This is an opinionated style guide for developing applications in ES6+ with React and/or Redux.
These styles are based on current best practices in the React and Redux communities, as well as real life experience with these tools in the field. It puts great focus on writing readable and maintainable code and using new JavaScript features in a responsible manner.
Much of this style guide is based on the Redux documentation and ideas from Dan Abramov in particular.
This guide is opinionated. You'll probably disagree on some points. Feel free to [open an issue] or file a pull request if you think something should be rephrased or changed. Please include a detailed reasoning about why your way is better, preferably with links to other sources. I change my mind often (for the better), so please convince me.
Developers who have recently discovered the power of ES6 often feel compelled to use them on every occasion. This is very good because it keeps us upstream with the community. But at the same we should be very cautious when using new features and syntax because it can quickly lead to unreadable code.
With great power comes great responsibility.
vardeprecated. There is no use case for it anymore.- Use
constby default. Immutable bindings are a good starting point. Code is easier to read and comprehend because you can be sure the variable is never re-assigned. - When your logic demands re-assignment use
let.
// deprecated:
var foo = 'foo';
// avoid:
let bar = 'bar';
// avoid:
const foo = 'foo',
bar = 'bar',
baz = 'baz';
// recommended:
const foo = 'foo';
const bar = 'bar';
const baz = 'baz';- Declare variables where you use them.
- Declare each variable as a full statement. It avoids the dangling comma discussion.
- Separate logical chunks of code with an empty line.
const foo = 'foo';
const bar = 'bar';
// use foo and bar
const baz = 'baz';
// use baz
const qux = 'qux';
// use quxCommon practice with var was to put all declarations at the top of their scope, even if they would only be used much
further down. The reason for this is that var declarations are automatically hoisted to the top of their scope, so
putting them there yourself makes the code more predictable(keep in mind var is deprecated in ES6).
However, let and const are not hoisted. Therefore there is no reason to declare them top of the file/scope. it's better to declare let/const just-in-time, right before the spot
where you are going to use them. This makes the code easier to follow. You won't have to go looking for usages before they are defined, because they can't be used there.
It also makes it easier to extract a bunch of statements into a function, which is a great way to improve readability even more.
file-names-like-this
ClassNamesLikeThis
methodNamesLikeThis
variableNamesLikeThis
parameterNamesLikeThis
propertyNamesLikeThis
SYMBOLIC_CONSTANTS_LIKE_THISUse ES6 imports import foo from 'foo'and use CommonJS imports if the ES6 import isn't supported by the library / use-cases like dynamically fetching image assests inside the codeconst assetUrl = isVerified ? require("../../verified.png") : require("../../non-verified.png").
Grouping There should be 3 clusters of imports:
third-party (aka vendor) libraries.React/Redux is "Third party" because we didn't write it.first-party libraries.If your team owns a separate repo(common-components) for sharing common component with other teams inside your organization then it becomes "First party", because it's not living inside our sorce code, but managed by us.local imports.all files inside our source code repo
import React, {Component} from "react";
import {connect} from redux;
import {Button, CheckBox} from "common-components";
import Util from "../common/util.js";
import mainThing, {
firstOtherThing,
secondOtherThing,
thirdOtherThing,
} from '../somewhere';Don't insert extra spaces between parens, brackets, or braces.
Yes:
// Literals:
const fancyPants = pants.map((pant) => ({...pant, isFancy: true}));
const toCartesian = (r, theta) => [r * cos(theta), r * sin(theta)];
// Destructuring:
const {StyleSheet, css} = valueObject;
const [x, y] = coordinates;
// Template strings:
const mission = `A ${price}, ${quality} education for ${clientele}.`;
// Parens:
if ((a === b) || (b === c)) {...}No:
// Literals:
const fancyPants = pants.map((pant) => ({ ...pant, isFancy: true }));
const toCartesian = (r, theta) => [ r * cos(theta), r * sin(theta) ];
// Destructuring:
const { StyleSheet, css } = require('aphrodite');
const [ x, y ] = coordinates;
// Template strings:
const mission = `A ${ price }, ${ quality } education for ${ clientele }.`;
// Parens:
if ( ( a === b ) || ( b === c ) ) {...}Lines should not exceed 79 characters. (This is called the "80 character rule," leaving 1 character for the newline.) This is consistent with our other language(python,...) style guide.
Inline style comments should be of the // variety, not the /* */ variety.
All files and classes should have JSDoc comments. JSDoc can be parsed by a number of open source tools, and must be well-formed. Syntax:
/**
* A JSDoc comment should begin with a slash and 2 asterisks.
