Warning: If you are upgrading across major versions, please read the release notes in the changelog.
I really dislike setting up build scripts. Most of the time I want to do the exact same thing:
While developing:
- easy to setup and run dev server
- transpile ES6+, JSX, Stylus code
- hotload (a.k.a. live reload) modules when changed
When ready to ship:
- minify and bundle all the things
- output minified, uniquely named static files into public directory
- be able to generate customized static HTML file(s) used to deliver my JS app
- be ready to just upload it all to something like surge.sh
- sometimes I want to pre-render all known HTML into static HTML files and have React take over once the clientside JS loads.
webpack can do most of those things pretty well out of the box. But, it sure is a pain to set it all up.
So, this is just a simplified, opinionated way to configure webpack for development and then build for production. That also supports easily generating more files.
If no one uses it but me, it will have still served its purpose.
A screencast showing how to use this module is here: http://learn.humanjavascript.com/react-ampersand/setting-up-webpack
npm install hjs-webpack
hjs-webpack
relies on a number of optional dependencies to add functionality for things like CSS preprocessing, ES2015 transpiling, templates, and plugins. It doesn't make sense to specifiy all the loaders as peerDependencies
since not every person will want every loader and missing peerDependencies
cause commands like npm ls
to error which isn't great.
So in order to get this additional functionality you should npm install
the loaders and plugins you want hjs-webpack
to use. If hjs-webpack
detects that they are installed, then they will be used automatically without any further configuration.
Here's some more information about the available loaders and plugins and what they each do. You should install each that you want with npm install --save-dev
.
Note that all of the CSS loaders and plugins require css-loader
postcss-loader
style-loader
to be installed.
less-loader
Require compiled less files. Extension: less
.
stylus-loader
Require compiled stylus files. Extension: styl
.
sass-loader
Require compiled sass files using the regular or indented syntax. Extensions: sass scss
.
yeticss
A plugin to add the yeticss library as a stylus
plugin.
autoprefixer
A plugin to auto prefix all your CSS with the necessary vendor prefixes.
babel-loader
Require transpiled JS with built-in support for ES2015 and JSX. Extensions: js jsx babel
.
json-loader
Require parsed JSON objects. Extension: json
.
coffee-loader
Require CoffeeScript. Extension: coffee
.
cjsx-loader
Require CoffeeScript with support for JSX. coffee-loader
must also be installed. Extension: cjsx
.
awesome-typescript-loader
Require TypeScript. Extension: ts
.
livescript-loader
Require LiveScript. Extension: ls
.
url-loader
Require assets that return data url if the size is less than the urlLoaderLimit
. Extensions: jpg jpeg png gif svg otf eot svg ttf woff
.
worker-loader
This lets us more easily write code for WebWorkers. Once you npm install worker-loader
you can write worker code using whatever other transpiler you're using and being able to require
or import
code from npm just like you would normally. If this is installed, simply name your file ending in worker.js
for example main.worker.js
or even just worker.js
and it will be loaded as a worker and packaged up as a separate file when built. In addition, worker-loader
supports inlining the code for workers and loading it as a data-uri as can be seen here: https://github.com/webpack/worker-loader/blob/master/createInlineWorker.js. To use this feature name your file SOMETHING.thread.js
or just thread.js
instead and it will be inlined if that feature is supported by the browser running your app (it will fallback to being loaded as a separate file otherwise).
pug-loader
Require pug files as compiled functions. Extension: pug
(legacy jade
also supported).
visualizer-plugin
A plugin to visualize and analyze your Webpack bundle to see which modules are taking up space and which might be duplicates.
npm install --save hjs-webpack
Put it at the root of your project, a typical config looks something like this:
var getConfig = require('hjs-webpack')
module.exports = getConfig({
// entry point for the app
in: 'src/app.js',
// Name or full path of output directory
// commonly named `www` or `public`. This
// is where your fully static site should
// end up for simple deployment.
out: 'public',
// This will destroy and re-create your
// `out` folder before building so you always
// get a fresh folder. Usually you want this
// but since it's destructive we make it
// false by default
clearBeforeBuild: true
})
I usually add something like the following scripts:
"scripts": {
"start": "hjs-dev-server",
"build": "webpack",
"deploy": "npm run build && surge -p public -d somedomain.com"
}
Assuming you've got some JS written that you've set as your in
in the webpack.config.js
you can run npm start
and open a browser to http://localhost:3000
and you everything should Just Work™.
When you're wanting to do a build, just run npm run build
. The build will generate your files into public
.
Now there's a static site in public
that can be deployed to something like Surge.sh, which I do by running npm run deploy
.
