports, not just functions.

For example, this is perfectly fine:

module.exports = 555

and so is this:

var numbers = [];
for (var i = 0; i < 100; i++) numbers.push(i);
 
module.exports = numbers;

There is another form of doing exports specifically for exporting items onto an object. Here, exports is used instead of module.exports:

exports.beep = function (n) { return n * 1000 }
exports.boop = 555

This program is the same as:

module.exports.beep = function (n) { return n * 1000 }
module.exports.boop = 555

because module.exports is the same as exports and is initially set to an empty object.

Note however that you can't do:

// this doesn't work 
exports = function (n) { return n * 1000 }

because the export value lives on the module object, and so assigning a new value for exports instead of module.exports masks the original reference.

Instead if you are going to export a single item, always do:

// instead 
module.exports = function (n) { return n * 1000 }

If you're still confused, try to understand how modules work in the background:

var module = {
  exports: {}
};
 
// If you require a module, it's basically wrapped in a function 
(function(moduleexports) {
  exports = function (n) { return n * 1000 };
}(module, module.exports))
 
console.log(module.exports); // it's still an empty object :( 

Most of the time, you will want to export a single function or constructor with module.exports because it's usually best for a module to do one thing.

The exports feature was originally the primary way of exporting functionality and module.exports was an afterthought, but module.exports proved to be much more useful in practice at being more direct, clear, and avoiding duplication.

In the early days, this style used to be much more common:

foo.js:

exports.foo = function (n) { return n * 111 }

main.js:

var foo = require('./foo.js');
console.log(foo.foo(5));

but note that the foo.foo is a bit superfluous. Using module.exports it becomes more clear:

foo.js:

module.exports = function (n) { return n * 111 }

main.js:

var foo = require('./foo.js');
console.log(foo(5));

To run a module in node, you've got to start from somewhere.

In node you pass a file to the node command to run a file:

$ node robot.js
beep boop

In browserify, you do this same thing, but instead of running the file, you generate a stream of concatenated javascript files on stdout that you can write to a file with the > operator:

$ browserify robot.js > bundle.js

Now bundle.js contains all the javascript that robot.js needs to work. Just plop it into a single script tag in some html:

<html>
  <body>
    <script src="bundle.js"></script> 
  </body>
</html>

Bonus: if you put your script tag right before the </body>, you can use all of the dom elements on the page without waiting for a dom onready event.

There are many more things you can do with bundling. Check out the bundling section elsewhere in this document.

Browserify starts at the entry point files that you give it and searches for any require() calls it finds using static analysis of the source code's abstract syntax tree.

For every require() call with a string in it, browserify resolves those module strings to file paths and then searches those file paths for require() calls recursively until the entire dependency graph is visited.

Each file is concatenated into a single javascript file with a minimal require() definition that maps the statically-resolved names to internal IDs.

This means that the bundle you generate is completely self-contained and has everything your application needs to work with a pretty negligible overhead.

For more details about how browserify works, check out the compiler pipeline section of this document.

node has a clever algorithm for resolving modules that is unique among rival platforms.

Instead of resolving packages from an array of system search paths like how $PATH works on the command line, node's mechanism is local by default.

If you require('./foo.js') from /beep/boop/bar.js, node will look for ./foo.js in /beep/boop/foo.js. Paths that start with a ./ or ../ are always local to the file that calls require().

If however you require a non-relative name such as require('xyz') from /beep/boop/foo.js, node searches these paths in order, stopping at the first match and raising an error if nothing is found:

/beep/boop/node_modules/xyz
/beep/node_modules/xyz
/node_modules/xyz

For each xyz directory that exists, node will first look for a xyz/package.json to see if a "main" field exists. The "main" field defines which file should take charge if you require() the directory path.

For example, if /beep/node_modules/xyz is the first match and /beep/node_modules/xyz/package.json has:

{
  "name": "xyz",
  "version": "1.2.3",
  "main": "lib/abc.js"
}

then the exports from /beep/node_modules/xyz/lib/abc.js will be returned by require('xyz').

If there is no package.json or no "main" field, index.js is assumed:

/beep/node_modules/xyz/index.js

If you need to, you can reach into a package to pick out a particular file. For example, to load the lib/clone.js file from the dat package, just do:

var clone = require('dat/lib/clone.js')

The recursive node_modules resolution will find the first dat package up the directory hierarchy, then the lib/clone.js file will be resolved from there. This require('dat/lib/clone.js') approach will work from any location where you can require('dat').

node also has a mechanism for searching an array of paths, but this mechanism is deprecated and you should be using node_modules/ unless you have a very good reason not to.

The great thing about node's algorithm and how npm installs packages is that you can never have a version conflict, unlike most every other platform. npm installs the dependencies of each package into node_modules.

Each library gets its own local node_modules/ directory where its dependencies are stored and each dependency's dependencies has its own node_modules/ directory, recursively all the way down.

This means that packages can successfully use different versions of libraries in the same application, which greatly decreases the coordination overhead necessary to iterate on APIs. This feature is very important for an ecosystem like npm where there is no central authority to manage how packages are published and organized. Everyone may simply publish as they see fit and not worry about how their dependency version choices might impact other dependencies included in the same application.

You can leverage how node_modules/ works to organize your own local application modules too. See the avoiding ../../../../../../.. section for more.

Browserify is a build step that runs on the server. It generates a single bundle file that has everything in it.

Here are some other ways of implementing module systems for the browser and what their strengths and weaknesses are:

Instead of a module system, each file defines properties on the window global object or develops an internal namespacing scheme.

