What is CRA?
Create React App is a toolchain built and maintained by developers at Facebook for bootstrapping React applications. You simply run one command and Create React App sets up the tools you need to start your React project.
Advantages of CRA
- Get started with a single command
npx create-react-app my-app
- Less to Learn. You can just focus on React alone and don't have to worry about webpack, babel, and other such build dependencies.
- Only one build dependency
react-scripts. This maintains all your build dependencies, so it's easy to maintain and upgrade with just one command.
npm install react-scripts@latest
Disadvantages of CRA
- Difficult to add custom build configs. One way to add custom configs is to eject the app, but then it overrides the Only one build dependency advantage. The other way is you can use packages like customize-cra or react-app-rewired but then they have limited capabilities.
- Abstracts everything. It's important to understand the things that need to run a React app. But due to it's Only one build dependency advantage, a beginner might think that
react-scriptsis the only dependency needed to run react apps and might not know that transpiler(babel), bundler(webpack) are the key dependencies which are used under the hood byreact-scripts. This happened to me until I read this awesome article. - CRA is bloated - IMO. For example, CRA comes with SASS support, if you are using
plain CSSorLessit's an extra dependency that you will never use. Here is a package.json of an ejected CRA app.
The alternative for CRA is to set up your own boilerplate. The only advantage that we can take from CRA is Get started with a single command and we can eliminate all of its disadvantages by setting up dependencies and configs by ourselves. We cannot take the other two advantages because it introduces two disadvantages(Abstracts everything and Difficult to add custom build configs).
This repo has all the code used in this blog post.
First, initialize your project with npm and git
npm init
git init
Let's quickly create a .gitignore file to ignore the following folders
node_modules
build
Now, let's look at what are the basic dependencies that are needed to run a React app.
react and react-dom
These are the only two runtime dependencies you need.
npm install react react-dom --save
Transpiler(Babel)
Transpiler converts ECMAScript 2015+ code into a backward-compatible version of JavaScript in current and older browsers. We also use this to transpile JSX by adding presets.
npm install @babel/core @babel/preset-env @babel/preset-react --save-dev
A simple babel config for a React app looks like this. You can add this config in .babelrc file or as a property in package.json.
{
"presets": [
"@babel/preset-env",
"@babel/preset-react"
]
}
You can add various presets and plugins based on your need.
Bundler(Webpack)
Bundler bundles your code and all its dependencies together in one bundle file(or more if you use code splitting).
npm install webpack webpack-cli webpack-dev-server babel-loader css-loader style-loader html-webpack-plugin --save-dev
A simple webpack.config.js for React application looks like this.
const path = require('path');
const HtmlWebPackPlugin = require('html-webpack-plugin');
module.exports = {
output: {
path: path.resolve(__dirname, 'build'),
filename: 'bundle.js',
},
resolve: {
modules: [path.join(__dirname, 'src'), 'node_modules'],
alias: {
react: path.join(__dirname, 'node_modules', 'react'),
},
},
module: {
rules: [
{
test: /\.(js|jsx)$/,
exclude: /node_modules/,
use: {
loader: 'babel-loader',
},
},
{
test: /\.css$/,
use: [
{
loader: 'style-loader',
},
{
loader: 'css-loader',
},
],
},
],
},
plugins: [
new HtmlWebPackPlugin({
template: './src/index.html',
}),
],
};
You can add various loaders based on your need. Check out my blog post on webpack optimizations where I talk about various webpack configs that you can add to make your React app production-ready.
That is all the dependencies we need. Now let's add an HTML template file and a react component.
Let's create src folder and add index.html
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<title>React Boilerplate</title>
</head>
<body>
<div id="root"></div>
</body>
</html>
Let's create a HelloWorld.js react component in the src folder
import React from 'react';
const HelloWorld = () => {
return (
<h3>Hello World</h3>
);
};
export default HelloWorld;
Let's add index.js file to the src folder
import React from 'react';
import { render } from 'react-dom';
import HelloWorld from './HelloWorld';
render(<HelloWorld />, document.getElementById('root'));
Finally, let's add the start and build scripts in package.json
"scripts": {
"start": "webpack-dev-server --mode=development --open --hot",
"build": "webpack --mode=production"
}
That is it. Now our react app is ready to run. Try the commands npm start and npm run build.