*/Top-level (top-of-file) comments are designed to orient readers unfamiliar with the code to what is in this file and any other disclaimers clients of the code should be given. It should provide a description of the file's contents and any dependencies or compatibility information. As an example:
/**
* Various utility functions to handle manage the commonly used functions across
* the application.
*
* These utilities were written to be a general purpose utility
* for the entire code base.
*/Class comments should be used for every class, and give a description along with appropriate type tags (see "Methods and properties" comments for more information on types on the constructor).
/**
* Class making something fun and easy.
*
* @param {string} arg1 An argument that makes this more interesting.
* @param {Array.<number>} arg2 List of numbers to be processed.
*/
function SomeFunClass(arg1, arg2) {
// ...
}All non-trivial methods and properties should also have JSDoc comments.
Type annotations are strongly encouraged; if there is even a slight chance that the type will be ambiguous to future readers, put in a type annotation.
Type annotations are based on the ES4/JS2 type system, and are documented in the Google JavaScript style guide.
@param and @return type annotations that have comments that do not fit on one line wrap to the next line and indent 4 spaces.
/**
- A UI component allows users to select badges from their full list
- of earned badges, displaying them in a container.
- Expects a Badges.BadgeList as a model. */
Class FeedTable extends Component{
/**
* Whether or not this is currently in edit mode
*/
editable: false,
/**
* The feed list to show in table .
* @type {{id: number, name: string}}
*/
data: null,
/**
* @param {rowIdx: number}.
* @return
*/
edit(rowIdx){
…
},
...
};Don't over comment and avoid commenting by having self-explanatory names for the variables and methods. Avoid inline comments for methods, commenting only in method header is a good practice.
Prefer === (strict equality) to == due to the numerous oddities
related to JavaScript's type coercion.
The only valid use of == is for comparing against null and undefined
at the same time:
// Check null and undefined, but distinguish between other falsey values
if (someVariable == null) {Always use [] and {} style literals to initialize arrays and
objects, not the Array and Object constructors.
Array constructors are error-prone due to their arguments: new Array(3) yields [undefined, undefined, undefined], not [3].
To avoid these kinds of weird cases, always use the more readable array literal.
Object constructors don't have the same problems, but follow the same
rule for consistency with arrays. Plus, {} is more readable.
In some cases, we do not yet allow a new language feature, if it's expensive to polyfill. In others, we require using the newer language feature and avoiding the old:
| Construct | Use... | ...instead of |
|---|---|---|
| copy objects | use clone methods of ramda/underscore/... libraries |
JSON.parse(JSON.stringify(...))/Object.assign({}...) |
| backticks | `http://${host}/${path}` |
"http://" + host + "/" + path |
| destructuring | var {x, y} = a; |
var x = a.x; var y = a.y; |
| fat arrow | foo(() => {...}) |
foo(function() {...}.bind(this)) |
| let/const | let a = 1; const b = "4EVAH"; a++; |
var a = 1; var b = "4EVAH"; a++; |
| includes | array.includes(item) |
array.indexOf(item) !== -1 |
| map/forEach/reduce/filter | list.forEach() / list.map |
for(var i=0,i<lidt.length:i++){...} |
| for/of | for (const [key, value] of Object.entries(obj)) {...} |
_.each(obj, function(value, key) {...}) |
| spread | {...a, ...b, c: d} |
_.extend({}, a, b, {c: d}) |
| rest params | function(bar, ...args) {foo(...args);} |
function(bar) {var args = Array.prototype.slice.call(arguments, 1); foo.apply(null, args);} |
Arrow functions are easier to read (and with Babel, more efficient)
than calling bind manually.
The magic arguments variable has some odd quirks. It's simpler to
use rest params like (...args) => foo(args).
+ is not forbidden, but backticks are encouraged!
- Consider anonymous functions deprecated. We have arrow functions for that now. With the addition of the arrow function syntax, we have gained a very elegant and convenient way to define anonymous functions. Because they use 'lexical this', usage of
thisis more predictable. Therefore we should consider arrow functions instead of regular anonymous functions. - Arrow functions don’t replace regular functions. You can't use an arrow function when a dynamic context is required: defining methods, create objects with constructors, get the target from
thiswhen handling events for more detail on this.
// deprecated
function (string) {
return string.toLocaleUpperCase();
}
// recommended
function upper(string) {
return string.toLocaleUpperCase();
}
// use wisely
string => string.toLocaleUpperCase()
// if it fits your style
const upper = string => string.toLocaleUpperCase()- Don't use shared mutable state or access variables from outer scope.