This approach does not scale well without extreme diligence since each new file needs an additional <script> tag in all of the html pages where the application will be rendered. Further, the files tend to be very order-sensitive because some files need to be included before other files the expect globals to already be present in the environment.

It can be difficult to refactor or maintain applications built this way. On the plus side, all browsers natively support this approach and no server-side tooling is required.

This approach tends to be very slow since each <script> tag initiates a new round-trip http request.

Instead of window globals, all the scripts are concatenated beforehand on the server. The code is still order-sensitive and difficult to maintain, but loads much faster because only a single http request for a single <script> tag needs to execute.

Without source maps, exceptions thrown will have offsets that can't be easily mapped back to their original files.

Instead of using <script> tags, every file is wrapped with a define() function and callback. This is AMD.

The first argument is an array of modules to load that maps to each argument supplied to the callback. Once all the modules are loaded, the callback fires.

define(['jquery'] , function ($) {
    return function () {};
});

You can give your module a name in the first argument so that other modules can include it.

There is a commonjs sugar syntax that stringifies each callback and scans it for require() calls with a regexp.

Code written this way is much less order-sensitive than concatenation or globals since the order is resolved by explicit dependency information.

For performance reasons, most of the time AMD is bundled server-side into a single file and during development it is more common to actually use the asynchronous feature of AMD.

If you're going to have a build step for performance and a sugar syntax for convenience, why not scrap the whole AMD business altogether and bundle commonjs? With tooling you can resolve modules to address order-sensitivity and your development and production environments will be much more similar and less fragile. The CJS syntax is nicer and the ecosystem is exploding because of node and npm.

You can seamlessly share code between node and the browser. You just need a build step and some tooling for source maps and auto-rebuilding.

Plus, we can use node's module lookup algorithms to save us from version mismatch insanity so that we can have multiple conflicting versions of different required packages in the same application and everything will still work. To save bytes down the wire you can dedupe, which is covered elsewhere in this document.

development

Concatenation has some downsides, but these can be very adequately addressed with development tooling.

Browserify supports a --debug/-d flag and opts.debug parameter to enable source maps. Source maps tell the browser to convert line and column offsets for exceptions thrown in the bundle file back into the offsets and filenames of the original sources.

The source maps include all the original file contents inline so that you can simply put the bundle file on a web server and not need to ensure that all the original source contents are accessible from the web server with paths set up correctly.

The downside of inlining all the source files into the inline source map is that the bundle is twice as large. This is fine for debugging locally but not practical for shipping source maps to production. However, you can use exorcist to pull the inline source map out into a separate bundle.map.js file:

browserify main.js --debug | exorcist bundle.js.map > bundle.js

Running a command to recompile your bundle every time can be slow and tedious. Luckily there are many tools to solve this problem.

watchify

You can use watchify interchangeably with browserify but instead of writing to an output file once, watchify will write the bundle file and then watch all of the files in your dependency graph for changes. When you modify a file, the new bundle file will be written much more quickly than the first time because of aggressive caching.

You can use -v to print a message every time a new bundle is written:

$ watchify browser.js -d -o static/bundle.js -v
610598 bytes written to static/bundle.js  0.23s
610606 bytes written to static/bundle.js  0.10s
610597 bytes written to static/bundle.js  0.14s
610606 bytes written to static/bundle.js  0.08s
610597 bytes written to static/bundle.js  0.08s
610597 bytes written to static/bundle.js  0.19s

Here is a handy configuration for using watchify and browserify with the package.json "scripts" field:

{
  "build": "browserify browser.js -o static/bundle.js",
  "watch": "watchify browser.js -o static/bundle.js --debug --verbose",
}

To build the bundle for production do npm run build and to watch files for during development do npm run watch.

Learn more about npm run.

beefy

If you would rather spin up a web server that automatically recompiles your code when you modify it, check out beefy.

Just give beefy an entry file:

beefy main.js

and it will set up shop on an http port.

If you are using express, check out browserify-middleware or enchilada.

They both provide middleware you can drop into an express application for serving browserify bundles.

You can just use the API directly from an ordinary http.createServer() for development too:

var browserify = require('browserify');
var http = require('http');
 
http.createServer(function (reqres) {
    if (req.url === '/bundle.js') {
        res.setHeader('content-type', 'application/javascript');
        var b = browserify(__dirname + '/main.js').bundle();
        b.on('error', console.error);
        b.pipe(res);
    }
    else res.writeHead(404, 'not found')
});

If you use grunt, you'll probably want to use the grunt-browserify plugin.

If you use gulp, you should use the browserify API directly.

Here is a guide for getting started with gulp and browserify.

Here is a guide on how to make browserify builds fast with watchify using gulp from the official gulp recipes.

builtins

In order to make more npm modules originally written for node work in the browser, browserify provides many browser-specific implementations of node core libraries:

events, stream, url, path, and querystring are particularly useful in a browser environment.

Additionally, if browserify detects the use of Buffer, process, global, __filename, or __dirname, it will include a browser-appropriate definition.

So even if a module does a lot of buffer and stream operations, it will probably just work in the browser, so long as it doesn't do any server IO.

If you haven't done any node before, here are some examples of what each of those globals can do. Note too that these globals are only actually defined when you or some module you depend on uses them.

Buffer

In node all the file and network APIs deal with Buffer chunks. In browserify the Buffer API is provided by buffer, which uses augmented typed arrays in a very performant way with fallbacks for old browsers.

