UPDATE (July 7th): babel-plugin-dual-import is now required to asynchronously import both css + js. Much Faster Builds! You likely want to read its intro article.
UPDATE (July 26th): babel-plugin-universal-import is what to use if you're using React Universal Component.
Like extract-text-webpack-plugin
, but creates multiple css files (one per chunk). Then, as part of server side rendering, you can deliver just the css chunks needed by the current request. The result is the most minimal CSS initially served compared to emerging "render path" solutions.
For a demo, git clone
: universal-demo
Note: this is a companion package to:
- webpack-flush-chunks
- react-universal-component
- babel-plugin-universal-import or babel-plugin-dual-import
yarn add react-universal-component webpack-flush-chunks
yarn add --dev extract-css-chunks-webpack-plugin babel-plugin-universal-import
.babelrc:
{
"plugins": ["universal-import"]
}
webpack.config.js:
const ExtractCssChunks = require("extract-css-chunks-webpack-plugin")
module.exports = {
module: {
rules: [
{
test: /\.css$/,
use: ExtractCssChunks.extract({
use: {
loader: 'css-loader',
options: {
modules: true,
localIdentName: '[name]__[local]--[hash:base64:5]'
}
}
})
}
]
},
plugins: [
new ExtractCssChunks,
]
}
Here's the sort of CSS you can expect to serve:
<head>
<link rel='stylesheet' href='/static/main.css' />
<link rel='stylesheet' href='/static/0.css' />
<link rel='stylesheet' href='/static/7.css' />
</head>
<body>
<div id="react-root"></div>
<script type='text/javascript' src='/static/vendor.js'></script>
<script type='text/javascript' src='/static/0.js'></script>
<script type='text/javascript' src='/static/7.js'></script>
<script type='text/javascript' src='/static/main.js'></script>
<!-- stylsheets that will be requested when import() on user navigation is called -->
<script>
window.__CSS_CHUNKS__ = {
Foo: '/static/Foo.css',
Bar: '/static/Bar.css'
}
</script>
</body>
webpack-flush-chunks will scoop up the exact stylesheets to embed in your response. It essentially automates producing the above.
Here's how you do it:
src/components/App.js:
const UniversalComponent = universal(props => import(`./${props.page}`))
<UniversalComponent page='Foo' />
server/render.js:
import { flushChunkNames } from 'react-universal-component/server'
import flushChunks from 'webpack-flush-chunks'
const app = ReactDOMServer.renderToString(<App />)
const { js, styles, cssHash } = flushChunks(webpackStats, {
chunkNames: flushChunkNames()
})
res.send(`
<!doctype html>
<html>
<head>
${styles}
</head>
<body>
<div id="root">${app}</div>
${cssHash}
${js}
</body>
</html>
`)
As for asynchronous calls to import()
on user navigation, babel-plugin-universal-import is required if you're using react-universal-component. And if you aren't, you must use: babel-plugin-dual-import.
These babel plugins request both your js + your css. Very Nice! This is the new feature of the 2.0. Read Sokra's (author of webpack) article on how on how this is the future of CSS for webpack. Use this and be in the future today.
- HMR: It also has first-class support for Hot Module Replacement across ALL those css files/chunks!!!
- cacheable stylesheets
- smallest total bytes sent compared to "render-path" css-in-js solutions that include your CSS definitions in JS
- Faster than the V1!
You can pass the same options as extract-text-webpack-plugin
to new ExtractCssChunks
, such as:
new ExtractCssChunk({
filename: '[name].[contenthash].css'
})
Keep in mind, by default [name].css
is used when process.env.NODE_ENV === 'deveopment'
and [name].[contenthash].css
during production, so you can likely forget about having to pass anything.
