styleguide/tsguide.html

4064 lines
138 KiB
HTML
Raw Permalink Normal View History

2021-01-02 23:30:59 +08:00
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8">
<title>Google TypeScript Style Guide</title>
<link rel="stylesheet" href="javaguide.css">
<script src="include/styleguide.js"></script>
<link rel="shortcut icon" href="https://www.google.com/favicon.ico">
<script src="include/jsguide.js"></script>
</head>
<body onload="initStyleGuide();">
<div id="content">
<h1>Google TypeScript Style Guide</h1>
<section>
<p>This guide is based on the internal Google TypeScript style guide, but it has
been slightly adjusted to remove Google-internal sections. Google's internal
environment has different constraints on TypeScript than you might find outside
of Google. The advice here is specifically useful for people authoring code they
intend to import into Google, but otherwise may not apply in your external
environment.</p>
2021-01-02 23:30:59 +08:00
<p>There is no automatic deployment process for this version as it's pushed
on-demand by volunteers.</p>
</section>
<h2 id="introduction" class="numbered">Introduction</h2>
<h3 id="terminology-notes" class="numbered">Terminology notes</h3>
2021-01-02 23:30:59 +08:00
<p>This Style Guide uses <a href="https://tools.ietf.org/html/rfc2119">RFC 2119</a>
terminology when using the phrases <em>must</em>, <em>must not</em>, <em>should</em>, <em>should not</em>,
and <em>may</em>. The terms <em>prefer</em> and <em>avoid</em> correspond to <em>should</em> and <em>should
not</em>, respectively. Imperative and declarative statements are prescriptive and
correspond to <em>must</em>.</p>
2021-01-02 23:30:59 +08:00
<h3 id="guide-notes" class="numbered">Guide notes</h3>
2021-01-02 23:30:59 +08:00
<p>All examples given are <strong>non-normative</strong> and serve only to illustrate the
normative language of the style guide. That is, while the examples are in Google
Style, they may not illustrate the <em>only</em> stylish way to represent the code.
Optional formatting choices made in examples must not be enforced as rules.</p>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<h2 id="source-file-basics" class="numbered">Source file basics</h2>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<p><a id="file-encoding"></a></p>
<h3 id="file-encoding-utf-8" class="numbered">File encoding: UTF-8</h3>
<p>Source files are encoded in <strong>UTF-8</strong>.</p>
<p><a id="special-characters"></a></p>
<h4 id="whitespace-characters" class="numbered">Whitespace characters</h4>
<p>Aside from the line terminator sequence, the ASCII horizontal space character
(0x20) is the only whitespace character that appears anywhere in a source file.
This implies that all other whitespace characters in string literals are
escaped.</p>
<h4 id="special-escape-sequences" class="numbered">Special escape sequences</h4>
<p>For any character that has a special escape sequence (<code>\'</code>, <code>\"</code>, <code>\\</code>, <code>\b</code>,
<code>\f</code>, <code>\n</code>, <code>\r</code>, <code>\t</code>, <code>\v</code>), that sequence is used rather than the
corresponding numeric escape (e.g <code>\x0a</code>, <code>\u000a</code>, or <code>\u{a}</code>). Legacy octal
escapes are never used.</p>
<h4 id="non-ascii-characters" class="numbered">Non-ASCII characters</h4>
<p>For the remaining non-ASCII characters, use the actual Unicode character (e.g.
<code></code>). For non-printable characters, the equivalent hex or Unicode escapes (e.g.
<code>\u221e</code>) can be used along with an explanatory comment.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">// Perfectly clear, even without a comment.
const units = 'μs';
2021-01-02 23:30:59 +08:00
// Use escapes for non-printable characters.
const output = '\ufeff' + content; // byte order mark
2021-01-02 23:30:59 +08:00
</code></pre>
<pre><code class="language-ts bad">// Hard to read and prone to mistakes, even with the comment.
const units = '\u03bcs'; // Greek letter mu, 's'
2021-01-02 23:30:59 +08:00
// The reader has no idea what this is.
const output = '\ufeff' + content;
2021-01-02 23:30:59 +08:00
</code></pre>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<p><a id="modules"></a>
<a id="source-organization"></a></p>
2021-01-02 23:30:59 +08:00
<h2 id="source-file-structure" class="numbered">Source file structure</h2>
2021-01-02 23:30:59 +08:00
<p>Files consist of the following, <strong>in order</strong>:</p>
2021-01-02 23:30:59 +08:00
<ol>
<li>Copyright information, if present</li>
<li>JSDoc with <code>@fileoverview</code>, if present</li>
<li>Imports, if present</li>
<li>The files implementation</li>
</ol>
2021-01-02 23:30:59 +08:00
<p><strong>Exactly one blank line</strong> separates each section that is present.</p>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<h3 id="file-copyright" class="numbered">Copyright information</h3>
2021-01-02 23:30:59 +08:00
<p>If license or copyright information is necessary in a file, add it in a JSDoc at
the top of the file. </p>
2021-01-02 23:30:59 +08:00
<p><a id="file-fileoverview"></a>
<a id="jsdoc-top-file-level-comments"></a></p>
2021-01-02 23:30:59 +08:00
<h3 id="fileoverview" class="numbered"><code>@fileoverview</code> JSDoc</h3>
2021-01-02 23:30:59 +08:00
<p>A file may have a top-level <code>@fileoverview</code> JSDoc. If present, it may provide a
description of the file's content, its uses, or information about its
dependencies. Wrapped lines are not indented.</p>
2021-01-02 23:30:59 +08:00
<p>Example:</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">/**
* @fileoverview Description of file. Lorem ipsum dolor sit amet, consectetur
* adipiscing elit, sed do eiusmod tempor incididunt.
*/
</code></pre>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<h3 id="imports" class="numbered">Imports</h3>
2021-01-02 23:30:59 +08:00
<p>There are four variants of import statements in ES6 and TypeScript:</p>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<section>
2021-01-02 23:30:59 +08:00
<table>
<thead>
<tr>
<th>Import type</th>
<th>Example</th>
<th>Use for</th>
</tr>
</thead>
2021-01-02 23:30:59 +08:00
<tbody>
<tr>
<td>module[<sup>module_import]</sup>
</td>
<td><code>import * as foo from
'...';</code></td>
<td>TypeScript imports
</td>
</tr>
<tr>
<td>named[<sup>destructuring_import]</sup>
</td>
<td><code>import {SomeThing}
from '...';</code></td>
<td>TypeScript imports
</td>
</tr>
<tr>
<td>default
2021-01-02 23:30:59 +08:00
</td>
<td><code>import SomeThing
from '...';</code>
</td>
<td>Only for other
external code that
requires them</td>
</tr>
<tr>
<td>side-effect
2021-01-02 23:30:59 +08:00
</td>
<td><code>import '...';</code>
</td>
<td>Only to import
libraries for their
side-effects on load
(such as custom
elements)</td>
</tr>
</tbody>
</table>
<pre><code class="language-ts good">// Good: choose between two options as appropriate (see below).
import * as ng from '@angular/core';
import {Foo} from './foo';
// Only when needed: default imports.
import Button from 'Button';
// Sometimes needed to import libraries for their side effects:
import 'jasmine';
import '@polymer/paper-button';
</code></pre>
</section>
2021-01-02 23:30:59 +08:00
<h4 id="import-paths" class="numbered">Import paths</h4>
2021-01-02 23:30:59 +08:00
<p>TypeScript code <em>must</em> use paths to import other TypeScript code. Paths <em>may</em> be
relative, i.e. starting with <code>.</code> or <code>..</code>,
or rooted at the base directory, e.g.
<code>root/path/to/file</code>.</p>
2021-01-02 23:30:59 +08:00
<p>Code <em>should</em> use relative imports (<code>./foo</code>) rather than absolute imports
<code>path/to/foo</code> when referring to files within the same (logical) project as this
allows to move the project around without introducing changes in these imports.</p>
2021-01-02 23:30:59 +08:00
<p>Consider limiting the number of parent steps (<code>../../../</code>) as those can make
module and path structures hard to understand.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">import {Symbol1} from 'path/from/root';
import {Symbol2} from '../parent/file';
import {Symbol3} from './sibling';
</code></pre>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<p><a id="module-versus-destructuring-import"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="namespace-versus-named-imports" class="numbered">Namespace versus named imports</h4>
2021-01-02 23:30:59 +08:00
<p>Both namespace and named imports can be used.</p>
2021-01-02 23:30:59 +08:00
<p>Prefer named imports for symbols used frequently in a file or for symbols that
have clear names, for example Jasmine's <code>describe</code> and <code>it</code>. Named imports can
be aliased to clearer names as needed with <code>as</code>.</p>
2021-01-02 23:30:59 +08:00
<p>Prefer namespace imports when using many different symbols from large APIs. A
namespace import, despite using the <code>*</code> character, is not comparable to a
<q>wildcard</q> import as seen in other languages. Instead, namespace imports give a
name to all the exports of a module, and each exported symbol from the module
becomes a property on the module name. Namespace imports can aid readability for
exported symbols that have common names like <code>Model</code> or <code>Controller</code> without the
need to declare aliases.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">// Bad: overlong import statement of needlessly namespaced names.
import {Item as TableviewItem, Header as TableviewHeader, Row as TableviewRow,
Model as TableviewModel, Renderer as TableviewRenderer} from './tableview';
2021-01-02 23:30:59 +08:00
let item: TableviewItem|undefined;
</code></pre>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">// Better: use the module for namespacing.
import * as tableview from './tableview';
2021-01-02 23:30:59 +08:00
let item: tableview.Item|undefined;
</code></pre>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">import * as testing from './testing';
2021-01-02 23:30:59 +08:00
// Bad: The module name does not improve readability.
testing.describe('foo', () =&gt; {
testing.it('bar', () =&gt; {
testing.expect(null).toBeNull();
testing.expect(undefined).toBeUndefined();
});
});
</code></pre>
<pre><code class="language-ts good">// Better: give local names for these common functions.
import {describe, it, expect} from './testing';
describe('foo', () =&gt; {
it('bar', () =&gt; {
expect(null).toBeNull();
expect(undefined).toBeUndefined();
});
});
</code></pre>
<h5 id="jspb-import-by-path" class="numbered">Special case: Apps JSPB protos</h5>
<p>Apps JSPB protos must use named imports, even when it leads to long import
lines.</p>
<p>This rule exists to aid in build performance and dead code elimination since
often <code>.proto</code> files contain many <code>message</code>s that are not all needed together.