Here's an example of using Buffer to convert a base64 string to hex:

var buf = Buffer('YmVlcCBib29w', 'base64');
var hex = buf.toString('hex');
console.log(hex);

This example will print:

6265657020626f6f70

process

In node, process is a special object that handles information and control for the running process such as environment, signals, and standard IO streams.

Of particular consequence is the process.nextTick() implementation that interfaces with the event loop.

In browserify the process implementation is handled by the process module which just provides process.nextTick() and little else.

Here's what process.nextTick() does:

setTimeout(function () {
    console.log('third');
}, 0);
 
process.nextTick(function () {
    console.log('second');
});
 
console.log('first');

This script will output:

first
second
third

process.nextTick(fn) is like setTimeout(fn, 0), but faster because setTimeout is artificially slower in javascript engines for compatibility reasons.

global

In node, global is the top-level scope where global variables are attached similar to how window works in the browser. In browserify, global is just an alias for the window object.

__filename

__filename is the path to the current file, which is different for each file.

To prevent disclosing system path information, this path is rooted at the opts.basedir that you pass to browserify(), which defaults to the current working directory.

If we have a main.js:

var bar = require('./foo/bar.js');
 
console.log('here in main.js, __filename is:', __filename);
bar();

and a foo/bar.js:

module.exports = function () {
    console.log('here in foo/bar.js, __filename is:', __filename);
};

then running browserify starting at main.js gives this output:

$ browserify main.js | node
here in main.js, __filename is: /main.js
here in foo/bar.js, __filename is: /foo/bar.js

__dirname

__dirname is the directory of the current file. Like __filename, __dirname is rooted at the opts.basedir.

Here's an example of how __dirname works:

main.js:

require('./x/y/z/abc.js');
console.log('in main.js __dirname=' + __dirname);

x/y/z/abc.js:

console.log('in abc.js, __dirname=' + __dirname);

output:

$ browserify main.js | node
in abc.js, __dirname=/x/y/z
in main.js __dirname=/

transforms

Instead of browserify baking in support for everything, it supports a flexible transform system that are used to convert source files in-place.

This way you can require() files written in coffee script or templates and everything will be compiled down to javascript.

To use coffeescript for example, you can use the coffeeify transform. Make sure you've installed coffeeify first with npm install coffeeify then do:

$ browserify -t coffeeify main.coffee > bundle.js

or with the API you can do:

var b = browserify('main.coffee');
b.transform('coffeeify');

The best part is, if you have source maps enabled with --debug or opts.debug, the bundle.js will map exceptions back into the original coffee script source files. This is very handy for debugging with firebug or chrome inspector.

Transforms implement a simple streaming interface. Here is a transform that replaces $CWD with the process.cwd():

var through = require('through2');
 
module.exports = function (file) {
    return through(function (bufencnext) {
        this.push(buf.toString('utf8').replace(/\$CWD/g, process.cwd());
        next();
    });
};

The transform function fires for every file in the current package and returns a transform stream that performs the conversion. The stream is written to and by browserify with the original file contents and browserify reads from the stream to obtain the new contents.

Simply save your transform to a file or make a package and then add it with -t ./your_transform.js.

For more information about how streams work, check out the stream handbook.

package.json

You can define a "browser" field in the package.json of any package that will tell browserify to override lookups for the main field and for individual modules.

If you have a module with a main entry point of main.js for node but have a browser-specific entry point at browser.js, you can do:

{
  "name": "mypkg",
  "version": "1.2.3",
  "main": "main.js",
  "browser": "browser.js"
}

Now when somebody does require('mypkg') in node, they will get the exports from main.js, but when they do require('mypkg') in a browser, they will get the exports from browser.js.

Splitting up whether you are in the browser or not with a "browser" field in this way is greatly preferrable to checking whether you are in a browser at runtime because you may want to load different modules based on whether you are in node or the browser. If the require() calls for both node and the browser are in the same file, browserify's static analysis will include everything whether you use those files or not.

You can do more with the "browser" field as an object instead of a string.

For example, if you only want to swap out a single file in lib/ with a browser-specific version, you could do:

{
  "name": "mypkg",
  "version": "1.2.3",
  "main": "main.js",
  "browser": {
    "lib/foo.js": "lib/browser-foo.js"
  }
}

or if you want to swap out a module used locally in the package, you can do:

{
  "name": "mypkg",
  "version": "1.2.3",
  "main": "main.js",
  "browser": {
    "fs": "level-fs-browser"
  }
}

You can ignore files (setting their contents to the empty object) by setting their values in the browser field to false:

{
  "name": "mypkg",
  "version": "1.2.3",
  "main": "main.js",
  "browser": {
    "winston": false
  }
}

The browser field only applies to the current package. Any mappings you put will not propagate down to its dependencies or up to its dependents. This isolation is designed to protect modules from each other so that when you require a module you won't need to worry about any system-wide effects it might have. Likewise, you shouldn't need to wory about how your local configuration might adversely affect modules far away deep into your dependency graph.

You can configure transforms to be automatically applied when a module is loaded in a package's browserify.transform field. For example, we can automatically apply the brfs transform with this package.json:

{
  "name": "mypkg",
  "version": "1.2.3",
  "main": "main.js",
  "browserify": {
    "transform": [ "brfs" ]
  }
}

Now in our main.js we can do:

var fs = require('fs');
var src = fs.readFileSync(__dirname + '/foo.txt', 'utf8');
 
module.exports = function (x) { return src.replace(x, 'zzz') };

and the fs.readFileSync() call will be inlined by brfs without consumers of the module having to know. You can apply as many transforms as you like in the transform array and they will be applied in order.

