universal-webpack
For beginners: consider trying Next.js first: it's user-friendly and is supposed to be a good start for people not wanting to deal with configuring Webpack manually. On the other hand, if you're an experienced Webpack user then setting up universal-webpack
shouldn't be too difficult.
This library generates client-side and server-side configuration for Webpack therefore enabling seamless client-side/server-side Webpack builds. Requires some initial set up and some prior knowledge of Webpack.
Install
npm install universal-webpack --save-dev
Example project
You may refer to this sample project as a reference example of using this library (see webpack
directory, package.json
and rendering-service/main.js
).
Use
Suppose you have a typical webpack.config.js
file. Create two new files called webpack.config.client.babel.js
and webpack.config.server.babel.js
with the following contents:
webpack.config.client.babel.js
import { client } from 'universal-webpack/config'
import settings from './universal-webpack-settings'
import configuration from './webpack.config'
// Create client-side Webpack config.
export default client(configuration, settings)
webpack.config.server.babel.js
import { server } from 'universal-webpack/config'
import settings from './universal-webpack-settings'
import configuration from './webpack.config'
// Create server-side Webpack config.
export default server(configuration, settings)
Where ./universal-webpack-settings.json
is a configuration file for universal-webpack
(see below, leave empty for now).
Now, use webpack.config.client.babel.js
instead of the old webpack.config.js
for client side Webpack builds. Your setup also most likely differentiates between a "development" client side Webpack build and a "production" one, in which case webpack.config.client.babel.js
is further split into two files — webpack.config.client.dev.babel.js
and webpack.config.client.prod.babel.js
— each of which inherits from webpack.config.client.babel.js
and makes the necessary changes to it as defined by your particular setup.
And, webpack.config.server.babel.js
file will be used for server-side Webpack builds. And, analogous to the client-side config, it also most likely is gonna be split into "development" and "production" configs, as defined by your particular setup.
Setting up the server side requires an additional step: creating the "entry" file for running the server. The reason is that client-side config is created from webpack.config.js
which already has Webpack "entry" defined. Usually it's something like ./src/index.js
which is the "main" file for the client-side application. Server-side needs such a "main" file too and the path to it must be configured in ./universal-webpack-settings.json
as server.input
:
universal-webpack-settings.json
{
"server":
{
"input": "./source/server.js",
"output": "./build/server/server.js"
}
}
With the server-side "entry" file path configured, the server-side config created by this library will have the Webpack "entry" parameter set up properly. A "server-side" Webpack build will now produce a "server-side" bundle (./build/server/server.js
) which can be run using Node.js. An example of an "entry" file:
source/server.js
import path from 'path'
import http from 'http'
import express from 'express'
import httpProxy from 'http-proxy'
// React routes.
// (shared with the client side)
import routes from '../client/routes.js'
// Redux reducers.
// (shared with the client side)
import reducers from '../client/reducers.js'
// Starts the server.
function startServer()
{
// Create HTTP server.
const app = new express()
const server = new http.Server(app)
// Serve static files.
app.use(express.static(path.join(__dirname, '..', 'build/assets')))
// Proxy API calls to API server.
const proxy = httpProxy.createProxyServer({ target: 'http://localhost:xxxx' })
app.use('/api', (req, res) => proxy.web(req, res))
// React application rendering.
app.use((req, res) => {
// Match current URL to the corresponding React page.
routerMatchURL(routes, req.originalUrl).then((error, routingResult) => {
if (error) {
throw error
}
// Render React page.
const page = createPageElement(routingResult, reducers)
res.status(200)
res.send('<!doctype html><html>...' + ReactDOM.renderToString(page) + '...</html>')
})
.catch((error) => {
res.status(500)
return res.send('Server error')
})
})
// Start the HTTP server.
server.listen()
}
// Run the server.
startServer()
The main use-case for universal-webpack
though is most likely "Server-Side Rendering" which means that the server is somehow gonna need to know the actual URLs for the compiled javascript and CSS files (which contain random-generated md5 hashes). Specifically for this case this library provides a special "runner" for the server-side bundle which requires that the server-side bundle just exports a "start server" function, without actually running the server, and then such "start server" function will be called with a special parameters
argument which holds the actual URLs for the compiled javascript and CSS files (see the "Chunks" section below).