Now, let's implement the Get started with a single command advantage from CRA. Basically, we are going to use an executable JS file that runs when we type a specific command(your boilerplate name) in the command line. Eg. reactjs-boilerplate new-project For this, we are going to use bin property in package.json.
Let's first create the executable JS file. Install fs-extra
npm i fs-extra
Create bin/start.js file on your project root with the following content.
#!/usr/bin/env node
const fs = require("fs-extra");
const path = require("path");
const https = require("https");
const { exec } = require("child_process");
const packageJson = require("../package.json");
const scripts = `"start": "webpack-dev-server --mode=development --open --hot",
"build": "webpack --mode=production"`;
const babel = `"babel": ${JSON.stringify(packageJson.babel)}`;
const getDeps = (deps) =>
Object.entries(deps)
.map((dep) => `${dep[0]}@${dep[1]}`)
.toString()
.replace(/,/g, " ")
.replace(/^/g, "")
// exclude the dependency only used in this file, nor relevant to the boilerplate
.replace(/fs-extra[^\s]+/g, "");
console.log("Initializing project..");
// create folder and initialize npm
exec(
`mkdir ${process.argv[2]} && cd ${process.argv[2]} && npm init -f`,
(initErr, initStdout, initStderr) => {
if (initErr) {
console.error(`Everything was fine, then it wasn't:
${initErr}`);
return;
}
const packageJSON = `${process.argv[2]}/package.json`;
// replace the default scripts
fs.readFile(packageJSON, (err, file) => {
if (err) throw err;
const data = file
.toString()
.replace(
'"test": "echo \\"Error: no test specified\\" && exit 1"',
scripts
)
.replace('"keywords": []', babel);
fs.writeFile(packageJSON, data, (err2) => err2 || true);
});
const filesToCopy = ["webpack.config.js"];
for (let i = 0; i < filesToCopy.length; i += 1) {
fs.createReadStream(path.join(__dirname, `../${filesToCopy[i]}`)).pipe(
fs.createWriteStream(`${process.argv[2]}/${filesToCopy[i]}`)
);
}
// npm will remove the .gitignore file when the package is installed, therefore it cannot be copied, locally and needs to be downloaded. Use your raw .gitignore once you pushed your code to GitHub.
https.get(
"https://raw.githubusercontent.com/Nikhil-Kumaran/reactjs-boilerplate/master/.gitignore",
(res) => {
res.setEncoding("utf8");
let body = "";
res.on("data", (data) => {
body += data;
});
res.on("end", () => {
fs.writeFile(
`${process.argv[2]}/.gitignore`,
body,
{ encoding: "utf-8" },
(err) => {
if (err) throw err;
}
);
});
}
);
console.log("npm init -- done\n");
// installing dependencies
console.log("Installing deps -- it might take a few minutes..");
const devDeps = getDeps(packageJson.devDependencies);
const deps = getDeps(packageJson.dependencies);
exec(
`cd ${process.argv[2]} && git init && node -v && npm -v && npm i -D ${devDeps} && npm i -S ${deps}`,
(npmErr, npmStdout, npmStderr) => {
if (npmErr) {
console.error(`Some error while installing dependencies
${npmErr}`);
return;
}
console.log(npmStdout);
console.log("Dependencies installed");
console.log("Copying additional files..");
// copy additional source files
fs.copy(path.join(__dirname, "../src"), `${process.argv[2]}/src`)
.then(() =>
console.log(
`All done!\n\nYour project is now ready\n\nUse the below command to run the app.\n\ncd ${process.argv[2]}\nnpm start`
)
)
.catch((err) => console.error(err));
}
);
}
);
Now let's map the executable JS file with a command. Paste this in your package.json
"bin": {
"your-boilerplate-name": "./bin/start.js"
}
Now let's link the package(boilerplate) locally by running
npm link
Now, when this command is typed in the terminal(command prompt), your-boilerplate-name my-app, our start.js executable is invoked and it creates a new folder named my-app, copies package.json, webpack.config.js, gitignore, src/ and installs the dependencies inside my-app project.
Great, now this works in your local. You can bootstrap React projects(with your own build configs) with just a single command.
You can also go one step further and publish your boilerplate to npm registry. First, commit and push your code to GitHub and follow these instructions.