Like the "browser" field, transforms configured in package.json will only apply to the local package for the same reasons.

Sometimes a transform takes configuration options on the command line. To apply these from package.json you can do the following.

on the command line

browserify -t coffeeify \
           -t [ browserify-ngannotate --ext .coffee ] \
           index.coffee > index.js

in package.json

"browserify"{
  "transform": [
    "coffeeify",
    ["browserify-ngannotate", {"ext": ".coffee"}]
  ]
}

finding good modules

Here are some useful heuristics for finding good modules on npm that work in the browser:

  • I can install it with npm

  • code snippet on the readme using require() - from a quick glance I should see how to integrate the library into what I'm presently working on

  • has a very clear, narrow idea about scope and purpose

  • knows when to delegate to other libraries - doesn't try to do too many things itself

  • written or maintained by authors whose opinions about software scope, modularity, and interfaces I generally agree with (often a faster shortcut than reading the code/docs very closely)

  • inspecting which modules depend on the library I'm evaluating - this is baked into the package page for modules published to npm

Other metrics like number of stars on github, project activity, or a slick landing page, are not as reliable.

People used to think that exporting a bunch of handy utility-style things would be the main way that programmers would consume code because that is the primary way of exporting and importing code on most other platforms and indeed still persists even on npm.

However, this kitchen-sink mentality toward including a bunch of thematically-related but separable functionality into a single package appears to be an artifact for the difficulty of publishing and discovery in a pre-github, pre-npm era.

There are two other big problems with modules that try to export a bunch of functionality all in one place under the auspices of convenience: demarcation turf wars and finding which modules do what.

Packages that are grab-bags of features waste a ton of time policing boundaries about which new features belong and don't belong. There is no clear natural boundary of the problem domain in this kind of package about what the scope is, it's all somebody's smug opinion.

Node, npm, and browserify are not that. They are avowedly ala-carte, participatory, and would rather celebrate disagreement and the dizzying proliferation of new ideas and approaches than try to clamp down in the name of conformity, standards, or "best practices".

Nobody who needs to do gaussian blur ever thinks "hmm I guess I'll start checking generic mathematics, statistics, image processing, and utility libraries to see which one has gaussian blur in it. Was it stats2 or image-pack-utils or maths-extra or maybe underscore has that one?" No. None of this. Stop it. They npm search gaussian and they immediately see ndarray-gaussian-filter and it does exactly what they want and then they continue on with their actual problem instead of getting lost in the weeds of somebody's neglected grand utility fiefdom.

organizing modules

Not everything in an application properly belongs on the public npm and the overhead of setting up a private npm or git repo is still rather large in many cases. Here are some approaches for avoiding the ../../../../../../../ relative paths problem.

People sometimes object to putting application-specific modules into node_modules because it is not obvious how to check in your internal modules without also checking in third-party modules from npm.

The answer is quite simple! If you have a .gitignore file that ignores node_modules:

node_modules

You can just add an exception with ! for each of your internal application modules:

node_modules/*
!node_modules/foo
!node_modules/bar

Please note that you can't unignore a subdirectory, if the parent is already ignored. So instead of ignoring node_modules, you have to ignore every directory inside node_modules with the node_modules/* trick, and then you can add your exceptions.

Now anywhere in your application you will be able to require('foo') or require('bar') without having a very large and fragile relative path.

If you have a lot of modules and want to keep them more separate from the third-party modules installed by npm, you can just put them all under a directory in node_modules such as node_modules/app:

node_modules/app/foo
node_modules/app/bar

Now you will be able to require('app/foo') or require('app/bar') from anywhere in your application.

In your .gitignore, just add an exception for node_modules/app:

node_modules/*
!node_modules/app

If your application had transforms configured in package.json, you'll need to create a separate package.json with its own transform field in your node_modules/foo or node_modules/app/foo component directory because transforms don't apply across module boundaries. This will make your modules more robust against configuration changes in your application and it will be easier to independently reuse the packages outside of your application.

Another handy trick if you are working on an application where you can make symlinks and don't need to support windows is to symlink a lib/ or app/ folder into node_modules. From the project root, do:

ln -s ../lib node_modules/app

and now from anywhere in your project you'll be able to require files in lib/ by doing require('app/foo.js') to get lib/foo.js.

You might see some places talk about using the $NODE_PATH environment variable or opts.paths to add directories for node and browserify to look in to find modules.

Unlike most other platforms, using a shell-style array of path directories with $NODE_PATH is not as favorable in node compared to making effective use of the node_modules directory.

This is because your application is more tightly coupled to a runtime environment configuration so there are more moving parts and your application will only work when your environment is setup correctly.

node and browserify both support but discourage the use of $NODE_PATH.

There are many browserify transforms you can use to do many things. Commonly, transforms are used to include non-javascript assets into bundle files.

One way of including any kind of asset that works in both node and the browser is brfs.

brfs uses static analysis to compile the results of fs.readFile() and fs.readFileSync() calls down to source contents at compile time.

For example, this main.js:

var fs = require('fs');
var html = fs.readFileSync(__dirname + '/robot.html', 'utf8');
console.log(html);

applied through brfs would become something like:

var fs = require('fs');
var html = "<b>beep boop</b>";
console.log(html);

when run through brfs.