So in this case the changes to the server file are gonna be:
source/server.js
...
export default function startServer(parameters) {
...
}
// Don't start the server manually.
// // Run the server.
// startServer()
And the server-side runner will be called like this:
source/start-server.js
// The runner.
var startServer = require('universal-webpack/server')
// The server-side bundle path info.
var settings = require('../universal-webpack-settings')
// Only `configuration.context` and `configuration.output.path`
// parameters are used from the whole Webpack config.
var configuration = require('../webpack.config')
// Run the server.
startServer(configuration, settings)
Running node source/start-server.js
will basically call the function exported from source/server.js
with the parameters
argument.
Finally, to run all the things required for "development" mode (in parallel):
# Client-side build.
webpack-dev-server --hot --config ./webpack.config.client.dev.babel.js
# Server-side build.
webpack --watch --config ./webpack.config.server.dev.babel.js --colors --display-error-details
# Run the server.
nodemon ./source/start-server --watch ./build/server
For production mode the command sequence would be:
# Build the client.
webpack --config "./webpack.config.client.babel.js" --colors --display-error-details
# Build the server.
webpack --config "./webpack.config.server.babel.js" --colors --display-error-details
# Run the server.
node "./source/start-server"
Chunks
This library will pass the chunks()
function parameter (inside the parameters
argument of the server-side function) which returns webpack-compiled chunks filename info:
build/webpack-chunks.json
{
javascript:
{
main: `/assets/main.785f110e7775ec8322cf.js`
},
styles:
{
main: `/assets/main.785f110e7775ec8322cf.css`
}
}
These filenames are required for <script src=.../>
and <link rel="style" href=.../>
tags in case of isomorphic (universal) rendering on the server-side.
Gotchas
- It emits no assets on the server side so make sure you include all assets on the client side (e.g. "favicon").
resolve.root
won't work out-of-the-box whileresolve.alias
es do. For those usingresolve.root
I recommend switching toresolve.alias
. By default no "modules" are bundled in a server-side bundle except forresolve.alias
es andexcludeFromExternals
matches (see below).
mini-css-extract-plugin
Using The third argument – options
object – may be passed to client()
configuration function. If options.development
is set to false
, then it will apply mini-css-extract-plugin
to CSS styles automatically, i.e. it will extract all CSS styles into separate *.css
files (one for each Webpack "chunk"): this is considered a slightly better approach for production deployment instead of just leaving all CSS in *.js
chunk files (due to static file caching in a browser). Using options.development=false
option is therefore just a convenience shortcut which one may use instead of adding mini-css-extract-plugin
to production client-side webpack configuration manually.
import { clientConfiguration } from 'universal-webpack'
import settings from './universal-webpack-settings'
import baseConfiguration from './webpack.config'
const configuration = clientConfiguration(baseConfiguration, settings, {
// Extract all CSS into separate `*.css` files (one for each chunk)
// using `mini-css-extract-plugin`
// instead of leaving that CSS embedded directly in `*.js` chunk files.
development: false
})
Advanced configuration
./universal-webpack-settings.json
configuration file also supports the following optional configuration parameters:
{
// By default, all `require()`d packages
// (e.g. everything from `node_modules`, `resolve.modules`),
// except for `resolve.alias`ed ones,
// are marked as `external` for server-side Webpack build
// which means they won't be processed and bundled by Webpack,
// instead being processed and `require()`d at runtime by Node.js.
//
// With this setting one can explicitly define which modules
// aren't gonna be marked as `external` dependencies.
// (and therefore are gonna be compiled and bundled by Webpack)
//
// Can be used, for example, for ES6-only `node_modules`.
// ( a more intelligent solution would be accepted
// https://github.com/catamphetamine/universal-webpack/issues/10 )
//
excludeFromExternals:
[
'lodash-es',
/^some-other-es6-only-module(\/.*)?$/
],
// As stated above, all files inside `node_modules`, when `require()`d,
// would be resolved as "externals" which means Webpack wouldn't use
// loaders to process them, and therefore `require()`ing them
// would result in an error when running the server-side bundle.