By leveraging destructured imports the build system can create finer grained
dependencies on Apps JSPB messages while preserving the ergonomics of path based
imports.</p>
<pre><code class="language-ts good">// Good: import the exact set of symbols you need from the proto file.
import {Foo, Bar} from './foo.proto';
function copyFooBar(foo: Foo, bar: Bar) {...}
</code></pre>
<h4 id="renaming-imports" class="numbered">Renaming imports</h4>
<p>Code <em>should</em> fix name collisions by using a namespace import or renaming the
exports themselves. Code <em>may</em> rename imports (<code>import {SomeThing as
SomeOtherThing}</code>) if needed.</p>
<p>Three examples where renaming can be helpful:</p>
<ol>
<li>If it's necessary to avoid collisions with other imported symbols.</li>
<li>If the imported symbol name is generated.</li>
<li>If importing symbols whose names are unclear by themselves, renaming can
improve code clarity. For example, when using RxJS the <code>from</code> function might
be more readable when renamed to <code>observableFrom</code>.</li>
</ol>
<h3 id="exports" class="numbered">Exports</h3>
<p>Use named exports in all code:</p>
<pre><code class="language-ts good">// Use named exports:
export class Foo { ... }
</code></pre>
<p>Do not use default exports. This ensures that all imports follow a uniform
pattern.</p>
<pre><code class="language-ts bad">// Do not use default exports:
export default class Foo { ... } // BAD!
</code></pre>
<section class="zippy">
<p>Why?</p>
<p>Default exports provide no canonical name, which makes central maintenance
difficult with relatively little benefit to code owners, including potentially
decreased readability:</p>
<pre><code class="language-ts bad">import Foo from './bar'; // Legal.
import Bar from './bar'; // Also legal.
</code></pre>
<p>Named exports have the benefit of erroring when import statements try to import
something that hasn't been declared. In <code>foo.ts</code>:</p>
<pre><code class="language-ts bad">const foo = 'blah';
export default foo;
</code></pre>
<p>And in <code>bar.ts</code>:</p>
<pre><code class="language-ts bad">import {fizz} from './foo';
</code></pre>
<p>Results in <code>error TS2614: Module '"./foo"' has no exported member 'fizz'.</code> While
<code>bar.ts</code>:</p>
<pre><code class="language-ts bad">import fizz from './foo';
</code></pre>
<p>Results in <code>fizz === foo</code>, which is probably unexpected and difficult to debug.</p>
<p>Additionally, default exports encourage people to put everything into one big
object to namespace it all together:</p>
<pre><code class="language-ts bad">export default class Foo {
static SOME_CONSTANT = ...
static someHelpfulFunction() { ... }
...
}
</code></pre>
<p>With the above pattern, we have file scope, which can be used as a namespace. We
also have a perhaps needless second scope (the class <code>Foo</code>) that can be
ambiguously used as both a type and a value in other files.</p>
<p>Instead, prefer use of file scope for namespacing, as well as named exports:</p>
<pre><code class="language-ts good">export const SOME_CONSTANT = ...
export function someHelpfulFunction()
export class Foo {
// only class stuff here
}
</code></pre>
</section>
<h4 id="export-visibility" class="numbered">Export visibility</h4>
<p>TypeScript does not support restricting the visibility for exported symbols.
Only export symbols that are used outside of the module. Generally minimize the
exported API surface of modules.</p>
<h4 id="mutable-exports" class="numbered">Mutable exports</h4>
<p>Regardless of technical support, mutable exports can create hard to understand
and debug code, in particular with re-exports across multiple modules. One way
to paraphrase this style point is that <code>export let</code> is not allowed.</p>
<section>
<pre><code class="language-ts bad">export let foo = 3;
// In pure ES6, foo is mutable and importers will observe the value change after a second.
// In TS, if foo is re-exported by a second file, importers will not see the value change.
window.setTimeout(() =&gt; {
foo = 4;
}, 1000 /* ms */);
</code></pre>
</section>
<p>If one needs to support externally accessible and mutable bindings, they
<em>should</em> instead use explicit getter functions.</p>
<pre><code class="language-ts good">let foo = 3;
window.setTimeout(() =&gt; {
foo = 4;
}, 1000 /* ms */);
// Use an explicit getter to access the mutable export.
export function getFoo() { return foo; };
</code></pre>
<p>For the common pattern of conditionally exporting either of two values, first do
the conditional check, then the export. Make sure that all exports are final
after the module's body has executed.</p>
<pre><code class="language-ts good">function pickApi() {
if (useOtherApi()) return OtherApi;
return RegularApi;
}
export const SomeApi = pickApi();
</code></pre>
<p><a id="static-containers"></a></p>
<h4 id="container-classes" class="numbered">Container classes</h4>
<p>Do not create container classes with static methods or properties for the sake
of namespacing.</p>
<pre><code class="language-ts bad">export class Container {
static FOO = 1;
static bar() { return 1; }
}
</code></pre>
<p>Instead, export individual constants and functions:</p>
<pre><code class="language-ts good">export const FOO = 1;
export function bar() { return 1; }
</code></pre>
<h3 id="import-export-type" class="numbered">Import and export type</h3>
<h4 id="import-type" class="numbered">Import type</h4>
<p>You may use <code>import type {...}</code> when you use the imported symbol only as a type.
Use regular imports for values:</p>
<pre><code class="language-ts good">import type {Foo} from './foo';
import {Bar} from './foo';
import {type Foo, Bar} from './foo';
</code></pre>
<section class="zippy">
<p>Why?</p>
<p>The TypeScript compiler automatically handles the distinction and does not
insert runtime loads for type references. So why annotate type imports?</p>
<p>The TypeScript compiler can run in 2 modes:</p>
<ul>
<li>In development mode, we typically want quick iteration loops. The compiler
transpiles to JavaScript without full type information. This is much faster,
but requires <code>import type</code> in certain cases.</li>
<li>In production mode, we want correctness. The compiler type checks everything
and ensures <code>import type</code> is used correctly.</li>
</ul>
<p>Note: If you need to force a runtime load for side effects, use <code>import '...';</code>.
See </p>
</section>
<h4 id="export-type" class="numbered">Export type</h4>
<p>Use <code>export type</code> when re-exporting a type, e.g.:</p>
<pre><code class="language-ts good">export type {AnInterface} from './foo';
</code></pre>
<section class="zippy">
<p>Why?</p>
<p><code>export type</code> is useful to allow type re-exports in file-by-file transpilation.
See
<a href="https://www.typescriptlang.org/tsconfig#exports-of-non-value-identifiers"><code>isolatedModules</code> docs</a>.</p>
<p><code>export type</code> might also seem useful to avoid ever exporting a value symbol for
an API. However it does not give guarantees, either: downstream code might still
import an API through a different path. A better way to split &amp; guarantee type
vs value usages of an API is to actually split the symbols into e.g.
<code>UserService</code> and <code>AjaxUserService</code>. This is less error prone and also better
communicates intent.</p>
</section>
<p><a id="namespaces-vs-modules"></a></p>
<h4 id="use-modules-not-namespaces" class="numbered">Use modules not namespaces</h4>
<p>TypeScript supports two methods to organize code: <em>namespaces</em> and <em>modules</em>,
but namespaces are disallowed. That
is, your code <em>must</em> refer to code in other files using imports and exports of
the form <code>import {foo} from 'bar';</code></p>
<p>Your code <em>must not</em> use the <code>namespace Foo { ... }</code> construct. <code>namespace</code>s
<em>may</em> only be used when required to interface with external, third party code.
To semantically namespace your code, use separate files.</p>
<p>Code <em>must not</em> use <code>require</code> (as in <code>import x = require('...');</code>) for imports.
Use ES6 module syntax.</p>
<pre><code class="language-ts bad">// Bad: do not use namespaces:
namespace Rocket {
function launch() { ... }
}
// Bad: do not use &lt;reference&gt;
/// &lt;reference path="..."/&gt;
// Bad: do not use require()
import x = require('mydep');
</code></pre>
<blockquote>
<p>NB: TypeScript <code>namespace</code>s used to be called internal modules and used to use
the <code>module</code> keyword in the form <code>module Foo { ... }</code>. Don't use that either.