This is handy because you can reuse the exact same code in node and the browser, which makes sharing modules and testing much simpler.

fs.readFile() and fs.readFileSync() accept the same arguments as in node, which makes including inline image assets as base64-encoded strings very easy:

var fs = require('fs');
var imdata = fs.readFileSync(__dirname + '/image.png', 'base64');
var img = document.createElement('img');
img.setAttribute('src', 'data:image/png;base64,' + imdata);
document.body.appendChild(img);

If you have some css you want to inline into your bundle, you can do that too with the assistence of a module such as insert-css:

var fs = require('fs');
var insertStyle = require('insert-css');
 
var css = fs.readFileSync(__dirname + '/style.css', 'utf8');
insertStyle(css);

Inserting css this way works fine for small reusable modules that you distribute with npm because they are fully-contained, but if you want a more wholistic approach to asset management using browserify, check out atomify and parcelify.

Putting these ideas about code organization together, we can build a reusable UI component that we can reuse across our application or in other applications.

Here is a bare-bones example of an empty widget module:

module.exports = Widget;
 
function Widget (opts) {
    if (!(this instanceof Widget)) return new Widget(opts);
    this.element = document.createElement('div');
}
 
Widget.prototype.appendTo = function (target) {
    if (typeof target === 'string') target = document.querySelector(target);
    target.appendChild(this.element);
};

Handy javascript constructor tip: you can include a this instanceof Widget check like above to let people consume your module with new Widget or Widget(). It's nice because it hides an implementation detail from your API and you still get the performance benefits and indentation wins of using prototypes.

To use this widget, just use require() to load the widget file, instantiate it, and then call .appendTo() with a css selector string or a dom element.

Like this:

var Widget = require('./widget.js');
var w = Widget();
w.appendTo('#container');

and now your widget will be appended to the DOM.

Creating HTML elements procedurally is fine for very simple content but gets very verbose and unclear for anything bigger. Luckily there are many transforms available to ease importing HTML into your javascript modules.

Let's extend our widget example using brfs. We can also use domify to turn the string that fs.readFileSync() returns into an html dom element:

var fs = require('fs');
var domify = require('domify');
 
var html = fs.readFileSync(__dirname + '/widget.html', 'utf8');
 
module.exports = Widget;
 
function Widget (opts) {
    if (!(this instanceof Widget)) return new Widget(opts);
    this.element = domify(html);
}
 
Widget.prototype.appendTo = function (target) {
    if (typeof target === 'string') target = document.querySelector(target);
    target.appendChild(this.element);
};

and now our widget will load a widget.html, so let's make one:

<div class="widget">
  <h1 class="name"></h1>
  <div class="msg"></div>
</div>

It's often useful to emit events. Here's how we can emit events using the built-in events module and the inherits module:

var fs = require('fs');
var domify = require('domify');
var inherits = require('inherits');
var EventEmitter = require('events').EventEmitter;
 
var html = fs.readFileSync(__dirname + '/widget.html', 'utf8');
 
inherits(Widget, EventEmitter);
module.exports = Widget;
 
function Widget (opts) {
    if (!(this instanceof Widget)) return new Widget(opts);
    this.element = domify(html);
}
 
Widget.prototype.appendTo = function (target) {
    if (typeof target === 'string') target = document.querySelector(target);
    target.appendChild(this.element);
    this.emit('append', target);
};

Now we can listen for 'append' events on our widget instance:

var Widget = require('./widget.js');
var w = Widget();
w.on('append', function (target) {
    console.log('appended to: ' + target.outerHTML);
});
w.appendTo('#container');

We can add more methods to our widget to set elements on the html:

var fs = require('fs');
var domify = require('domify');
var inherits = require('inherits');
var EventEmitter = require('events').EventEmitter;
 
var html = fs.readFileSync(__dirname + '/widget.html', 'utf8');
 
inherits(Widget, EventEmitter);
module.exports = Widget;
 
function Widget (opts) {
    if (!(this instanceof Widget)) return new Widget(opts);
    this.element = domify(html);
}
 
Widget.prototype.appendTo = function (target) {
    if (typeof target === 'string') target = document.querySelector(target);
    target.appendChild(this.element);
};
 
Widget.prototype.setName = function (name) {
    this.element.querySelector('.name').textContent = name;
}
 
Widget.prototype.setMessage = function (msg) {
    this.element.querySelector('.msg').textContent = msg;
}

If setting element attributes and content gets too verbose, check out hyperglue.

Now finally, we can toss our widget.js and widget.html into node_modules/app-widget. Since our widget uses the brfs transform, we can create a package.json with:

{
  "name": "app-widget",
  "version": "1.0.0",
  "private": true,
  "main": "widget.js",
  "browserify": {
    "transform": [ "brfs" ]
  },
  "dependencies": {
    "brfs": "^1.1.1",
    "inherits": "^2.0.1"
  }
}

And now whenever we require('app-widget') from anywhere in our application, brfs will be applied to our widget.js automatically! Our widget can even maintain its own dependencies. This way we can update dependencies in one widgets without worrying about breaking changes cascading over into other widgets.

Make sure to add an exclusion in your .gitignore for node_modules/app-widget:

node_modules/*
!node_modules/app-widget

You can read more about shared rendering in node and the browser if you want to learn about sharing rendering logic between node and the browser using browserify and some streaming html libraries.

testing in node and the browser

Testing modular code is very easy! One of the biggest benefits of modularity is that your interfaces become much easier to instantiate in isolation and so it's easy to make automated tests.

Unfortunately, few testing libraries play nicely out of the box with modules and tend to roll their own idiosyncratic interfaces with implicit globals and obtuse flow control that get in the way of a clean design with good separation.