The 2 exceptions are: allChunks
will no longer do anything, and fallback
will no longer do anything when passed to to extract
. Basically just worry about passing your css-loader
string and localIdentName
🤓
The most common workflow when working with webpack is to write a "development" / "production" value in to the 'process.env.NODE_ENV' namespace, typically, using webpack's built-in "DefinePlugin" plugin. e.g:
new webpack.DefinePlugin( {
'process.env': {
NODE_ENV: `"${config.devMode ? 'development' : 'production'}"`,
},
} )
The value set by the aformentioned plugin will only be available in the runtime ( when webpack's javascript output is excuted ). In order for this plugin to work with hot module reloading, The npm script / gulp task / grunt task etc, should be invoked with the same environment variable as shown above.
For example, when running the build using some form of npm script:
{
"scripts": {
"build": "cross-env NODE_ENV=development webpack --config build/webpack.config.js"
}
}
cross-env is optional but recomanded
If you effectively use code-splitting, Exract Css Chunks can be a far better option than using emerging solutions like StyleTron, StyledComponents, and slightly older tools like Aphrodite, Glamor, etc. We aren't fans of either rounds of tools because of several issues, but particularly because they all have a runtime overhead. Every time your React component is rendered with those, CSS is generated and updated within the DOM. On the server, you're going to also see unnecessary cycles for flushing the CSS along the critical render path. Next.js's styled-jsx
, by the way, doesn't even work on the server--not so good when it comes to flash of unstyled content (FOUC).
The reason Extract CSS Chunk can be a better option is because we also generate multiple sets of CSS based on what is actually "used", but without the runtime overhead. The difference is our definition of "used" is modules determined statically (which may not in fact be rendered) vs. what is in the "critical render path" (as is the case with the other tools).
So yes, our CSS may be mildly larger and include unnecessary css, but our no_css.js
bundles will be a lot smaller as they don't need to inject any styles. See, even though those solutions have the smallest possible CSS file size, their javascript bundles, in order to continue to render your components styled properly on the client, must contain the necessary CSS for all posibilities! Those solutions serve both your entire bundle's CSS (in your javascript) and the CSS flushed from the critical render path.
On top of that, those are extra packages all with a huge number of issues in their Github repos corresponding to various limitations in the CSS they generate--something that is prevented when your definition for "CSS-in-JS" is simply importing CSS files compiled as normal by powerful proven CSS-specific processors.
Lastly, those solutions don't provide cacheable stylesheets. They do a lot of work--but they will continue doing it for you when you could have been done in one go long ago. Cloudflare is free--serve them through their CDN and you're winning. I love true javascript in css--don't get me wrong--but first I'd have to see they generate cacheable stylesheets. In my opinion, for now, it's best for environments that natively support it such as React Native.
Now with that out of the way, for completeness let's compare what bytes of just the CSS are sent over the wire. The difference basically is minimal. Whereas solutions that flush from the critical render path will capture no more than the precise bits of CSS from the if/else
branches followed, Extract Css Chunks is all about you effectively using code-splitting to insure, for example, you don't serve your Administration Panel's CSS in your public-facing site, etc. In other words, it's all about avoiding serving large swaths of CSS from completely different sections of your app. I.e. the biggest gains available to you.
This is where the real problem lies--where the real amount of unnecessary CSS comes from. If you effectively code-split your app, you may end up with 10 chunks. Now you can serve just the corresonding CSS for that chunk. If you try to go even farther and remove the css not used in the render path, you're likely acheiving somewhere between 1-20% of the gains you achieved by thorough code-splitting. In other words, code splitting fulfills the 80/20 rule and attains the simple sweetspot of 80% optimization you--as a balanced, level-headed, non-neurotic and professional developer--are looking to achieve.
In short, by putting code splitting in appropriate places you have a lot of control over the CSS files that are created and can send a sensible minimal amount of associated bytes over the wire for the first request, perhaps even the smallest amount of all options.
It's our perspective that you have achieved 80-99% of the performance gains (i.e. the creation of small relevant css files) at this static stage. Offloading this work to the runtime stage is ultimately nit-picking and results in diminishing returns. When you factor that your JS bundle has to contain that JS anyway, those solutions make less and less sense from the perspective of reducing bytes delivered in the initial request.