Always use ES6 imports.</p>
</blockquote>
<p><a id="language-rules"></a></p>
<h2 id="language-features" class="numbered">Language features</h2>
<p>This section delineates which features may or may not be used, and any
additional constraints on their use.</p>
<p>Language features which are not discussed in this style guide <em>may</em> be used with
no recommendations of their usage.</p>
<p><a id="features-local-variable-declarations"></a></p>
<h3 id="local-variable-declarations" class="numbered">Local variable declarations</h3>
<p><a id="variables"></a>
<a id="features-use-const-and-let"></a></p>
<h4 id="use-const-and-let" class="numbered">Use const and let</h4>
<p>Always use <code>const</code> or <code>let</code> to declare variables. Use <code>const</code> by default, unless
a variable needs to be reassigned. Never use <code>var</code>.</p>
<pre><code class="language-ts good">const foo = otherValue; // Use if "foo" never changes.
let bar = someValue; // Use if "bar" is ever assigned into later on.
</code></pre>
<p><code>const</code> and <code>let</code> are block scoped, like variables in most other languages.
<code>var</code> in JavaScript is function scoped, which can cause difficult to understand
bugs. Don't use it.</p>
<pre><code class="language-ts bad">var foo = someValue; // Don't use - var scoping is complex and causes bugs.
</code></pre>
<p>Variables <em>must not</em> be used before their declaration.</p>
<p><a id="features-one-variable-per-declaration"></a></p>
<h4 id="one-variable-per-declaration" class="numbered">One variable per declaration</h4>
<p>Every local variable declaration declares only one variable: declarations such
as <code class="badcode">let a = 1, b = 2;</code> are not used.</p>
<p><a id="features-array-literals"></a></p>
<h3 id="array-literals" class="numbered">Array literals</h3>
<p><a id="features-arrays-ctor"></a></p>
<h4 id="array-constructor" class="numbered">Do not use the <code>Array</code> constructor</h4>
<p><em>Do not</em> use the <code>Array()</code> constructor, with or without <code>new</code>. It has confusing
and contradictory usage:</p>
<pre><code class="language-ts bad">const a = new Array(2); // [undefined, undefined]
const b = new Array(2, 3); // [2, 3];
</code></pre>
<p>Instead, always use bracket notation to initialize arrays, or <code>from</code> to
initialize an <code>Array</code> with a certain size:</p>
<pre><code class="language-ts good">const a = [2];
const b = [2, 3];
// Equivalent to Array(2):
const c = [];
c.length = 2;
// [0, 0, 0, 0, 0]
Array.from&lt;number&gt;({length: 5}).fill(0);
</code></pre>
<p><a id="features-arrays-non-numeric-properties"></a></p>
<h4 id="do-not-define-properties-on-arrays" class="numbered">Do not define properties on arrays</h4>
<p>Do not define or use non-numeric properties on an array (other than <code>length</code>).
Use a <code>Map</code> (or <code>Object</code>) instead.</p>
<p><a id="features-arrays-spread-operator"></a></p>
<h4 id="array-spread-syntax" class="numbered">Using spread syntax</h4>
<p>Using spread syntax <code>[...foo];</code> is a convenient shorthand for shallow-copying or
concatenating iterables.</p>
<pre><code class="language-ts good">const foo = [
1,
];
const foo2 = [
...foo,
6,
7,
];
const foo3 = [
5,
...foo,
];
foo2[1] === 6;
foo3[1] === 1;
</code></pre>
<p>When using spread syntax, the value being spread <em>must</em> match what is being
created. When creating an array, only spread iterables. Primitives (including
<code>null</code> and <code>undefined</code>) <em>must not</em> be spread.</p>
<pre><code class="language-ts bad">const foo = [7];
const bar = [5, ...(shouldUseFoo &amp;&amp; foo)]; // might be undefined
// Creates {0: 'a', 1: 'b', 2: 'c'} but has no length
const fooStrings = ['a', 'b', 'c'];
const ids = {...fooStrings};
</code></pre>
<pre><code class="language-ts good">const foo = shouldUseFoo ? [7] : [];
const bar = [5, ...foo];
const fooStrings = ['a', 'b', 'c'];
const ids = [...fooStrings, 'd', 'e'];
</code></pre>
<p><a id="features-arrays-destructuring"></a></p>
<h4 id="array-destructuring" class="numbered">Array destructuring</h4>
<p>Array literals may be used on the left-hand side of an assignment to perform
destructuring (such as when unpacking multiple values from a single array or
iterable). A final <q>rest</q> element may be included (with no space between the
<code>...</code> and the variable name). Elements should be omitted if they are unused.</p>
<pre><code class="language-ts good">const [a, b, c, ...rest] = generateResults();
let [, b,, d] = someArray;
</code></pre>
<p>Destructuring may also be used for function parameters. Always specify <code>[]</code> as
the default value if a destructured array parameter is optional, and provide
default values on the left hand side:</p>
<pre><code class="language-ts good">function destructured([a = 4, b = 2] = []) { … }
</code></pre>
<p>Disallowed:</p>
<pre><code class="language-ts bad">function badDestructuring([a, b] = [4, 2]) { … }
</code></pre>
<p>Tip: For (un)packing multiple values into a functions parameter or return,
prefer object destructuring to array destructuring when possible, as it allows
naming the individual elements and specifying a different type for each.</p>
<p><a id="features-object-literals"></a></p>
<h3 id="object-literals" class="numbered">Object literals</h3>
<p><a id="features-objects-ctor"></a></p>
<h4 id="object-constructor" class="numbered">Do not use the <code>Object</code> constructor</h4>
<p>The <code>Object</code> constructor is disallowed. Use an object literal (<code>{}</code> or <code>{a: 0,
b: 1, c: 2}</code>) instead.</p>
<h4 id="iterating-objects" class="numbered">Iterating objects</h4>
<p>Iterating objects with <code>for (... in ...)</code> is error prone. It will include
enumerable properties from the prototype chain.</p>
<p>Do not use unfiltered <code>for (... in ...)</code> statements:</p>
<pre><code class="language-ts bad">for (const x in someObj) {
// x could come from some parent prototype!
}
</code></pre>
<p>Either filter values explicitly with an <code>if</code> statement, or use <code>for (... of
Object.keys(...))</code>.</p>
<pre><code class="language-ts good">for (const x in someObj) {
if (!someObj.hasOwnProperty(x)) continue;
// now x was definitely defined on someObj
}
for (const x of Object.keys(someObj)) { // note: for _of_!
// now x was definitely defined on someObj
}
for (const [key, value] of Object.entries(someObj)) { // note: for _of_!
// now key was definitely defined on someObj
}
</code></pre>
<p><a id="using-the-spread-operator"></a></p>
<h4 id="object-spread-syntax" class="numbered">Using spread syntax</h4>
<p>Using spread syntax <code>{...bar}</code> is a convenient shorthand for creating a shallow
copy of an object. When using spread syntax in object initialization, later
values replace earlier values at the same key.</p>
<pre><code class="language-ts good">const foo = {
num: 1,
};
const foo2 = {
...foo,
num: 5,
};
const foo3 = {
num: 5,
...foo,
}
foo2.num === 5;
foo3.num === 1;
</code></pre>
<p>When using spread syntax, the value being spread <em>must</em> match what is being
created. That is, when creating an object, only objects may be spread; arrays
and primitives (including <code>null</code> and <code>undefined</code>) <em>must not</em> be spread. Avoid
spreading objects that have prototypes other than the Object prototype (e.g.
class definitions, class instances, functions) as the behavior is unintuitive
(only enumerable non-prototype properties are shallow-copied).</p>
<pre><code class="language-ts bad">const foo = {num: 7};
const bar = {num: 5, ...(shouldUseFoo &amp;&amp; foo)}; // might be undefined
// Creates {0: 'a', 1: 'b', 2: 'c'} but has no length
const fooStrings = ['a', 'b', 'c'];
const ids = {...fooStrings};
</code></pre>
<pre><code class="language-ts good">const foo = shouldUseFoo ? {num: 7} : {};
const bar = {num: 5, ...foo};
</code></pre>
<p><a id="features-objects-computed-property-names"></a></p>
<h4 id="computed-property-names" class="numbered">Computed property names</h4>
<p>Computed property names (e.g. <code>{['key' + foo()]: 42}</code>) are allowed, and are
considered dict-style (quoted) keys (i.e., must not be mixed with non-quoted
keys) unless the computed property is a
<a href="https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Symbol">symbol</a>
(e.g. <code>[Symbol.iterator]</code>).</p>
<p><a id="features-objects-destructuring"></a></p>
<h4 id="object-destructuring" class="numbered">Object destructuring</h4>
<p>Object destructuring patterns may be used on the left-hand side of an assignment
to perform destructuring and unpack multiple values from a single object.</p>
<p>Destructured objects may also be used as function parameters, but should be kept
as simple as possible: a single level of unquoted shorthand properties. Deeper
levels of nesting and computed properties may not be used in parameter
destructuring. Specify any default values in the left-hand-side of the
destructured parameter (<code>{str = 'some default'} = {}</code>, rather than
<code class="badcode">{str} = {str: 'some default'}</code>), and if a
destructured object is itself optional, it must default to <code>{}</code>.</p>
<p>Example:</p>
<pre><code class="language-ts good">interface Options {
/** The number of times to do something. */
num?: number;
/** A string to do stuff to. */
str?: string;
}
function destructured({num, str = 'default'}: Options = {}) {}
</code></pre>
<p>Disallowed:</p>
<pre><code class="language-ts bad">function nestedTooDeeply({x: {num, str}}: {x: Options}) {}
function nontrivialDefault({num, str}: Options = {num: 42, str: 'default'}) {}
</code></pre>
<p><a id="formatting-class-literals"></a>
<a id="features-classes"></a></p>
<h3 id="classes" class="numbered">Classes</h3>
<h4 id="class-declarations" class="numbered">Class declarations</h4>
<p>Class declarations <em>must not</em> be terminated with semicolons:</p>
<pre><code class="language-ts good">class Foo {
}
</code></pre>
<pre><code class="language-ts bad">class Foo {
}; // Unnecessary semicolon
</code></pre>
<p>In contrast, statements that contain class expressions <em>must</em> be terminated with
a semicolon:</p>
<pre><code class="language-ts good">export const Baz = class extends Bar {
method(): number {
return this.x;
}
}; // Semicolon here as this is a statement, not a declaration
</code></pre>
<pre><code class="language-ts bad">exports const Baz = class extends Bar {
method(): number {
return this.x;
}
}
</code></pre>
<p>It is neither encouraged nor discouraged to have blank lines separating class
declaration braces from other class content:</p>
<pre><code class="language-ts good">// No spaces around braces - fine.