People also make a huge fuss about "mocking" but it's usually not necessary if you design your modules with testing in mind. Keeping IO separate from your algorithms, carefully restricting the scope of your module, and accepting callback parameters for different interfaces can all make your code much easier to test.

For example, if you have a library that does both IO and speaks a protocol, consider separating the IO layer from the protocol using an interface like streams.

Your code will be easier to test and reusable in different contexts that you didn't initially envision. This is a recurring theme of testing: if your code is hard to test, it is probably not modular enough or contains the wrong balance of abstractions. Testing should not be an afterthought, it should inform your whole design and it will help you to write better interfaces.

tape

Tape was specifically designed from the start to work well in both node and browserify. Suppose we have an index.js with an async interface:

module.exports = function (xcb) {
    setTimeout(function () {
        cb(* 100);
    }, 1000);
};

Here's how we can test this module using tape. Let's put this file in test/beep.js:

var test = require('tape');
var hundreder = require('../');
 
test('beep', function (t) {
    t.plan(1);
    
    hundreder(5, function (n) {
        t.equal(n, 500, '5*100 === 500');
    });
});

Because the test file lives in test/, we can require the index.js in the parent directory by doing require('../'). index.js is the default place that node and browserify look for a module if there is no package.json in that directory with a main field.

We can require() tape like any other library after it has been installed with npm install tape.

The string 'beep' is an optional name for the test. The 3rd argument to t.equal() is a completely optional description.

The t.plan(1) says that we expect 1 assertion. If there are not enough assertions or too many, the test will fail. An assertion is a comparison like t.equal(). tape has assertion primitives for:

  • t.equal(a, b) - compare a and b strictly with ===
  • t.deepEqual(a, b) - compare a and b recursively
  • t.ok(x) - fail if x is not truthy

and more! You can always add an additional description argument.

Running our module is very simple! To run the module in node, just run node test/beep.js:

$ node test/beep.js
TAP version 13
# beep
ok 1 5*100 === 500
 
1..1
# tests 1
# pass  1
 
# ok

The output is printed to stdout and the exit code is 0.

To run our code in the browser, just do:

$ browserify test/beep.js > bundle.js

then plop bundle.js into a <script> tag:

<script src="bundle.js"></script>

and load that html in a browser. The output will be in the debug console which you can open with F12, ctrl-shift-j, or ctrl-shift-k depending on the browser.

This is a bit cumbersome to run our tests in a browser, but you can install the testling command to help. First do:

npm install -g testling

And now just do browserify test/beep.js | testling:

$ browserify test/beep.js | testling
 
TAP version 13
# beep
ok 1 5*100 === 500
 
1..1
# tests 1
# pass  1
 
# ok

testling will launch a real browser headlessly on your system to run the tests.

Now suppose we want to add another file, test/boop.js:

var test = require('tape');
var hundreder = require('../');
 
test('fraction', function (t) {
    t.plan(1);
 
    hundreder(1/20, function (n) {
        t.equal(n, 5, '1/20th of 100');
    });
});
 
test('negative', function (t) {
    t.plan(1);
 
    hundreder(-3, function (n) {
        t.equal(n, -300, 'negative number');
    });
});

Here our test has 2 test() blocks. The second test block won't start to execute until the first is completely finished, even though it is asynchronous. You can even nest test blocks by using t.test().

We can run test/boop.js with node directly as with test/beep.js, but if we want to run both tests, there is a minimal command-runner we can use that comes with tape. To get the tape command do:

npm install -g tape

and now you can run:

$ tape test/*.js
TAP version 13
# beep
ok 1 5*100 === 500
# fraction
ok 2 1/20th of 100
# negative
ok 3 negative number
 
1..3
# tests 3
# pass  3
 
# ok

and you can just pass test/*.js to browserify to run your tests in the browser:

$ browserify test/* | testling
 
TAP version 13
# beep
ok 1 5*100 === 500
# fraction
ok 2 1/20th of 100
# negative
ok 3 negative number
 
1..3
# tests 3
# pass  3
 
# ok

Putting together all these steps, we can configure package.json with a test script:

{
  "name": "hundreder",
  "version": "1.0.0",
  "main": "index.js",
  "devDependencies": {
    "tape": "^2.13.1",
    "testling": "^1.6.1"
  },
  "scripts": {
    "test": "tape test/*.js",
    "test-browser": "browserify test/*.js | testlingify"
  }
}

Now you can do npm test to run the tests in node and npm run test-browser to run the tests in the browser. You don't need to worry about installing commands with -g when you use npm run: npm automatically sets up the $PATH for all packages installed locally to the project.

If you have some tests that only run in node and some tests that only run in the browser, you could have subdirectories in test/ such as test/server and test/browser with the tests that run both places just in test/. Then you could just add the relevant directory to the globs:

{
  "name": "hundreder",
  "version": "1.0.0",
  "main": "index.js",
  "devDependencies": {
    "tape": "^2.13.1",
    "testling": "^1.6.1"
  },
  "scripts": {
    "test": "tape test/*.js test/server/*.js",
    "test-browser": "browserify test/*.js test/browser/*.js | testlingify"
  }
}

and now server-specific and browser-specific tests will be run in addition to the common tests.

If you want something even slicker, check out prova once you have gotten the basic concepts.

The core assert module is a fine way to write simple tests too, although it can sometimes be tricky to ensure that the correct number of callbacks have fired.

You can solve that problem with tools like macgyver but it is appropriately DIY.

bundling

This section covers bundling in more detail.

Bundling is the step where starting from the entry files, all the source files in the dependency graph are walked and packed into a single output file.