- Don't cause any side effects by running a function.
- Don't mutate any passed arguments, return a new value instead.
// avoid side-effects
let x;
function inc() {
x++;
}
// recommended
function inc(x) {
return x + 1;
}
// avoid mutation
function inc(object) {
object.value++;
}
// recommended
function inc(object) {
return { ...object, value: object.value + 1 };
}A sure way to reduce code complexity and enhance testability and readability is to keep your functions pure. This is one of the primary aspects of functional programming. It involves writing functions in a purely input-output fashion. Pure functions only work with the data they are given through their arguments and return some new data, without causing side effects. That means they cannot access outside scope and they cannot mutate their arguments.
React components come in two flavors: presentation components and container components, also known as [dumb vs smart components]. We simply refer to them as 'components' and 'containers'.
React v0.14 introduced a new way to write components. In the past we would write every component as a class. However, since presentation components are very lightweight, they don't need all the power that a class provides (i.e. lifecycle methods and state).
The new way to write components uses a function instead of a class. Components defined that way are known as 'functional stateless components' because they cannot have any state and they are pure functions. This means they will receive their props as an argument and must return the rendered component. It basically boils down to having only a render method, which receives the props as an argument rather than accessing it as a class property. They also receive context as a second argument. Container components have application logic, but do not emit HTML themselves, will return HTML through dump components.
- Separate 'components'(dump) from 'containers'(smart).
- Prefer using a function for dump-components(shouldn't have state), a class for containers.
- Use PascalCase for component names.
- Put each component in its own directory and component file and style should be same name as directory name.
- Expose components as the default export.
- Expose secondary functions as named exports for unit testing.
- Never export anonymous (arrow) functions.
// components/TodoItem/index.js
export default function TodoItem(props, context) {
return <li onClick={() => context.onTodoClick()}>{props.label}</li>;
}
// containers/TodoList/index.js
export default class TodoList extends React.Component {
render() {
return <ul>{this.props.todos.map(todo => <TodoItem {...todo} />)}</ul>;
}
}-
Use static properties for
defaultProps. -
Use an instance property for
state. -
Use flow for
propsandstate. -
Autobind event handlers and callbacks.
Example:
import React, {Component} from 'react'; class Foo extends Component { static defaultProps = {} state = {} state: {} props: {} handleClick = (e) => { // handle the click } }
-
If
statedepends onprops, define it in the constructor.Example:
class Bar extends Component { constructor(props) { super(props); // must be called first this.state = { value: props.value, }; } }
-
Open elements on the same line.
The 80-character line limit is a bit tight, so we opt to conserve the extra 4. Yes:
return <div>
...
</div>;No:
return ( // "div" is not on the same line as "return"
<div>
...
</div>
);- Parentheses
-
Wrap JSX tags in parentheses when they span more than one line. eslint:
react/jsx-wrap-multilines// bad render() { return <MyComponent variant="long body" foo="bar"> <MyChild /> </MyComponent>; } // good render() { return ( <MyComponent variant="long body" foo="bar"> <MyChild /> </MyComponent> ); } // good, when single line render() { const body = <div>hello</div>; return <MyComponent>{body}</MyComponent>; }
8.Align and sort HTML properties.
Fit them all on the same line if you can. If you can't, put each property on a line of its own, indented four spaces, in sorted order. The closing angle brace should be on a line of its own, indented the same as the opening angle brace. This makes it easy to see the props at a glance.
Yes:
<div className="highlight" key="highlight-div">
<div
className="highlight"
key="highlight-div"
>
<Image
className="highlight"
key="highlight-div"
/>No:
<div className="highlight" // property not on its own line
key="highlight-div"
>
<div // closing brace not on its own line
className="highlight"
key="highlight-div">
<div // property is not sorted
key="highlight-div"
className="highlight"
>-
Use higher order components instead of mixins.
ES6 style classes do not support mixins. See the mixins considered harmful blog post for details of how to convert mixins to higher order components.
Ordering within a React component is strict. The following example illustrates the precise ordering of various component methods and properties:
class Foo extends Component {
// Static properties
static defaultProps = {}
// The `constructor` method
constructor() {
super();
}
// Instance properties
state = { hi: 5}
// Flow `state` annotation
// NOTE: `state` is special. All other Flow annotations live below.
state: { hi: number }
// React lifecycle hooks.