class Baz {
method(): number {
return this.x;
}
}
// A single space around both braces - also fine.
class Foo {
method(): number {
return this.x;
}
}
</code></pre>
<h4 id="class-method-declarations" class="numbered">Class method declarations</h4>
<p>Class method declarations <em>must not</em> use a semicolon to separate individual
method declarations:</p>
<pre><code class="language-ts good">class Foo {
doThing() {
console.log("A");
}
}
</code></pre>
<pre><code class="language-ts bad">class Foo {
doThing() {
console.log("A");
}; // &lt;-- unnecessary
}
</code></pre>
<p>Method declarations should be separated from surrounding code by a single blank
line:</p>
<pre><code class="language-ts good">class Foo {
doThing() {
console.log("A");
}
getOtherThing(): number {
return 4;
}
}
</code></pre>
<pre><code class="language-ts bad">class Foo {
doThing() {
console.log("A");
}
getOtherThing(): number {
return 4;
}
}
</code></pre>
<p><a id="features-classes-overriding-tostring"></a></p>
<h5 id="overriding-tostring" class="numbered">Overriding toString</h5>
<p>The <code>toString</code> method may be overridden, but must always succeed and never have
visible side effects.</p>
<p>Tip: Beware, in particular, of calling other methods from toString, since
exceptional conditions could lead to infinite loops.</p>
<p><a id="features-classes-static-methods"></a></p>
<h4 id="static-methods" class="numbered">Static methods</h4>
<h5 id="avoid-private-static-methods" class="numbered">Avoid private static methods</h5>
<p>Where it does not interfere with readability, prefer module-local functions over
private static methods.</p>
<h5 id="avoid-static-method-dynamic-dispatch" class="numbered">Do not rely on dynamic dispatch</h5>
<p>Code <em>should not</em> rely on dynamic dispatch of static
methods. Static methods <em>should</em> only be called on the base class
itself (which defines it directly). Static methods <em>should not</em> be called on
variables containing a dynamic instance that may be either the constructor or a
subclass constructor (and <em>must</em> be defined with <code>@nocollapse</code> if this is done),
and <em>must not</em> be called directly on a subclass that doesnt define the method
itself.</p>
<p>Disallowed:</p>
<pre><code class="language-ts bad">// Context for the examples below (this class is okay by itself)
class Base {
/** @nocollapse */ static foo() {}
}
class Sub extends Base {}
// Discouraged: don't call static methods dynamically
function callFoo(cls: typeof Base) {
cls.foo();
}
// Disallowed: don't call static methods on subclasses that don't define it themselves
Sub.foo();
2021-01-02 23:30:59 +08:00
// Disallowed: don't access this in static methods.
class MyClass {
static foo() {
return this.staticField;
}
2021-01-02 23:30:59 +08:00
}
MyClass.staticField = 1;
</code></pre>
2021-01-02 23:30:59 +08:00
<h5 id="static-this" class="numbered">Avoid static <code>this</code> references</h5>
2021-01-02 23:30:59 +08:00
<p>Code <em>must not</em> use <code>this</code> in a static context.</p>
2021-01-02 23:30:59 +08:00
<p>JavaScript allows accessing static fields through <code>this</code>. Different from other
languages, static fields are also inherited.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="bad">class ShoeStore {
static storage: Storage = ...;
static isAvailable(s: Shoe) {
// Bad: do not use `this` in a static method.
return this.storage.has(s.id);
}
}
class EmptyShoeStore extends ShoeStore {
static storage: Storage = EMPTY_STORE; // overrides storage from ShoeStore
2021-01-02 23:30:59 +08:00
}
</code></pre>
<section class="zippy">
2021-01-02 23:30:59 +08:00
<p>Why?</p>
<p>This code is generally surprising: authors might not expect that static fields
can be accessed through the this pointer, and might be surprised to find that
they can be overridden - this feature is not commonly used.</p>
<p>This code also encourages an anti-pattern of having substantial static state,
which causes problems with testability.</p>
</section>
<p><a id="disallowed-features-omitting-parents-with-new"></a>
<a id="features-classes-constructors"></a></p>
<h4 id="constructors" class="numbered">Constructors</h4>
<p>Constructor calls <em>must</em> use parentheses, even when no arguments are passed:</p>
<pre><code class="language-ts bad">const x = new Foo;
</code></pre>
<pre><code class="language-ts good">const x = new Foo();
</code></pre>
<p>Omitting parentheses can lead to subtle mistakes. These two lines are not
equivalent:</p>
<pre><code class="language-js good">new Foo().Bar();
new Foo.Bar();
</code></pre>
<p>It is unnecessary to provide an empty constructor or one that simply delegates
into its parent class because ES2015 provides a default class constructor if one
is not specified. However constructors with parameter properties, visibility
modifiers or parameter decorators <em>should not</em> be omitted even if the body of
the constructor is empty.</p>
<pre><code class="language-ts bad">class UnnecessaryConstructor {
constructor() {}
}
</code></pre>
<pre><code class="language-ts bad">class UnnecessaryConstructorOverride extends Base {
constructor(value: number) {
super(value);
}
}
</code></pre>
<pre><code class="language-ts good">class DefaultConstructor {
}
class ParameterProperties {
constructor(private myService) {}
}
class ParameterDecorators {
constructor(@SideEffectDecorator myService) {}
}
class NoInstantiation {
private constructor() {}
}
</code></pre>
<p>The constructor should be separated from surrounding code both above and below
by a single blank line:</p>
<pre><code class="language-ts good">class Foo {
myField = 10;
constructor(private readonly ctorParam) {}
doThing() {
console.log(ctorParam.getThing() + myField);
}
}
</code></pre>
<pre><code class="language-ts bad">class Foo {
myField = 10;
constructor(private readonly ctorParam) {}
doThing() {
console.log(ctorParam.getThing() + myField);
}
}
</code></pre>
<h4 id="class-members" class="numbered">Class members</h4>
<h5 id="private-fields" class="numbered">No #private fields</h5>
<p>Do not use private fields (also known as private identifiers):</p>
<pre><code class="language-ts bad">class Clazz {
#ident = 1;
}
</code></pre>
<p>Instead, use TypeScript's visibility annotations:</p>
<pre><code class="language-ts good">class Clazz {
private ident = 1;
}
</code></pre>
<section class="zippy">
<p>Why?</p>
<p> Private identifiers cause substantial emit size and
performance regressions when down-leveled by TypeScript, and are unsupported
before ES2015. They can only be downleveled to ES2015, not lower. At the same
time, they do not offer substantial benefits when static type checking is used
to enforce visibility.</p>
</section>
<h5 id="use-readonly" class="numbered">Use readonly</h5>
<p>Mark properties that are never reassigned outside of the constructor with the
<code>readonly</code> modifier (these need not be deeply immutable).</p>
<h5 id="parameter-properties" class="numbered">Parameter properties</h5>
<p>Rather than plumbing an obvious initializer through to a class member, use a
TypeScript
<a href="https://www.typescriptlang.org/docs/handbook/2/classes.html#parameter-properties">parameter property</a>.</p>
<pre><code class="language-ts bad">class Foo {
private readonly barService: BarService;
constructor(barService: BarService) {
this.barService = barService;
}
}
</code></pre>
<pre><code class="language-ts good">class Foo {
constructor(private readonly barService: BarService) {}
}
</code></pre>
<p>If the parameter property needs documentation,
<a href="#parameter-property-comments">use an <code>@param</code> JSDoc tag</a>.</p>
<h5 id="field-initializers" class="numbered">Field initializers</h5>
<p>If a class member is not a parameter, initialize it where it's declared, which
sometimes lets you drop the constructor entirely.</p>
<pre><code class="language-ts bad">class Foo {
private readonly userList: string[];
constructor() {
this.userList = [];
}
}
</code></pre>
<pre><code class="language-ts good">class Foo {
private readonly userList: string[] = [];
}
</code></pre>
<p>Tip: Properties should never be added to or removed from an instance after the
constructor is finished, since it significantly hinders VMs ability to optimize
classes' <q>shape</q>. Optional fields that may be filled in later should be
explicitly initialized to <code>undefined</code> to prevent later shape changes.</p>
<h5 id="properties-used-outside-of-class-lexical-scope" class="numbered">Properties used outside of class lexical scope</h5>
<p>Properties used from outside the lexical scope of their containing class, such
as an Angular component's properties used from a template, <em>must not</em> use
<code>private</code> visibility, as they are used outside of the lexical scope of their
containing class.</p>
<p>Use either <code>protected</code> or <code>public</code> as appropriate to the property in question.