One of the first things you'll want to tweak is how the files that npm installs are placed on disk to avoid duplicates.

When you do a clean install in a directory, npm will ordinarily factor out similar versions into the topmost directory where 2 modules share a dependency. However, as you install more packages, new packages will not be factored out automatically. You can however use the npm dedupe command to factor out packages for an already-installed set of packages in node_modules/. You could also remove node_modules/ and install from scratch again if problems with duplicates persist.

browserify will not include the same exact file twice, but compatible versions may differ slightly. browserify is also not version-aware, it will include the versions of packages exactly as they are laid out in node_modules/ according to the require() algorithm that node uses.

You can use the browserify --list and browserify --deps commands to further inspect which files are being included to scan for duplicates.

You can generate UMD bundles with --standalone that will work in node, the browser with globals, and AMD environments.

Just add --standalone NAME to your bundle command:

$ browserify foo.js --standalone xyz > bundle.js

This command will export the contents of foo.js under the external module name xyz. If a module system is detected in the host environment, it will be used. Otherwise a window global named xyz will be exported.

You can use dot-syntax to specify a namespace hierarchy:

$ browserify foo.js --standalone foo.bar.baz > bundle.js

If there is already a foo or a foo.bar in the host environment in window global mode, browserify will attach its exports onto those objects. The AMD and module.exports modules will behave the same.

Note however that standalone only works with a single entry or directly-required file.

In browserify parlance, "ignore" means: replace the definition of a module with an empty object. "exclude" means: remove a module completely from a dependency graph.

Another way to achieve many of the same goals as ignore and exclude is the "browser" field in package.json, which is covered elsewhere in this document.

Ignoring is an optimistic strategy designed to stub in an empty definition for node-specific modules that are only used in some codepaths. For example, if a module requires a library that only works in node but for a specific chunk of the code:

var fs = require('fs');
var path = require('path');
var mkdirp = require('mkdirp');
 
exports.convert = convert;
function convert (src) {
    return src.replace(/beep/g, 'boop');
}
 
exports.write = function (srcdstcb) {
    fs.readFile(src, function (errsrc) {
        if (err) return cb(err);
        mkdirp(path.dirname(dst), function (err) {
            if (err) return cb(err);
            var out = convert(src);
            fs.writeFile(dst, out, cb);
        });
    });
};

browserify already "ignores" the 'fs' module by returning an empty object, but the .write() function here won't work in the browser without an extra step like a static analysis transform or a runtime storage fs abstraction.

However, if we really want the convert() function but don't want to see mkdirp in the final bundle, we can ignore mkdirp with b.ignore('mkdirp') or browserify --ignore mkdirp. The code will still work in the browser if we don't call write() because require('mkdirp') won't throw an exception, just return an empty object.

Generally speaking it's not a good idea for modules that are primarily algorithmic (parsers, formatters) to do IO themselves but these tricks can let you use those modules in the browser anyway.

To ignore foo on the command-line do:

browserify --ignore foo

To ignore foo from the api with some bundle instance b do:

b.ignore('foo')

Another related thing we might want is to completely remove a module from the output so that require('modulename') will fail at runtime. This is useful if we want to split things up into multiple bundles that will defer in a cascade to previously-defined require() definitions.

For example, if we have a vendored standalone bundle for jquery that we don't want to appear in the primary bundle:

$ npm install jquery
$ browserify -r jquery --standalone jquery > jquery-bundle.js

then we want to just require('jquery') in a main.js:

var $ = require('jquery');
$(window).click(function () { document.body.bgColor = 'red' });

defering to the jquery dist bundle so that we can write:

<script src="jquery-bundle.js"></script>
<script src="bundle.js"></script>

and not have the jquery definition show up in bundle.js, then while compiling the main.js, you can --exclude jquery:

browserify main.js --exclude jquery > bundle.js

To exclude foo on the command-line do:

browserify --exclude foo

To exclude foo from the api with some bundle instance b do:

b.exclude('foo')

shimming

Unfortunately, some packages are not written with node-style commonjs exports. For modules that export their functionality with globals or AMD, there are packages that can help automatically convert these troublesome packages into something that browserify can understand.

One way to automatically convert non-commonjs packages is with browserify-shim.

browserify-shim is loaded as a transform and also reads a "browserify-shim" field from package.json.

Suppose we need to use a troublesome third-party library we've placed in ./vendor/foo.js that exports its functionality as a window global called FOO. We can set up our package.json with:

{
  "browserify": {
    "transform": "browserify-shim"
  },
  "browserify-shim": {
    "./vendor/foo.js": "FOO"
  }
}

and now when we require('./vendor/foo.js'), we get the FOO variable that ./vendor/foo.js tried to put into the global scope, but that attempt was shimmed away into an isolated context to prevent global pollution.

We could even use the browser field to make require('foo') work instead of always needing to use a relative path to load ./vendor/foo.js:

{
  "browser": {
    "foo": "./vendor/foo.js"
  },
  "browserify": {
    "transform": "browserify-shim"
  },
  "browserify-shim": {
    "foo": "FOO"
  }
}

Now require('foo') will return the FOO export that ./vendor/foo.js tried to place on the global scope.

partitioning

Most of the time, the default method of bundling where one or more entry files map to a single bundled output file is perfectly adequate, particularly considering that bundling minimizes latency down to a single http request to fetch all the javascript assets.

However, sometimes this initial penalty is too high for parts of a website that are rarely or never used by most visitors such as an admin panel. This partitioning can be accomplished with the technique covered in the ignoring and excluding section, but factoring out shared dependencies manually can be tedious for a large and fluid dependency graph.