Hurray! We created our alternative to create-react-app within a few minutes, which is not bloated(you can add dependencies as per your requirement) and easier to add/modify build configs.
Of course, our set up is very minimal, and it's certainly not ready for production. You have to add a few more webpack configs to optimize your build.
I've created a reactjs-boilerplate with the production-ready build set up, with linters and pre-commit hooks. Give it a try. Suggestions and contributions are welcome.
Recap
- We saw the advantages and disadvantages of CRA.
- We decided to take Get started with a single command advantage from CRA and implement it in our project and eliminate all of its drawbacks.
- We added minimal webpack and babel configs required to run a react application
- We created a HelloWorld.js react component, ran it using dev server, and build it.
- We created an executable JS file and mapped it with a command name via bin property in the package.json.
- We used
npm linkto link our boilerplate and made our boilerplate to bootstrap new react projects with a single command.
That's it, folks, Thanks for reading this blog post. Hope it's been useful for you. Please do comment your questions and suggestions.
References
- https://medium.com/netscape/a-guide-to-create-a-nodejs-command-line-package-c2166ad0452e
- https://github.com/Nikhil-Kumaran/reactjs-boilerplate
- https://reactjs.org/docs/create-a-new-react-app.html#creating-a-toolchain-from-scratch
- https://medium.com/the-node-js-collection/modern-javascript-explained-for-dinosaurs-f695e9747b70
Top comments (62)
Not all advantages and disadvantages are created equal: having to create and maintain your own build stack is just a single disadvantage, but it's a huge one.
Maintaining is not that difficult. You just have to be informed with new webpack and babel releases just like you do for React and other feature releases.
There is a reason why a great many developers abhor the word 'just'.
CRA is not boilerplate. It's build tool itself. You can DIY same kind of setup but you are really underestimating how much work it is to keep it up to date. CRA is managing build toolchain versioning for all "hidden" dependencies. When project lasts more than a couple of months, it's huge benefit.
Thank you for putting this together. A really worthwhile exercise that- even if you don't use it- helps students understand what is going on inside all of those packages and extra files in CRA.
IMO, dealing with Facebook baked-in vulnerabilities is an even bigger problem.
I really have to disagree with the premise. Maintaining the whole build process is a lot of work and very error prone.
I usually go a step futher and use the react-boilerplate template for create-react-app:
This way I have a full structure and thoughtfully setup react app in minutes. With typescript, redux, redux-saga, styled-components included.
Sure, it ultimately boils down to your requirement. I'd like to have control over the configs and I'm confident about maintaining the build process so I gave it a try🙂
I like this article and agree with some of the points. Most important one is that you need to know what is running under the hood.
Considering that, I would like to point out the main disadvantage - you need to update manually all the build dependencies, or maintain outdated configs.
Create React App is not to be used for every app, it depends on the use case. Gatsby and Next.Js might also be great candidates for this 0 build tooling approach.
I wrote about it here:
When to pick Gatsby, Next.Js or Create React App
Alexandru-Dan Pop ・ Jul 13 '20 ・ 4 min read
Great article.
Thanks!
I couldn't agree more! CRA is waaaaaay too bloated in my opinion. It took me a while to learn Webpack when I wanted to do some customization, and I think relying on CRA was part of it.
I'd say it's ok and probably a must-do at some point of every React and FE developer journey; to do some experimenting with the tooling we use every day, which we don't even care about it (webpack, babel, eslint, etc).
But that's the point to use said technologies, they were created for us to configure once and forget about them, so we can focus on our business logic intead.
IMO it isn't wise to encourage people to do this for every project. Yes, do it once and for learning purposes, but this is clearly a foot gun.
The experience will teach you to not do this. Once the projects grows, your dependencies get obsolete and because your build system is totally custom, you're many steps behind, not only to upgrade something like your React version (which might be straightforward), but also to gain advantages of many cool improvements the people behind CRA came up with, and you're losing that train.
"Let's measure how many lines of code we didn't have to write instead"
Well said - I think its very important to know the tools we use under the hood on a daily basis. However projects like CRA were made and maintained by the some of the greatest minds in web development - might be best to stick with it unless it's for educational purposes.