Angular and AngularJS template properties should use <code>protected</code>, but Polymer
should use <code>public</code>.</p>
<p>TypeScript code <em>must not</em> use <code>obj['foo']</code> to bypass the visibility of a
property.</p>
<section class="zippy">
<p>Why?</p>
<p>When a property is <code>private</code>, you are declaring to both automated systems and
humans that the property accesses are scoped to the methods of the declaring
class, and they will rely on that. For example, a check for unused code will
flag a private property that appears to be unused, even if some other file
manages to bypass the visibility restriction.</p>
<p>Though it might appear that <code>obj['foo']</code> can bypass visibility in the TypeScript
compiler, this pattern can be broken by rearranging the build rules, and also
violates <a href="#optimization-compatibility">optimization compatibility</a>.</p>
</section>
<p><a id="features-classes-getters-and-setters"></a></p>
<h5 id="classes-getters-and-setters" class="numbered">Getters and setters</h5>
<p>Getters and setters, also known as accessors, for class members <em>may</em> be used.
The getter method <em>must</em> be a
<a href="https://en.wikipedia.org/wiki/Pure_function">pure function</a> (i.e., result is
consistent and has no side effects: getters <em>must not</em> change observable state).
They are also useful as a means of restricting the visibility of internal or
verbose implementation details (shown below).</p>
<pre><code class="language-ts good">class Foo {
constructor(private readonly someService: SomeService) {}
get someMember(): string {
return this.someService.someVariable;
}
set someMember(newValue: string) {
this.someService.someVariable = newValue;
}
}
</code></pre>
<pre><code class="language-ts bad">class Foo {
nextId = 0;
get next() {
return this.nextId++; // Bad: getter changes observable state
2021-01-02 23:30:59 +08:00
}
}
</code></pre>
<p>If an accessor is used to hide a class property, the hidden property <em>may</em> be
prefixed or suffixed with any whole word, like <code>internal</code> or <code>wrapped</code>. When
using these private properties, access the value through the accessor whenever
possible. At least one accessor for a property <em>must</em> be non-trivial: do not
define <q>pass-through</q> accessors only for the purpose of hiding a property.
Instead, make the property public (or consider making it <code>readonly</code> rather than
just defining a getter with no setter).</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">class Foo {
private wrappedBar = '';
get bar() {
return this.wrappedBar || 'bar';
}
2021-01-02 23:30:59 +08:00
set bar(wrapped: string) {
this.wrappedBar = wrapped.trim();
}
}
2021-01-02 23:30:59 +08:00
</code></pre>
<pre><code class="language-ts bad">class Bar {
private barInternal = '';
// Neither of these accessors have logic, so just make bar public.
get bar() {
return this.barInternal;
}
set bar(value: string) {
this.barInternal = value;
2021-01-02 23:30:59 +08:00
}
}
</code></pre>
<p>Getters and setters <em>must not</em> be defined using <code>Object.defineProperty</code>, since
this interferes with property renaming.</p>
2021-01-02 23:30:59 +08:00
<p><a id="features-classes-computed-properties"></a></p>
2021-01-02 23:30:59 +08:00
<h5 id="class-computed-properties" class="numbered">Computed properties</h5>
2021-01-02 23:30:59 +08:00
<p>Computed properties may only be used in classes when the property is a symbol.
Dict-style properties (that is, quoted or computed non-symbol keys) are not
allowed (see
<a href="#features-objects-mixing-keys">rationale for not mixing key types</a>. A
<code>[Symbol.iterator]</code> method should be defined for any classes that are logically
iterable. Beyond this, <code>Symbol</code> should be used sparingly.</p>
2021-01-02 23:30:59 +08:00
<p>Tip: be careful of using any other built-in symbols (e.g.
<code>Symbol.isConcatSpreadable</code>) as they are not polyfilled by the compiler and will
therefore not work in older browsers.</p>
2021-01-02 23:30:59 +08:00
<h4 id="visibility" class="numbered">Visibility</h4>
2021-01-02 23:30:59 +08:00
<p>Restricting visibility of properties, methods, and entire types helps with
keeping code decoupled.</p>
<ul>
<li>Limit symbol visibility as much as possible.</li>
<li>Consider converting private methods to non-exported functions within the
same file but outside of any class, and moving private properties into a
separate, non-exported class.</li>
<li>TypeScript symbols are public by default. Never use the <code>public</code> modifier
except when declaring non-readonly public parameter properties (in
constructors).</li>
</ul>
<pre><code class="language-ts bad">class Foo {
public bar = new Bar(); // BAD: public modifier not needed
constructor(public readonly baz: Baz) {} // BAD: readonly implies it's a property which defaults to public
}
</code></pre>
<pre><code class="language-ts good">class Foo {
bar = new Bar(); // GOOD: public modifier not needed
constructor(public baz: Baz) {} // public modifier allowed
}
</code></pre>
<p>See also <a href="#export-visibility">export visibility</a>.</p>
2021-01-02 23:30:59 +08:00
<h4 id="disallowed-class-patterns" class="numbered">Disallowed class patterns</h4>
2021-01-02 23:30:59 +08:00
<p><a id="features-classes-prototypes"></a></p>
2021-01-02 23:30:59 +08:00
<h5 id="class-prototypes" class="numbered">Do not manipulate <code>prototype</code>s directly</h5>
<p>The <code>class</code> keyword allows clearer and more readable class definitions than
defining <code>prototype</code> properties. Ordinary implementation code has no business
manipulating these objects. Mixins and modifying the prototypes of builtin
objects are explicitly forbidden.</p>
<p><strong>Exception</strong>: Framework code (such as Polymer, or Angular) may need to use <code>prototype</code>s, and should not resort
to even-worse workarounds to avoid doing so.</p>
<p><a id="features-functions"></a></p>
<h3 id="functions" class="numbered">Functions</h3>
<h4 id="terminology" class="numbered">Terminology</h4>
<p>There are many different types of functions, with subtle distinctions between
them. This guide uses the following terminology, which aligns with
<a href="https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Functions">MDN</a>:</p>
<ul>
<li><q>function declaration</q>: a declaration (i.e. not an expression) using the
<code>function</code> keyword</li>
<li><q>function expression</q>: an expression, typically used in an assignment or
passed as a parameter, using the <code>function</code> keyword</li>
<li><q>arrow function</q>: an expression using the <code>=&gt;</code> syntax</li>
<li><q>block body</q>: right hand side of an arrow function with braces</li>
<li><q>concise body</q>: right hand side of an arrow function without braces</li>
</ul>
<p>Methods and classes/constructors are not covered in this section.</p>
<h4 id="function-declarations" class="numbered">Prefer function declarations for named functions</h4>
<p>Prefer function declarations over arrow functions or function expressions when
defining named functions.</p>
<pre><code class="language-.ts good">function foo() {
return 42;
}
2021-01-02 23:30:59 +08:00
</code></pre>
<pre><code class="language-ts bad">const foo = () =&gt; 42;
2021-01-02 23:30:59 +08:00
</code></pre>
<p>Arrow functions <em>may</em> be used, for example, when an explicit type annotation is
required.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">interface SearchFunction {
(source: string, subString: string): boolean;
2021-01-02 23:30:59 +08:00
}
const fooSearch: SearchFunction = (source, subString) =&gt; { ... };
2021-01-02 23:30:59 +08:00
</code></pre>
<p><a id="features-functions-nested-functions"></a></p>
<h4 id="nested-functions" class="numbered">Nested functions</h4>
<p>Functions nested within other methods or functions <em>may</em> use function
declarations or arrow functions, as appropriate. In method bodies in particular,
arrow functions are preferred because they have access to the outer <code>this</code>.</p>
<p><a id="use-arrow-functions-in-expressions"></a></p>
<h4 id="function-expressions" class="numbered">Do not use function expressions</h4>
<p>Do not use function expressions. Use arrow functions instead.</p>
<pre><code class="language-ts good">bar(() =&gt; { this.doSomething(); })
2021-01-02 23:30:59 +08:00
</code></pre>
<pre><code class="language-ts bad">bar(function() { ... })
</code></pre>
2021-01-02 23:30:59 +08:00
<p><strong>Exception:</strong> Function expressions <em>may</em> be used <em>only if</em> code has to
dynamically rebind <code>this</code> (but this is <a href="#rebinding-this">discouraged</a>), or for
generator functions (which do not have an arrow syntax).</p>
2021-01-02 23:30:59 +08:00
<p><a id="features-functions-arrow-functions"></a>
<a id="expression-bodies-vs-block-bodies"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="arrow-function-bodies" class="numbered">Arrow function bodies</h4>
<p>Use arrow functions with concise bodies (i.e. expressions) or block bodies as
appropriate.</p>
<pre><code class="language-ts good">// Top level functions use function declarations.
function someFunction() {
// Block bodies are fine:
const receipts = books.map((b: Book) =&gt; {
const receipt = payMoney(b.price);
recordTransaction(receipt);
return receipt;
});
// Concise bodies are fine, too, if the return value is used:
const longThings = myValues.filter(v =&gt; v.length &gt; 1000).map(v =&gt; String(v));
function payMoney(amount: number) {
// function declarations are fine, but must not access `this`.
}
// Nested arrow functions may be assigned to a const.
const computeTax = (amount: number) =&gt; amount * 0.12;
2021-01-02 23:30:59 +08:00
}
</code></pre>
<p>Only use a concise body if the return value of the function is actually used.