// They should follow their chronological ordering:
// 1. componentWillMount
// 2. componentDidMount
// 3. componentWillReceiveProps
// 4. shouldComponentUpdate
// 5. componentWillUpdate
// 6. componentDidUpdate
// 7. componentWillUnmount
componentDidMount() { ... }
// All other Flow annotations
props: { ... }
someRandomData: string,
// All other instance methods
handleClick = (e) => { ... }
// Finally, the render method
render() { ... }
}- Start your render method with a destructuring assignment on props and/or context.
- Use the spread operator to assign multiple attributes in one go.
const { label, completed } = this.props;
const { onTodoClick } = this.context;
return <TodoItem {...{ label, completed, onTodoClick }} />Prefer props to state. You almost always want to use props. By avoiding state when possible, you minimize redundancy, making it easier to reason about your application.
- Always define explicit defaultProps for all non-required props.
Why? propTypes are a form of documentation, and providing defaultProps means the reader of your code doesn’t have to assume as much. In addition, it can mean that your code can omit certain type checks.
// bad
function SFC({ foo, bar, children }) {
return <div>{foo}{bar}{children}</div>;
}
SFC.propTypes = {
foo: PropTypes.number.isRequired,
bar: PropTypes.string,
children: PropTypes.node,
};
// good
function SFC({ foo, bar, children }) {
return <div>{foo}{bar}{children}</div>;
}
SFC.propTypes = {
foo: PropTypes.number.isRequired,
bar: PropTypes.string,
children: PropTypes.node,
};
SFC.defaultProps = {
bar: '',
children: null,
};Notes for use: Filter out unnecessary props when possible. Also, use prop-types-exact to help prevent bugs.
//good
render() {
const { irrelevantProp, ...relevantProps } = this.props;
return <WrappedComponent {...relevantProps} />
}
//bad
render() {
const { irrelevantProp, ...relevantProps } = this.props;
return <WrappedComponent {...this.props} />
}Example:
<Component onClick={this.handleClick} onLaunchMissiles={this.handleLaunchMissiles} />This is consistent with React's event naming: onClick, onDrag,
onChange, etc.
Example:
<Component onLaunchMissiles={this.handleLaunchMissiles} />Always use ref callbacks. eslint: react/no-string-refs
```jsx
// bad
<Foo
ref="myRef"
/>
// good
<Foo
ref={(ref) => { this.myRef = ref; }}
/>
```
children is something of a special prop in React. Most often,
you'll want to pass a React element (or an array of React elements).
This can be annotated like so:
children: PropTypes.arrayOf(PropTypes.element) | PropTypes.elementNote that this is only the most common use-case for children. Children can also be other types (like a string, or a function).
Do not use data- attributes or classes. All information should be stored in JavaScript, either in the React component itself,or in a React store if using a framework such as Redux.
-
Avoid using an array index as
keyprop, prefer a unique ID. (why?)// bad {todos.map((todo, index) => <Todo {...todo} key={index} /> )} // good {todos.map(todo => ( <Todo {...todo} key={todo.id} /> ))}
Every .jsx file should export a single React class.
Note that the file can still define multiple classes, it just can't export more than one.
- Do not use
isMounted. eslint:react/no-is-mounted
Why?
isMountedis an anti-pattern, is not available when using ES6 classes, and is on its way to being officially deprecated.
Consider Type-checking with Flow over PropTypes
Flow is a type-checker that runs at compile-time to catch issues and prevent bugs. It should be used on all new components.
For more information on how we use Flow, check the Javascript style guide
Props can now be validated with Flow instead of React's PropTypes.Flow provides a much more expressive way to set types for props,with the additional benefits of being able to annotate state (and any additional methods or data).
Types can be defined on the class itself:
class Foo extends Component {
props: {
name: string,
uniqueId: number,
complexData: {
setOfThings: Array<string>,
},
}
}If you're writing a Stateless Functional Component, or if you use
lifecycle methods that reference props (eg. componentDidUpdate),
you may find it helpful to define a Props type, and reference it:
type Props = {
numbers: Array<number>,
};
class Foo extends Component {
props: Props
shouldComponentUpdate(nextProps: Props) {
...
}
}
const StatelessFoo = ({numbers}: Props) => {
...
};- Group related action creators in one file.
- Use redux-promise / redux-thunk for asynchronous actions, or redux-saga if you need more flexibility.
- Expose each action creator as a named export.
- Don't use a default export.