Since we're using webpack under the hood, this is done the "webpack way".
Basically you can require
your styles as if they were JavaScript files.
Simply do this in your application code:
require('./path/to/your/css/main.css')
Be sure to include the extension: .css
in your require statment. If you use .styl
you can write Stylus seamlessly and at the top of your stylus files you've got access to yeti.css for easy styling.
Try creating a file called main.styl
containing:
@import 'yeticss'
Require it from your main application file (see in
section below) and you should get some nice default styles.
Note in development mode these will be live-reloaded (hot loaded). In production, these will be extracted into their own files, including intelligent handling of referenced URLs within your stylesheets. Things like font-files and images will be extracted if they're over a certain size. You shouldn't have to worry about this too much. It should just work seamlessly.
Option #1: requiring files
Webpack lets us do var url = require('something.png')
from within our app code and url
is something you can safely set as the src
of an image tag, for example. When you build the project, it uses the url-loader and will base64 encode and inline it if it's smaller than the urlLoaderLimit and hash and export it otherwise.
When you do this, webpack will hash the file and use that as a name. If you basically just want to require a file so webpack knows about it, the following syntax will copy the favicon to the out
directory (at the root) but leave the name unchanged: require('!!file?name=favicon.ico!./real/path/to/your/favicon.ico')
. The !!
at the beginning will tell webpack to ignore other configured loaders so that your favicon won't get base64 encoded by the url-loader
. See the webpack documentation about loader order for more info.
But, letting webpack handle images isn't always what you want to do. Sometimes you want just a simple folder of static assets and be able to reference them like you're used to. That's why there's another option:
Option #2: just put 'em in your out
directory
You can also just put your assests in the out
directory and tell hjs-webpack to ignore them by setting a glob pattern as the clearBeforeBuild
option.
Assume an out
directory called public
that looks like this:
public/
some-other-generated-file.html
index.html
yourapp.1.1.1.css
yourapp.1.1.1.js
favicon.ico
images/
some-pic.png
Then, instead of setting clearBeforeBuild: true
you can set it to a glob string like so: clearBeforeBuild: '!(images|favicon.ico)'
.
Now when you build it'll clear everything that matches the glob pattern an nothing else.
In this case, it'd leave the images
directory and your favicon.ico
alone (more details in options section below).
note The development server will treat the out
directory as the contentBase
which means in this case the favicon would be available at /favicon.ico
despite being in public
.
There are 3 example projects in the /examples directory with various config setups:
- Only generating CSS/JS
- Generating CSS/JS and using included HTML template
- Pre-rendering app layout and the public homepage HTML with React as part of the build process
The main export you get when you require('hjs-webpack')
is simply a pre-configured webpack.config.js
. You could take the result of that and add other plugins if you so chose, but shouldn't be necessary for most common tasks.
This should just be the path to the file that serves as the main entry point of your application.
Path to directory where we're going to put generated files.
A boolean to specify whether to clear the out
folder before building.
If you wish to only clear some of this directory you can also pass a glob string. Globs are the file path matching strings you've probably seen in on the command line or in a .gitigore
(i.e. **/*.js*
).
The most common thing you'd probably want to do while using this module would be to exclude a directory from being cleared. The following example would clear out the public
directory but leave the public/images
and public/static
folders intact if they exist.
getConfig({
in: 'src/app.js',
out: 'public',
clearBeforeBuild: '!(images|static)'
})
So, just to be clear, everything that matches the glob string within the out folder will be deleted when building.
A boolean to indicate whether or not everything is in production mode (minified, etc.) or development mode (everything hotloaded and unminified).
By default this value is true
if the command you ran contains hjs-dev-server
and false
otherwise. The option exists here in case you need to override the default.
A webpack developer tool to enhance debugging. See the webpack docs for more options.
Options passed directly to the UglifyJSPlugin
. Only used if isDev
is false
. Default:
{
compress: { warnings: false },
output: { comments: false },
sourceMap: false
}
This is passed directly to webpack
, so you can use all the configuration options available there.
By default a filename is created for you based on the following rules:
- If
isDev
istrue
, then the filename isapp.js
- If
isDev
isfalse
, then the filenameNAME.VERSION.js
whereNAME
andVERSION
are pulled from yourpackage.json
file - If
output.hash
is true, then instead ofVERSION
your filename will contain theHASH
of the compiled file
This is passed directly to the extract-text-webpack-plugin
, so you can use all the configuration options available there. Note: this is only used if isDev
is true
, since in development mode the css bundle is inserted dynamically into the document by the style-loader
.