The block body makes sure the return type is <code>void</code> then and prevents potential
side effects.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">// BAD: use a block body if the return value of the function is not used.
myPromise.then(v =&gt; console.log(v));
// BAD: this typechecks, but the return value still leaks.
let f: () =&gt; void;
f = () =&gt; 1;
</code></pre>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">// GOOD: return value is unused, use a block body.
myPromise.then(v =&gt; {
console.log(v);
});
// GOOD: code may use blocks for readability.
const transformed = [1, 2, 3].map(v =&gt; {
const intermediate = someComplicatedExpr(v);
const more = acrossManyLines(intermediate);
return worthWrapping(more);
});
// GOOD: explicit `void` ensures no leaked return value
myPromise.then(v =&gt; void console.log(v));
</code></pre>
2021-01-02 23:30:59 +08:00
<p>Tip: The <code>void</code> operator can be used to ensure an arrow function with an
expression body returns <code>undefined</code> when the result is unused.</p>
<h4 id="rebinding-this" class="numbered">Rebinding <code>this</code></h4>
<p>Function expressions and function declarations <em>must not</em> use <code>this</code> unless they
specifically exist to rebind the <code>this</code> pointer. Rebinding <code>this</code> can in most
cases be avoided by using arrow functions or explicit parameters.</p>
<pre><code class="language-ts bad">function clickHandler() {
// Bad: what's `this` in this context?
this.textContent = 'Hello';
2021-01-02 23:30:59 +08:00
}
// Bad: the `this` pointer reference is implicitly set to document.body.
document.body.onclick = clickHandler;
2021-01-02 23:30:59 +08:00
</code></pre>
<pre><code class="language-ts good">// Good: explicitly reference the object from an arrow function.
document.body.onclick = () =&gt; { document.body.textContent = 'hello'; };
// Alternatively: take an explicit parameter
const setTextFn = (e: HTMLElement) =&gt; { e.textContent = 'hello'; };
document.body.onclick = setTextFn.bind(null, document.body);
</code></pre>
2021-01-02 23:30:59 +08:00
<p>Prefer arrow functions over other approaches to binding <code>this</code>, such as
<code>f.bind(this)</code>, <code>goog.bind(f, this)</code>, or <code>const self = this</code>.</p>
<h4 id="functions-as-callbacks" class="numbered">Prefer passing arrow functions as callbacks</h4>
<p>Callbacks can be invoked with unexpected arguments that can pass a type check
but still result in logical errors.</p>
<p>Avoid passing a named callback to a higher-order function, unless you are sure
of the stability of both functions' call signatures. Beware, in particular, of
less-commonly-used optional parameters.</p>
<pre><code class="language-ts bad">// BAD: Arguments are not explicitly passed, leading to unintended behavior
// when the optional `radix` argument gets the array indices 0, 1, and 2.
const numbers = ['11', '5', '10'].map(parseInt);
// &gt; [11, NaN, 2];
2021-01-02 23:30:59 +08:00
</code></pre>
<p>Instead, prefer passing an arrow-function that explicitly forwards parameters to
the named callback.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">// GOOD: Arguments are explicitly passed to the callback
const numbers = ['11', '5', '3'].map((n) =&gt; parseInt(n));
// &gt; [11, 5, 3]
2021-01-02 23:30:59 +08:00
// GOOD: Function is locally defined and is designed to be used as a callback
function dayFilter(element: string|null|undefined) {
return element != null &amp;&amp; element.endsWith('day');
}
2021-01-02 23:30:59 +08:00
const days = ['tuesday', undefined, 'juice', 'wednesday'].filter(dayFilter);
</code></pre>
2021-01-02 23:30:59 +08:00
<h4 id="arrow-functions-as-properties" class="numbered">Arrow functions as properties</h4>
2021-01-02 23:30:59 +08:00
<p>Classes usually <em>should not</em> contain properties initialized to arrow functions.
Arrow function properties require the calling function to understand that the
callee's <code>this</code> is already bound, which increases confusion about what <code>this</code>
is, and call sites and references using such handlers look broken (i.e. require
non-local knowledge to determine that they are correct). Code <em>should</em> always
use arrow functions to call instance methods (<code>const handler = (x) =&gt; {
this.listener(x); };</code>), and <em>should not</em> obtain or pass references to instance
methods (<del><code>const handler = this.listener; handler(x);</code></del>).</p>
2021-01-02 23:30:59 +08:00
<blockquote>
<p>Note: in some specific situations, e.g. when binding functions in a template,
arrow functions as properties are useful and create much more readable code.
Use judgement with this rule. Also, see the
<a href="#event-handlers"><code>Event Handlers</code></a> section below.</p>
</blockquote>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">class DelayHandler {
constructor() {
// Problem: `this` is not preserved in the callback. `this` in the callback
// will not be an instance of DelayHandler.
setTimeout(this.patienceTracker, 5000);
}
private patienceTracker() {
this.waitedPatiently = true;
}
}
</code></pre>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">// Arrow functions usually should not be properties.
class DelayHandler {
constructor() {
// Bad: this code looks like it forgot to bind `this`.
setTimeout(this.patienceTracker, 5000);
}
private patienceTracker = () =&gt; {
this.waitedPatiently = true;
2021-01-02 23:30:59 +08:00
}
}
</code></pre>
<pre><code class="language-ts good">// Explicitly manage `this` at call time.
class DelayHandler {
constructor() {
// Use anonymous functions if possible.
setTimeout(() =&gt; {
this.patienceTracker();
}, 5000);
}
private patienceTracker() {
this.waitedPatiently = true;
}
2021-01-02 23:30:59 +08:00
}
</code></pre>
<h4 id="event-handlers" class="numbered">Event handlers</h4>
2021-01-02 23:30:59 +08:00
<p>Event handlers <em>may</em> use arrow functions when there is no need to uninstall the
handler (for example, if the event is emitted by the class itself). If the
handler requires uninstallation, arrow function properties are the right
approach, because they automatically capture <code>this</code> and provide a stable
reference to uninstall.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">// Event handlers may be anonymous functions or arrow function properties.
class Component {
onAttached() {
// The event is emitted by this class, no need to uninstall.
this.addEventListener('click', () =&gt; {
this.listener();
});
// this.listener is a stable reference, we can uninstall it later.
window.addEventListener('onbeforeunload', this.listener);
}
onDetached() {
// The event is emitted by window. If we don't uninstall, this.listener will
// keep a reference to `this` because it's bound, causing a memory leak.
window.removeEventListener('onbeforeunload', this.listener);
}
// An arrow function stored in a property is bound to `this` automatically.
private listener = () =&gt; {
confirm('Do you want to exit the page?');
}
}
</code></pre>
2021-01-02 23:30:59 +08:00
<p>Do not use <code>bind</code> in the expression that installs an event handler, because it
creates a temporary reference that can't be uninstalled.</p>
<pre><code class="language-ts bad">// Binding listeners creates a temporary reference that prevents uninstalling.
class Component {
onAttached() {
// This creates a temporary reference that we won't be able to uninstall
window.addEventListener('onbeforeunload', this.listener.bind(this));
}
onDetached() {
// This bind creates a different reference, so this line does nothing.