// actions/something.js
export function doSomething() {
return {
type: 'DO_SOMETHING',
payload: { foo: 'bar' },
meta: { baz: 'qux' }
};
}Action creators are functions which return an action object. An action object contains a type and optionally a payload and metadata. Action objects should follow the [Reux Standard Action] schema.
We could also use the createAction helper of redux-actions, but other than enforcing FSA it doesn't do much here in
terms of reducing boilerplate. In fact createAction is less readable and doesn't let us easily export named functions
(exporting anonymous or arrow functions should be avoided). Note that only type is mandatory.
'Ducks' is a proposal to bundle action creators with their reducers into a single file. Such a bundle is called a 'duck'. The goal is to move towards a modular application structure. So far the common practice is to group files by type rather than by feature. A similar transition has happened in the Angular world. This style guide still follows the group-by-type pattern, but may move towards using ducks in the future.
- Expose reducers as the default export.
- Expose handlers as named exports for unit testing.
- Use constants instead of inline strings for action types.
- Always define an
getInitialState()method for initial state.
// using redux-actions
// reducers/something.js
import { handleActions } from 'redux-actions';
import { Todo } from '../constants/actionTypes';
const getInitialState () => {
...
};
export default handleActions({
[Todo.add]: addTodo
}, state = getInitialState());
export function addTodo(state, action) {
return { ...state, todo: action.payload.todo };
}A reducer handles incoming actions. All reducers are triggered for all actions, so it's up to each reducer to decide whether or not to act on the incoming action. This is commonly done by switching on the action type. Reducers receive a current state and an action object and must return a new (updated or not) state. Their function signature is as follows:
(state, action) => stateReducers in Redux are pure functions, which means they cannot access outside information other than the arguments they are given. It's also not allowed to mutate outside state, be it by referencing outer scope or by mutating objects which they receive as arguments. In other words they must take the data they are given and return a new state without using mutation. This can often be achieved by using the array and object spread operators.
The simplest way to create a reducer is to use the handleActions method of redux-actions. This avoids having to write a switch statement and a default handler. Another benefit is that is enforces the use of FSA-compliant action objects. The final reducer should be exposed as the default export. Individual handler functions should be exposed as named exports in order to simplify unit testing.
- Always use a state selector function that returns the minimal state subset you need.
- Expose the connected component as default export.
- Expose the unconnected component as named export for unit testing.
export class TodoList extends React.Component { ... }
const stateSelector = ({ todos, filters }) => ({ todos, filters });
export default connect(stateSelector)(TodoList);The reselect library works well for more complex selectors and to achieve better performance.
- All methods should return a promise.
- Expose the entire service object as a default export.
- Expose Individual service methods as named exports for unit testing.
- Declare methods in alphabetical order.
- Declare helper functions underneath the methods, in chronological order.
- Never export anonymous (arrow) functions.
// services/SomethingService.js
export default {
getSomething
};
export function getSomething() {
return new Promise(resolve => resolve(helper()));
}
function helper() {
return 'let me help you with that';
}A service is where you commonly do AJAX requests to the back-end. All service methods should return a promise, even if the result is calculated synchronously. This ensures a consistent and predictable API across all services and allows changing to an asynchronous implementation later, without having to update service method calls all over the codebase.
Services should expose themselves via a default export to allow consumers to import the service as a whole. The reason for this is that multiple services may expose the same generic method names, which can lead to name collisions. Services should also expose each method individually using a named export in order to aid unit testing.
Service methods should be listed in alphabetical order. This keeps things tidy and avoids merge conflicts. Any helper functions should be defined underneath all exported methods and listed in the order in which they are used (i.e. chronological order, see 'Arrow functions'). Consider moving helper functions to utils and reusing them between services.
- Group related util functions under a common name.
- List util functions in alphabetical order.
- Expose each function as named export.
- Don't use a default export.
- Util functions must be pure.
- Util functions should be reusable, but have a single purpose.
// utils/serialization.js
export function upper(string) {
return string.toLocaleUpperCase();
}Utils is a collection of various supporting functions. It is a good idea to extract a function to a util when you are doing the same thing in multiple places throughout the codebase. If you do are not using the function in several files, it probably should not be a util function.
Exported functions should be defined in alphabetical order to make them easy to locate and avoid merge issues. Each function should be exposed using a named export. A utils file should not expose a default export, as it does not represent an entity as a whole but is merely a bag of related functions.
Utility functions may never keep internal state, nor can they use services. They should be pure functions. Each exported function should be suitable for multiple uses (i.e. generic enough to be reused), but they should have a single well defined purpose (i.e. do one thing, and do it well).