I see your point. But hear me out, my project grows and I create my own design component library and I want this library to be bundled separately from other vendor bundles. If I have my own build configs I can easily add
cacheGroupsin webpack to achieve this but it's difficult to do so in CRA(without ejecting). It ultimately boils down to your project requirements🙂To be honest, that's when you should start considering how you structure your project. You are probably adding some unnecessary complexity into its build system.
I'd use a separate project for my UI library, heck, I wouldn't even use webpack on it, I prefer rollup for libraries, but that's totally my opinion.
Yes - IMO. Look at this way. These are the advs of CRA
react-scripts- Easy to upgradeBecause of the first two points, we are sacrificing two things
So we cannot incorporate the first two advs, so "The only advantage that we can take from CRA is Get started with a single command".
In the rest of the article, I wrote a step by step guide to add your own reactjs setup with webpack and babel. And finally implemented the "Single command" to bootstrap react app advantage using JS executable file and bin property.
If there is any other advantage of CRA, please share. Let's discuss how to implement them.
The problem with react-scripts is that react-scripts uses an older, out of date webpack. If you want to use the newest version, the app would fail to render using react-scripts to run the project. React-scripts forces you to use 4.44.2, whereas the newest version of Webpack is 5.38.1. To me, that's enough of a jump to be considered a huge negative if you want to use webpack.
As always, you can install react-scripts on your own. The advantage I see to using this custom setup is that one, you don't have unused dependencies bloating your project. Using CRA, my boilerplate redux-typescript-mocha setup was 1.17 GB in size. without CRA's bloating, that size shrunk to 839. That's 331 MB less in size. react-scripts alone adds an additional 120~ mb.
So, I can get behind setting up my own projects instead of using CRA. if you need a template, you could always use CRA and then delete what you don't need afterwords.
Awesome, Brian. I would love to checkout your redux-typescript-mocha setup. Can you share the link to the repo, if it's open source?
Sure, github.com/c0dezer019/Stormcaster/... It's in the client folder.
Cool
Some months ago I made a project to do, in some way, the same you do here. With some more experience in the field and more knowledge I can assure you that I learnt a lot, but it was a PAIN.
If you want to do it to understand what is behind webpack, react and babel its a good exercise. Elsewhere, you are just wasting time. IMO.
Trying to follow through this post, however I encountered
Error: Cannot find module 'webpack-cli/bin/config-yargs'error while running thenpm start.Changing from
"start": "webpack-dev-server --mode=development --open --hot"To
"start": "webpack serve --mode=development --open --hot"resolved the issue.
Seems like the interface in not compatible with webpack5 now. webpack-dev-server
Yeah, i've also encountered that issue (and solved it the same way).
Which btw? Is exactly the argument some people made in the comments above in regards for high maintainance, and error prone code when NOT using CRA, and going for pure Webpack solution.
Yeah, but migration to newer versions won't be that difficult if you have the habit of checking for any potential refactoring in your project regularly.
Thanks. Will update it.
Use Vite over CRA
medium.com/codex/you-should-choose...
CRA is devil! This is good article and I totally agree.
Here is my thoughts:
CRA is toy for dummies, it creates noise in frontend community and some folks think it's a best practies and afraid to admit that they spend days hacking around this crap...
Yes, you can learn how to cook it, and there are many people who will argue with me, but I will never admit that CRA is efficient and reliable way to run project... eventual you do "eject" and live with tones of crap that will be deprecated very soon...
I recommend you to use CRA for learning, but not for real project or you end up fighting against its limitations and "best practies"
The only advantage is that it comes with hot reload feature, but you can do it yourself anyway or just grab it from github
I'm not crazy (CRAzy?) about CRA either, but I do use it and have found workarounds to two of the points you mention:
there are several npm packages that allow more customization of the CRA tsconfig.json, so you can do things like link external project folders
depending on your deployment platform, you can make a monorepo work. I have a server node instance that runs separate of CRA, and the CRA instance proxies unhandled requests back to it (CRA has built-in support for this). At build+deployment time, the CRA project builds to a dist folder that the main project assumes responsibility to serve, so the whole thing can run as a single server instance
Neither of these steps are very confidence inspiring, but then neither is maintaining your own custom build stack. I've done that for personal projects in the past, and it runs into a problem where you have to futz with it just infrequently enough that you forget what you did last time, and have to relearn it all over each time. Between that and the CRA boilerplate, I'm not sure what's better.