By default a filename is created for you based on the following rules:
- If
isDev
istrue
, then the filename isapp.css
- If
isDev
isfalse
, then the filenameNAME.VERSION.css
whereNAME
andVERSION
are pulled from yourpackage.json
file - If
output.hash
is true, then instead ofVERSION
your filename will contain theHASH
of the compiled file
This is used in conjunction with the output.filename
and output.cssFilename
options above, and is only used if isDev
is false
. If hash
is true
then the filenames of your JS and CSS files will contain the hash of the compiled file. This is useful to fingerprint your asset files so that they can be cached for as long as possible.
Note that as per the suggestion in the webpack docs, the OccurenceOrderPlugin
is already used so you will get consistent ordering of modules.
This is the default threshold to use for whether URLs referenced in stylesheets will be inlined or extracted during build (we're just pre-configuring the url-loader).
These options are passed through to the hjs-dev-server
with a few defaults. Some of these options are passed directly to webpack-dev-middleware
, see those docs for all available options.
{
port, // pulled from top level option "port"
hostname, // // pulled from top level option "hostname"
historyApiFallback: true,
hot: true,
compress: true, // enable express compression to faster index reload (default: false)
// The following options are for webpack-dev-middleware
noInfo: true,
quiet: false,
lazy: false,
publicPath // pulled from top level option "output.publicPath"
}
This is used to start hjs-dev-server
with its self signed certificate, so you can load the application with an https url. It also configures hot module replacement to also use https.
You can supply an object of require names with paths to the files you want that name to represent. This makes it easy to do things like swapping out config files based on build mode, etc.
Adding this to your config would mean that any time you did: require('config')
within your application code, you'd end up with the file specified by the path.
{
'config': '/some/path/config.json'
}
This option is true
by default. This means, by default, we'll serve and generate a very basic HTML file that looks like this:
<!doctype html>
<meta charset="utf-8"/>
<meta name="viewport" content="width=device-width, initial-scale=1, user-scalable=no"/>
<link rel="stylesheet" href="/package-name.1.0.0.css"/>
<body><div id="root"></div><script src="/package-name.1.0.0.js"></script></body>
Note the <meta charset>
tag and mobile viewport settings are there by default.
The <body>
followed by the main script tag is also intentional. The ordering ensures we don't have to wait for DOMReady
in our clientside code, you can safely assume that both document.body
and document.head
will be available when your script executes.
If you just want to do JS and CSS and handle all the html yourself, simply add html: false
to your config (see examples directory for example).
using an html
function to generate specific files
This is where it gets interesting. Imagine pre-rendering all known structural content for a Native Web App to static files. Users get pixels on the screen immediately, your JS takes over when downloaded. If you're using React, this "taking over" can be completely seamless and invisible to the user. It's also possible with this approach to write an app that works entirely without JS. See the prerendered-html-files example.
Your function should produce an object.
Each key in the object is a filename, and its value is a string to be written to disc.
If you simply specify html: true
it will do the following by default:
html: function (context) {
return {
'index.html': context.defaultTemplate()
}
}
So if you want to produce other files, you can do so by adding them to the returned object:
html: function (context) {
return {
'index.html': context.defaultTemplate(),
// if you build it entirely yourself it should be a complete HTML document
// using whatever templating system you want
'other.html': '<!DOCTYPE><body><h1>Hello World</h1></body>'
}
}
async version
html: function (context, callback) {
// do whatever async stuff generate result object
// and pass it to the callback instead
db.fetchData(function (err, data) {
callback(null, {
'index.html': buildHTML(data),
'other.html': doSomethingElse(data)
})
})
}
The context argument
Your html
function will be called with a context object that contains the following:
context.main
: the name of the generated JS filecontext.css
: the name of the generated CSS file. This only exists ifisDev
isfalse
, since in development mode the css bundle is inserted dynamically into the document by thestyle-loader
.context.defaultTemplate()
a convenience method you can call to generate the basic HTML shown above. This takes a few options too if you just want to make minor tweaks. If you want to do more, just don't use the default template, generate your own instead. The options are:
{html: '<div id="my-container">Some custom markup</div>'}
This markup will be added inside the<body>
tag. By default it adds a<div id="root"></div>
as a mount target for React apps.{charset: 'utf-8'}
what charset to set{title: 'your app'}
sets<title>
{head: 'any string'}
anything else you want to put in thehead
, other meta tags, or whatnot.{metaViewport: boolean|object}
set to false if you don't want the default viewport tag. Set to an object withuserScalable
true if you don't want to block user-zoom on mobile{publicPath: 'http://mycdn.com/'}
(default/
) pass in path that will prefix the generated css/js files in the template. Note, there isoutput.publicPath
provided by webpack, but doesn't easily allow for switching based on envirnoment. In this method we've got access tocontext.isDev
and can easily switch based on that.{metaTags: {}}
lets you easily add<meta>
tags to the document head. Takes an object where the key is thename
and the value is thecontent
.{lang: 'en-US'}
sets thelang
attribute on the<html>
tag.