window.removeEventListener('onbeforeunload', this.listener.bind(this));
}
private listener() {
confirm('Do you want to exit the page?');
2021-01-02 23:30:59 +08:00
}
}
</code></pre>
<p><a id="features-functions-default-parameters"></a></p>
<h4 id="parameter-initializers" class="numbered">Parameter initializers</h4>
<p>Optional function parameters <em>may</em> be given a default initializer to use when
the argument is omitted. Initializers <em>must not</em> have any observable side
effects. Initializers <em>should</em> be kept as simple as possible.</p>
<pre><code class="language-ts good">function process(name: string, extraContext: string[] = []) {}
function activate(index = 0) {}
</code></pre>
<pre><code class="language-ts bad">// BAD: side effect of incrementing the counter
let globalCounter = 0;
function newId(index = globalCounter++) {}
// BAD: exposes shared mutable state, which can introduce unintended coupling
// between function calls
class Foo {
private readonly defaultPaths: string[];
frobnicate(paths = defaultPaths) {}
2021-01-02 23:30:59 +08:00
}
</code></pre>
<p>Use default parameters sparingly. Prefer
<a href="#features-objects-destructuring">destructuring</a> to create readable APIs when
there are more than a small handful of optional parameters that do not have a
natural order.</p>
2021-01-02 23:30:59 +08:00
<p><a id="features-functions-rest-parameters"></a>
<a id="features-functions-spread-operator"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="rest-and-spread" class="numbered">Prefer rest and spread when appropriate</h4>
2021-01-02 23:30:59 +08:00
<p>Use a <em>rest</em> parameter instead of accessing <code>arguments</code>. Never name a local
variable or parameter <code>arguments</code>, which confusingly shadows the built-in name.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">function variadic(array: string[], ...numbers: number[]) {}
</code></pre>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<p>Use function spread syntax instead of <code>Function.prototype.apply</code>.</p>
2021-01-02 23:30:59 +08:00
<p><a id="features-functions-generators"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="formatting-functions" class="numbered">Formatting functions</h4>
2021-01-02 23:30:59 +08:00
<p>Blank lines at the start or end of the function body are not allowed.</p>
2021-01-02 23:30:59 +08:00
<p>A single blank line <em>may</em> be used within function bodies sparingly to create
<em>logical groupings</em> of statements.</p>
2021-01-02 23:30:59 +08:00
<p>Generators should attach the <code>*</code> to the <code>function</code> and <code>yield</code> keywords, as in
<code>function* foo()</code> and <code>yield* iter</code>, rather than <del><code>function *foo()</code></del> or
<del><code>yield *iter</code></del>.</p>
2021-01-02 23:30:59 +08:00
<p>Parentheses around the left-hand side of a single-argument arrow function are
recommended but not required.</p>
2021-01-02 23:30:59 +08:00
<p>Do not put a space after the <code>...</code> in rest or spread syntax.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">function myFunction(...elements: number[]) {}
myFunction(...array, ...iterable, ...generator());
2021-01-02 23:30:59 +08:00
</code></pre>
<h3 id="features-this" class="numbered">this</h3>
2021-01-02 23:30:59 +08:00
<p>Only use <code>this</code> in class constructors and methods, functions that have an
explicit <code>this</code> type declared (e.g. <code>function func(this: ThisType, ...)</code>), or in
arrow functions defined in a scope where <code>this</code> may be used.</p>
2021-01-02 23:30:59 +08:00
<p>Never use <code>this</code> to refer to the global object, the context of an <code>eval</code>, the
target of an event, or unnecessarily <code>call()</code>ed or <code>apply()</code>ed functions.</p>
<pre><code class="language-ts bad">this.alert('Hello');
</code></pre>
<h3 id="interfaces" class="numbered">Interfaces</h3>
<h3 id="primitive-literals" class="numbered">Primitive literals</h3>
<p><a id="features-string-literals"></a></p>
<h4 id="string-literals" class="numbered">String literals</h4>
<p><a id="features-strings-use-single-quotes"></a></p>
<h5 id="use-single-quotes" class="numbered">Use single quotes</h5>
<p>Ordinary string literals are delimited with single quotes (<code>'</code>), rather than
double quotes (<code>"</code>).</p>
2021-01-02 23:30:59 +08:00
<p>Tip: if a string contains a single quote character, consider using a template
string to avoid having to escape the quote.</p>
2021-01-02 23:30:59 +08:00
<p><a id="syntax-no-line-continuations"></a>
<a id="features-strings-no-line-continuations"></a></p>
2021-01-02 23:30:59 +08:00
<h5 id="no-line-continuations" class="numbered">No line continuations</h5>
2021-01-02 23:30:59 +08:00
<p>Do not use <em>line continuations</em> (that is, ending a line inside a string literal
with a backslash) in either ordinary or template string literals. Even though
ES5 allows this, it can lead to tricky errors if any trailing whitespace comes
after the slash, and is less obvious to readers.</p>
2021-01-02 23:30:59 +08:00
<p>Disallowed:</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">const LONG_STRING = 'This is a very very very very very very very long string. \
It inadvertently contains long stretches of spaces due to how the \
continued lines are indented.';
</code></pre>
2021-01-02 23:30:59 +08:00
<p>Instead, write</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">const LONG_STRING = 'This is a very very very very very very long string. ' +
'It does not contain long stretches of spaces because it uses ' +
'concatenated strings.';
const SINGLE_STRING =
'http://it.is.also/acceptable_to_use_a_single_long_string_when_breaking_would_hinder_search_discoverability';
2021-01-02 23:30:59 +08:00
</code></pre>
<p><a id="features-strings-template-strings"></a></p>
2021-01-02 23:30:59 +08:00
<h5 id="template-literals" class="numbered">Template literals</h5>
2021-01-02 23:30:59 +08:00
<p>Use template literals (delimited with <code>`</code>) over complex string
concatenation, particularly if multiple string literals are involved. Template
literals may span multiple lines.</p>
2021-01-02 23:30:59 +08:00
<p>If a template literal spans multiple lines, it does not need to follow the
indentation of the enclosing block, though it may if the added whitespace does
not matter.</p>
2021-01-02 23:30:59 +08:00
<p>Example:</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">function arithmetic(a: number, b: number) {
return `Here is a table of arithmetic operations:
${a} + ${b} = ${a + b}
${a} - ${b} = ${a - b}
${a} * ${b} = ${a * b}
${a} / ${b} = ${a / b}`;
}
2021-01-02 23:30:59 +08:00
</code></pre>
<p><a id="features-number-literals"></a></p>
<h4 id="number-literals" class="numbered">Number literals</h4>
<p>Numbers may be specified in decimal, hex, octal, or binary. Use exactly <code>0x</code>,
<code>0o</code>, and <code>0b</code> prefixes, with lowercase letters, for hex, octal, and binary,
respectively. Never include a leading zero unless it is immediately followed by
<code>x</code>, <code>o</code>, or <code>b</code>.</p>
<h4 id="type-coercion" class="numbered">Type coercion</h4>
2021-01-02 23:30:59 +08:00
<p>TypeScript code <em>may</em> use the <code>String()</code> and <code>Boolean()</code> (note: no <code>new</code>!)
2021-01-02 23:30:59 +08:00
functions, string template literals, or <code>!!</code> to coerce types.</p>
<pre><code class="language-ts good">const bool = Boolean(false);
const str = String(aNumber);
const bool2 = !!str;
const str2 = `result: ${bool2}`;
</code></pre>
<p>Values of enum types (including unions of enum types and other types) <em>must not</em>
be converted to booleans with <code>Boolean()</code> or <code>!!</code>, and must instead be compared
explicitly with comparison operators.</p>
<pre><code class="language-ts bad">enum SupportLevel {
NONE,
BASIC,
ADVANCED,
}
const level: SupportLevel = ...;
let enabled = Boolean(level);
const maybeLevel: SupportLevel|undefined = ...;
enabled = !!maybeLevel;
</code></pre>
<pre><code class="language-ts good">enum SupportLevel {
NONE,
BASIC,
ADVANCED,
}
const level: SupportLevel = ...;
let enabled = level !== SupportLevel.NONE;
const maybeLevel: SupportLevel|undefined = ...;
enabled = level !== undefined &amp;&amp; level !== SupportLevel.NONE;
</code></pre>
<section class="zippy">
<p>Why?</p>
<p>For most purposes, it doesn't matter what number or string value an enum name is
mapped to at runtime, because values of enum types are referred to by name in
source code. Consequently, engineers are accustomed to not thinking about this,
and so situations where it <em>does</em> matter are undesirable because they will be
surprising. Such is the case with conversion of enums to booleans; in
particular, by default, the first declared enum value is falsy (because it is 0)
while the others are truthy, which is likely to be unexpected. Readers of code
that uses an enum value may not even know whether it's the first declared value
or not.</p>
</section>
2021-01-02 23:30:59 +08:00
<p>Using string concatenation to cast to string is discouraged, as we check that
operands to the plus operator are of matching types.</p>
<p>Code <em>must</em> use <code>Number()</code> to parse numeric values, and <em>must</em> check its return
2021-01-02 23:30:59 +08:00
for <code>NaN</code> values explicitly, unless failing to parse is impossible from context.</p>
<p>Note: <code>Number('')</code>, <code>Number(' ')</code>, and <code>Number('\t')</code> would return <code>0</code> instead
of <code>NaN</code>. <code>Number('Infinity')</code> and <code>Number('-Infinity')</code> would return <code>Infinity</code>
and <code>-Infinity</code> respectively. Additionally, exponential notation such as
<code>Number('1e+309')</code> and <code>Number('-1e+309')</code> can overflow into <code>Infinity</code>. These
cases may require special handling.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">const aNumber = Number('123');
if (!isFinite(aNumber)) throw new Error(...);
2021-01-02 23:30:59 +08:00
</code></pre>
<p>Code <em>must not</em> use unary plus (<code>+</code>) to coerce strings to numbers. Parsing
numbers can fail, has surprising corner cases, and can be a code smell (parsing
at the wrong layer). A unary plus is too easy to miss in code reviews given
this.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">const x = +y;
</code></pre>
<p>Code also <em>must not</em> use <code>parseInt</code> or <code>parseFloat</code> to parse numbers, except for
2021-01-02 23:30:59 +08:00
non-base-10 strings (see below). Both of those functions ignore trailing
characters in the string, which can shadow error conditions (e.g. parsing <code>12
dwarves</code> as <code>12</code>).</p>
<pre><code class="language-ts bad">const n = parseInt(someString, 10); // Error prone,
const f = parseFloat(someString); // regardless of passing a radix.
</code></pre>
<p>Code that requires parsing with a radix <em>must</em> check that its input contains
only appropriate digits for that radix before calling into <code>parseInt</code>;</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">if (!/^[a-fA-F0-9]+$/.test(someString)) throw new Error(...);
// Needed to parse hexadecimal.