Luckily, there are plugins that can automatically factor browserify output into separate bundle payloads.

factor-bundle splits browserify output into multiple bundle targets based on entry-point. For each entry-point, an entry-specific output file is built. Files that are needed by two or more of the entry files get factored out into a common bundle.

For example, suppose we have 2 pages: /x and /y. Each page has an entry point, x.js for /x and y.js for /y.

We then generate page-specific bundles bundle/x.js and bundle/y.js with bundle/common.js containing the dependencies shared by both x.js and y.js:

browserify x.js y.js -p [ factor-bundle -o bundle/x.js -o bundle/y.js ] \
  -o bundle/common.js

Now we can simply put 2 script tags on each page. On /x we would put:

<script src="/bundle/common.js"></script>
<script src="/bundle/x.js"></script>

and on page /y we would put:

<script src="/bundle/common.js"></script>
<script src="/bundle/y.js"></script>

You could also load the bundles asynchronously with ajax or by inserting a script tag into the page dynamically but factor-bundle only concerns itself with generating the bundles, not with loading them.

partition-bundle handles splitting output into multiple bundles like factor-bundle, but includes a built-in loader using a special loadjs() function.

partition-bundle takes a json file that maps source files to bundle files:

{
  "entry.js": ["./a"],
  "common.js": ["./b"],
  "common/extra.js": ["./e", "./d"]
}

Then partition-bundle is loaded as a plugin and the mapping file, output directory, and destination url path (required for dynamic loading) are passed in:

browserify -p [ partition-bundle --map mapping.json \
  --output output/directory --url directory ]

Now you can add:

<script src="entry.js"></script>

to your page to load the entry file. From inside the entry file, you can dynamically load other bundles with a loadjs() function:

a.addEventListener('click', function() {
  loadjs(['./e', './d'], function(ed) {
    console.log(e, d);
  });
});

compiler pipeline

Since version 5, browserify exposes its compiler pipeline as a labeled-stream-splicer.

This means that transformations can be added or removed directly into the internal pipeline. This pipeline provides a clean interface for advanced customizations such as watching files or factoring bundles from multiple entry points.

For example, we could replace the built-in integer-based labeling mechanism with hashed IDs by first injecting a pass-through transform after the "deps" have been calculated to hash source files. Then we can use the hashes we captured to create our own custom labeler, replacing the built-in "label" transform:

var browserify = require('browserify');
var through = require('through2');
var shasum = require('shasum');
 
var b = browserify('./main.js');
 
var hashes = {};
var hasher = through.obj(function (rowencnext) {
    hashes[row.id] = shasum(row.source);
    this.push(row);
    next();
});
b.pipeline.get('deps').push(hasher);
 
var labeler = through.obj(function (rowencnext) {
    row.id = hashes[row.id];
    
    Object.keys(row.deps).forEach(function (key) {
        row.deps[key] = hashes[row.deps[key]];
    });
    
    this.push(row);
    next();
});
b.pipeline.get('label').splice(0, 1, labeler);
 
b.bundle().pipe(process.stdout);

Now instead of getting integers for the IDs in the output format, we get file hashes:

$ node bundle.js
(function e(t,n,r){function s(o,u){if(!n[o]){if(!t[o]){var a=typeof require=="function"&&require;if(!u&&a)return a(o,!0);if(i)return i(o,!0);var f=new Error("Cannot find module '"+o+"'");throw f.code="MODULE_NOT_FOUND",f}var l=n[o]={exports:{}};t[o][0].call(l.exports,function(e){var n=t[o][1][e];return s(n?n:e)},l,l.exports,e,t,n,r)}return n[o].exports}var i=typeof require=="function"&&require;for(var o=0;o<r.length;o++)s(r[o]);return s})({"5f0a0e3a143f2356582f58a70f385f4bde44f04b":[function(require,module,exports){
var foo = require('./foo.js');
var bar = require('./bar.js');
 
console.log(foo(3) + bar(4));
 
},{"./bar.js":"cba5983117ae1d6699d85fc4d54eb589d758f12b","./foo.js":"736100869ec2e44f7cfcf0dc6554b055e117c53c"}],"cba5983117ae1d6699d85fc4d54eb589d758f12b":[function(require,module,exports){
module.exports = function (n) { return n * 100 };
 
},{}],"736100869ec2e44f7cfcf0dc6554b055e117c53c":[function(require,module,exports){
module.exports = function (n) { return n + 1 };
 
},{}]},{},["5f0a0e3a143f2356582f58a70f385f4bde44f04b"]);

Note that the built-in labeler does other things like checking for the external, excluded configurations so replacing it will be difficult if you depend on those features. This example just serves as an example for the kinds of things you can do by hacking into the compiler pipeline.

Each phase in the browserify pipeline has a label that you can hook onto. Fetch a label with .get(name) to return a labeled-stream-splicer handle at the appropriate label. Once you have a handle, you can .push(), .pop(), .shift(), .unshift(), and .splice() your own transform streams into the pipeline or remove existing transform streams.

The recorder is used to capture the inputs sent to the deps phase so that they can be replayed on subsequent calls to .bundle(). Unlike in previous releases, v5 can generate bundle output multiple times. This is very handy for tools like watchify that re-bundle when a file has changed.

The deps phase expects entry and require() files or objects as input and calls module-deps to generate a stream of json output for all of the files in the dependency graph.

module-deps is invoked with some customizations here such as:

  • setting up the browse