context.isDev
: boolean specifying whether or not we're in dev mode.context.package
: the parsedpackage.json
file as an object.context.stats
: the stats object returned by webpack. Of likely interest iscontext.stats.hash
(a hash of current build).context.stats.assets
is an array of all the assets that will be generated. This can be useful for generating cache manifests, etc. Overall, this is a big object that lists all the modules in your whole app. You likely won't need most of it, but it's all there in case you do. (A sample can be found here).
By default, if you supply an html
function it will always be used, whether you're in development mode or not.
Set this option to false
to only use your html
function when building for production. Note, that .isDev
is attached to the context object passed to the html
function as described above, so alternately you could just use that value to branch your logic within that function. Using this option circumvents the custom html
function entirely during development.
The dev server uses http-proxy-middleware to optionally proxy requests to a separate, possibly external, backend server. Proxies can be specified with devServer.proxy
. This can be a single proxy, or an array of proxies. The proxy context and options are passed directly to http-proxy-middleware
.
getConfig({
in: 'src/app.js',
out: 'public',
clearBeforeBuild: true,
// Use devServer.proxy to specify proxies
devServer: {
proxy: {
context: "/api",
options: {
target: "http://localhost:3001",
pathRewrite: {
"^/api": ""
}
}
}
}
})
If you're building an app that you want to look good on all devices it's nice to be able to run them all at once.
Hotloading makes this extremely nice and convenient.
If you're on a Mac, this is fairly simple. Just add a hostname
option to your config like so:
module.exports = getConfig({
in: 'src/app.js',
out: 'public',
// set this to whatever your machine name is
// plus `.local`
// my machine is `loki` so I do:
hostname: 'loki.local'
})
Now when you run the development instead of going to localhost open: http://{{yourmachine}}.local:3000
on any device that's on your local network, they should all connect and all hotload your style and JS changes.
hjs-webpack
is not designed to take all the same options as webpack
. Instead it is designed to take the config options listed above and return an object that is then consumed by webpack
. That means that if you want to change/add/remove anything in the config, it is the same as manipulating any JavaScript object.
Here's an example where hjs-webpack
is used to create the base webpack
config, and then it is manipulated to add a new loader, plugin, and option.
var webpack = require('webpack')
var getConfig = require('hjs-webpack')
var config = getConfig(myHjsWebpackOptions)
// Add xml-loader
config.module.loaders.push({ test: /\.xml$/, loader: 'xml-loader' })
// Add module noParse option
config.module.noParse = [/dont-parse-this/, /also-this/]
// Add webpack PrefetchPlugin
config.plugins.push(new webpack.PrefetchPlugin([context], request))
// Add a separate entry point for jQuery
config.resolve.alias = { jquery:'jquery/src/jquery.js' }
config.plugins.push(
new webpack.ProvidePlugin({
jQuery: 'jquery',
$: 'jquery',
'window.jQuery':'jquery'
}),
new webpack.optimize.CommonsChunkPlugin('vendors', 'vendors.js')
);
config.entry = {
// Add entries for vendors
vendors: ['jquery'],
// Reassign previous single entry to main entry
main: config.entry
};
// Export the newly manipulated config
module.exports = config
Since hjs-webpack
already has a babel loader, the easiest way to tweak Babel settings is to create a file at the root of your project called .babelrc
that contains config settings. See bablerc docs for more options.
There are some babel presets that work well with hjs-webpack
. You can check out an example of using presets in the examples directory. There's one with hot reloading and one without. You'll need to install these presets just like any other dev dependencies.
Here's a quick example if you like copy/pasting:
npm install babel-preset-es2015 babel-preset-react babel-preset-react-hmre --save-dev
and then your .babelrc
{
"presets": ["es2015", "react"],
"env": {
"development": {
"presets": ["react-hmre"]
}
}
}
This is mostly just some add-ons to webpack so most of the credit goes there.
If you're interested in building apps this way, watch the free section of the tutorials at http://learn.humanjavascript.com. It shows basic usage of this module. Also, you can follow me on twitter @HenrikJoreteg.
Big thanks to co-maintainer @LukeKarrys for helping find/fix some really annoying bugs.
Beware that this is all highly opinionated and contains a lot of personal preferences. If you want to add or remove major things, feel free to open issues or send PRs, but you may just want to fork it.
See the CHANGELOG.md
MIT