There may be other reasons to use those tools (e.g. you don't like setting up webpack configs, or somehow you're really fond of pre-creating many <div />
elements with their styles), but we prefer a simple standards-based way (without HoCs or specialized style components) to import styles just as you would in React Native. However to give the other tools credit, many of them likely started out with a different problem motivating them: avoiding webpack configs so you can include packages and their contained CSS without client apps being required to setup something like CSS loaders in Webpack. Having your CSS completely contained in true JS has its use cases, but at the application level--especially when you're already using something like Webpack--we fail to see its benefits. About all they share is a solution to avoiding flashes of unstyled content (FOUC), except one can save you a lot more bytes in what you send over the wire and save you from a continual runtime overhead where it's not needed. Honorable Mention: StyleTron's concept of "atomic declaration-level deduplication" where it will make a class out of, say, color: blue
so you don't need to send redundant styles certainly is a novel innovation, but again if the code still exists in your JS and you're building an application using Webpack (instead of a package), what's the point. In fact, it just makes editing the stylesheets in your browser developer tools more complicated. One benefit of critical render path solutions is the browser can spend less time matching the smaller number of styles to new DOM nodes as they appear, but then again it also has to spend the time injecting and parsing the new styles constantly, which is likely costlier.
As an aside, so many apps share code between web and React Native--so the answer to the styles problem must be one that is identical for both. From that perspective importing a styles object still makes a lot of sense. You're not missing out on the fundamental aspect of CSS-in-JSS: isolated component-level styles and the ability to import them just like any other javascript code. Put them in other files, and use tools like extract-css-chunks-webpack-plugin
and your pre-processors of choice that innately get styles right for the browser. As long as you're using CSS Modules, you're still at the cutting edge of CSS-in-JSS. Let's just say the other tools took one wrong turn and took it too far, when we already were at our destination.
SUMMARY OF BENEFITS COMPARED TO "CRITICAL-RENDER-PATH" SOLUTIONS:
- no continual runtime overhead during
render
using HoCs that inject styles - smaller JS bundles without CSS injection
- You DO NOT need to clutter your component code with a specialized way of applying CSS (HoCs, styled elements)!
- The way you import module-based styles is exactly how you would import styles in React Native that exist in a separate file, which allows for extremely interchangeable code between React Native and regular React. Hurray!
- pretty much already does everything covered by @vjeux's 2014 css-in-js talk, besides dead code elimination. Dead code elimination is only solved for the other tools--as per the explanation above how the CSS is in the JS anyway--so much as Webpack and Uglify can remove JS that is not used. Either way, it's not a stretch to eventually add this feature to
extract-text-webpack-plugin
as well as this plugin. Hey, maybe it already has it??
Emotion is different. They allow for the extraction of static styles via their extract mode. We're very much looking forward to this being the perfect companion to the css chunks approach.
Currently however extract mode does not support IE11. So that means it's a no go, but we have hopes that in the future they'll solve that problem.
The reason Emotion doesn't work in IE11+ is because they currently try to preserve any dynamic aspects of your CSS-in-JS by converting it CSS vars, which isn't supported in IE11. That was a very smart approach, but unfortunately not good enough.
The vision we'd like to see for that package is where dynamic css stays inline, and where only static CSS is extracted into stylesheets, in which case CSS vars aren't needed. I've heard from them they have some "hidden flags" that allow for something close to this. When, and if, they take this feature all the way, look forward to us pushing it as our recommended approach. Go Emotion!
We love CSS modules; no less, no more.
Long live the dream of Code Splitting Everywhere!
We use commitizen, so run npm run cm
to make commits. A command-line form will appear, requiring you answer a few questions to automatically produce a nicely formatted commit. Releases, semantic version numbers, tags and changelogs will automatically be generated based on these commits thanks to semantic-release. Be good.
- redux-first-router. It's made to work perfectly with Universal. Together they comprise our "frameworkless" Redux-based approach to what Next.js does (splitting, SSR, prefetching, and routing). People are lovin it by the way 😎