//
// E.g. for CSS files Node.js would just throw `SyntaxError: Unexpected token .`
// because these CSS files need to be compiled by Webpack's `css-loader` first.
//
// Hence the "exclude from externals" file extensions list
// which by default is initialized with some common asset types:
//
loadExternalModuleFileExtensions:
[
'css',
'png',
'jpg',
'svg',
'xml'
],
// Enable `silent` flag to prevent client side webpack build
// from outputting chunk stats to the console.
silent: true,
// By default, chunk_info_filename is `webpack-chunks.json`
chunk_info_filename: 'submodule-webpack-chunks.json'
}
Source maps
I managed to get source maps working in my Node.js server-side code using source-map-support
module.
source/start-server.js
// Enables proper source map support in Node.js
require('source-map-support/register')
// The rest is the same as in the above example
var startServer = require('universal-webpack/server')
var settings = require('../universal-webpack-settings')
var configuration = require('../webpack.config')
startServer(configuration, settings)
Without source-map-support
enabled it would give me No element indexed by XXX
error (which means that by default Node.js thinks there are references to other source maps and tries to load them but there are no such source maps).
devtool
is set to source-map
for server-side builds.
Nodemon
I recommend using nodemon for running server-side Webpack bundle. Nodemon has a --watch <directory>
command line parameter which restarts Node.js process each time the <directory>
is updated (e.g. each time any file in that directory is modified).
In other words, Nodemon will relaunch the server every time the code is rebuilt with Webpack.
There's one little gotcha though: for the --watch
feature to work the watched folder needs to exist by the time Nodemon is launched. That means that the server must be started only after the settings.server.output
path folder has been created.
To accomplish that this library provides a command line tool: universal-webpack
. No need to install in globally: it is supposed to work locally through npm scripts. Usage example:
package.json
...
"scripts": {
"start": "npm-run-all prepare-server-build start-development-workflow",
"start-development-workflow": "npm-run-all --parallel development-webpack-build-for-client development-webpack-build-for-server development-start-server",
"prepare-server-build": "universal-webpack --settings ./universal-webpack-settings.json prepare",
...
The prepare
command creates settings.server.output
path folder, or clears it if it already exists.
Note: In a big React project server restart times can reach ~10 seconds.
Flash of unstyled content
A "flash of unstyled content" is a well-known dev-mode Webpack phenomenon. One can observe it when refreshing the page in development mode: because Webpack's style-loader
adds styles to the page dynamically there's a short period of time (a second maybe) when there are no CSS styles applied to the webpage (in production mode mini-css-extract-plugin
is used instead of style-loader
so there's no "flash of unstyled content").
It's not really a bug, because it's only for development mode. Still, if you're a perfectionist then it can be annoying. The most basic workaround for this is to simply show a white "smoke screen" and then hide it after a pre-defined timeout.
import { smokeScreen, hideSmokeScreenAfter } from 'universal-webpack'
<body>
${smokeScreen}
</body>
<script>
${hideSmokeScreenAfter(100)}
</script>
resolve.moduleDirectories
If you were using resolve.moduleDirectories
for global paths instead of relative paths in your code then consider using resolve.alias
instead
resolve:
{
alias:
{
components: path.resolve(__dirname, '../src/components'),
...
}
}
universal-webpack
vs webpack-isomorphic-tools
Note: If you never heard of webpack-isomorphic-tools
then you shouldn't read this section.
webpack-isomorphic-tools
runs on the server-side and hooks into Node.js require()
function with the help of require-hacker
and does what needs to be done.
universal-webpack
doesn't hook into require()
function - it's just a helper for transforming client-side Webpack configuration to a server-side Webpack configuration. It doesn't run on the server-side or something. It's just a Webpack configuration generator - turned out that Webpack has a target: "node"
parameter which makes it output code that runs on Node.js without any issues.
I wrote webpack-isomorphic-tools
before universal-webpack
, so universal-webpack
is the recommended tool. However many people still use webpack-isomorphic-tools
(including me) and find it somewhat less complicated for beginners.
GitHub
On March 9th, 2020, GitHub, Inc. silently banned my account (and all my libraries) without any notice. I opened a support ticked but they didn't answer. Because of that, I had to move all my libraries to GitLab.