// tslint:disable-next-line:ban
const n = parseInt(someString, 16); // Only allowed for radix != 10
</code></pre>
<p>Use <code>Number()</code> followed by <code>Math.floor</code> or <code>Math.trunc</code> (where available) to
parse integer numbers:</p>
<pre><code class="language-ts good">let f = Number(someString);
if (isNaN(f)) handleError();
f = Math.floor(f);
</code></pre>
<p><a id="type-coercion-implicit"></a></p>
<h5 id="implicit-coercion" class="numbered">Implicit coercion</h5>
2021-01-02 23:30:59 +08:00
<p>Do not use explicit boolean coercions in conditional clauses that have implicit
boolean coercion. Those are the conditions in an <code>if</code>, <code>for</code> and <code>while</code>
statements.</p>
<pre><code class="language-ts bad">const foo: MyInterface|null = ...;
if (!!foo) {...}
while (!!foo) {...}
</code></pre>
<pre><code class="language-ts good">const foo: MyInterface|null = ...;
if (foo) {...}
while (foo) {...}
</code></pre>
<p><a href="#type-coercion">As with explicit conversions</a>, values of enum types (including
unions of enum types and other types) <em>must not</em> be implicitly coerced to
booleans, and must instead be compared explicitly with comparison operators.</p>
<pre><code class="language-ts bad">enum SupportLevel {
NONE,
BASIC,
ADVANCED,
}
const level: SupportLevel = ...;
if (level) {...}
const maybeLevel: SupportLevel|undefined = ...;
if (level) {...}
</code></pre>
<pre><code class="language-ts good">enum SupportLevel {
NONE,
BASIC,
ADVANCED,
}
const level: SupportLevel = ...;
if (level !== SupportLevel.NONE) {...}
const maybeLevel: SupportLevel|undefined = ...;
if (level !== undefined &amp;&amp; level !== SupportLevel.NONE) {...}
</code></pre>
<p>Other types of values may be either implicitly coerced to booleans or compared
explicitly with comparison operators:</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">// Explicitly comparing &gt; 0 is OK:
if (arr.length &gt; 0) {...}
// so is relying on boolean coercion:
if (arr.length) {...}
2021-01-02 23:30:59 +08:00
</code></pre>
<p><a id="features-control-structures"></a></p>
2021-01-02 23:30:59 +08:00
<h3 id="control-structures" class="numbered">Control structures</h3>
2021-01-02 23:30:59 +08:00
<p><a id="formatting-braces"></a>
<a id="formatting-braces-all"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="control-flow-statements-blocks" class="numbered">Control flow statements and blocks</h4>
2021-01-02 23:30:59 +08:00
<p>Control flow statements (<code>if</code>, <code>else</code>, <code>for</code>, <code>do</code>, <code>while</code>, etc) always use
braced blocks for the containing code, even if the body contains only a single
statement. The first statement of a non-empty block must begin on its own line.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">for (let i = 0; i &lt; x; i++) {
doSomethingWith(i);
}
2021-01-02 23:30:59 +08:00
if (x) {
doSomethingWithALongMethodNameThatForcesANewLine(x);
2021-01-02 23:30:59 +08:00
}
</code></pre>
<pre><code class="language-ts bad">if (x)
doSomethingWithALongMethodNameThatForcesANewLine(x);
for (let i = 0; i &lt; x; i++) doSomethingWith(i);
2021-01-02 23:30:59 +08:00
</code></pre>
<p><strong>Exception:</strong> <code>if</code> statements fitting on one line <em>may</em> elide the block.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">if (x) x.doFoo();
</code></pre>
<h5 id="assignment-in-control-statements" class="numbered">Assignment in control statements</h5>
2021-01-02 23:30:59 +08:00
<p>Prefer to avoid assignment of variables inside control statements. Assignment
can be easily mistaken for equality checks inside control statements.</p>
<pre><code class="language-ts bad">if (x = someFunction()) {
// Assignment easily mistaken with equality check
// ...
}
</code></pre>
<pre><code class="language-ts good">x = someFunction();
if (x) {
// ...
}
</code></pre>
<p>In cases where assignment inside the control statement is preferred, enclose the
assignment in additional parenthesis to indicate it is intentional.</p>
<pre><code class="language-ts">while ((x = someFunction())) {
// Double parenthesis shows assignment is intentional
// ...
}
</code></pre>
<p><a id="features-for-loops"></a></p>
<h5 id="iterating-containers" class="numbered">Iterating containers</h5>
2021-01-02 23:30:59 +08:00
<p>Prefer <code>for (... of someArr)</code> to iterate over arrays. <code>Array.prototype.forEach</code> and vanilla <code>for</code>
loops are also allowed:</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">for (const x of someArr) {
// x is a value of someArr.
}
2021-01-02 23:30:59 +08:00
for (let i = 0; i &lt; someArr.length; i++) {
// Explicitly count if the index is needed, otherwise use the for/of form.
const x = someArr[i];
// ...
}
for (const [i, x] of someArr.entries()) {
// Alternative version of the above.
2021-01-02 23:30:59 +08:00
}
</code></pre>
<p><code>for</code>-<code>in</code> loops may only be used on dict-style objects (see
<a href="#optimization-compatibility-for-property-access">below</a> for more info). Do not
use <code>for (... in ...)</code> to iterate over arrays as it will counterintuitively give
the array's indices (as strings!), not values:</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">for (const x in someArray) {
// x is the index!
2021-01-02 23:30:59 +08:00
}
</code></pre>
<p><code>Object.prototype.hasOwnProperty</code> should be used in <code>for</code>-<code>in</code> loops to exclude
unwanted prototype properties. Prefer <code>for</code>-<code>of</code> with <code>Object.keys</code>,
<code>Object.values</code>, or <code>Object.entries</code> over <code>for</code>-<code>in</code> when possible.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">for (const key in obj) {
if (!obj.hasOwnProperty(key)) continue;
doWork(key, obj[key]);
2021-01-02 23:30:59 +08:00
}
for (const key of Object.keys(obj)) {
doWork(key, obj[key]);
}
for (const value of Object.values(obj)) {
doWorkValOnly(value);
}
for (const [key, value] of Object.entries(obj)) {
doWork(key, value);
2021-01-02 23:30:59 +08:00
}
</code></pre>
<p><a id="formatting-grouping-parentheses"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="grouping-parentheses" class="numbered">Grouping parentheses</h4>
2021-01-02 23:30:59 +08:00
<p>Optional grouping parentheses are omitted only when the author and reviewer
agree that there is no reasonable chance that the code will be misinterpreted
without them, nor would they have made the code easier to read. It is <em>not</em>
reasonable to assume that every reader has the entire operator precedence table
memorized.</p>
2021-01-02 23:30:59 +08:00
<p>Do not use unnecessary parentheses around the entire expression following
<code>delete</code>, <code>typeof</code>, <code>void</code>, <code>return</code>, <code>throw</code>, <code>case</code>, <code>in</code>, <code>of</code>, or <code>yield</code>.</p>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<p><a id="exceptions"></a>
<a id="features-exceptions"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="exception-handling" class="numbered">Exception handling</h4>
2021-01-02 23:30:59 +08:00
<p>Exceptions are an important part of the language and should be used whenever
exceptional cases occur.</p>
2021-01-02 23:30:59 +08:00
<p>Custom exceptions provide a great way to convey additional error information
from functions. They should be defined and used wherever the native <code>Error</code> type
is insufficient.</p>
<p>Prefer throwing exceptions over ad-hoc error-handling approaches (such as
passing an error container reference type, or returning an object with an error
property).</p>
2021-01-02 23:30:59 +08:00
<h5 id="instantiate-errors-using-new" class="numbered">Instantiate errors using <code>new</code></h5>
2021-01-02 23:30:59 +08:00
<p>Always use <code>new Error()</code> when instantiating exceptions, instead of just calling
<code>Error()</code>. Both forms create a new <code>Error</code> instance, but using <code>new</code> is more
consistent with how other objects are instantiated.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">throw new Error('Foo is not a valid bar.');
2021-01-02 23:30:59 +08:00
</code></pre>
<pre><code class="language-ts bad">throw Error('Foo is not a valid bar.');
</code></pre>
2021-01-02 23:30:59 +08:00
<h5 id="only-throw-errors" class="numbered">Only throw errors</h5>
2021-01-02 23:30:59 +08:00
<p>JavaScript (and thus TypeScript) allow throwing or rejecting a Promise with
arbitrary values. However if the thrown or rejected value is not an <code>Error</code>, it
does not populate stack trace information, making debugging hard. This treatment
extends to <code>Promise</code> rejection values as <code>Promise.reject(obj)</code> is equivalent to
<code>throw obj;</code> in async functions.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">// bad: does not get a stack trace.
throw 'oh noes!';
// For promises
new Promise((resolve, reject) =&gt; void reject('oh noes!'));
Promise.reject();
Promise.reject('oh noes!');
2021-01-02 23:30:59 +08:00
</code></pre>
<p>Instead, only throw (subclasses of) <code>Error</code>:</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">// Throw only Errors
throw new Error('oh noes!');
// ... or subtypes of Error.
class MyError extends Error {}
throw new MyError('my oh noes!');
// For promises
new Promise((resolve) =&gt; resolve()); // No reject is OK.
new Promise((resolve, reject) =&gt; void reject(new Error('oh noes!')));
Promise.reject(new Error('oh noes!'));
</code></pre>
2021-01-02 23:30:59 +08:00
<h5 id="catching-and-rethrowing" class="numbered">Catching and rethrowing</h5>
2021-01-02 23:30:59 +08:00
<p>When catching errors, code <em>should</em> assume that all thrown errors are instances
of <code>Error</code>.</p>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<section>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">function assertIsError(e: unknown): asserts e is Error {
if (!(e instanceof Error)) throw new Error("e is not an Error");
}
try {
doSomething();
} catch (e: unknown) {
// All thrown errors must be Error subtypes. Do not handle
// other possible values unless you know they are thrown.
assertIsError(e);
displayError(e.message);
// or rethrow:
throw e;
2021-01-02 23:30:59 +08:00
}
</code></pre>
</section>
2021-01-02 23:30:59 +08:00
<p>Exception handlers <em>must not</em> defensively handle non-<code>Error</code> types unless the
called API is conclusively known to throw non-<code>Error</code>s in violation of the above
rule. In that case, a comment should be included to specifically identify where
the non-<code>Error</code>s originate.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">try {
badApiThrowingStrings();
} catch (e: unknown) {
// Note: bad API throws strings instead of errors.
if (typeof e === 'string') { ... }
}
2021-01-02 23:30:59 +08:00
</code></pre>
<section class="zippy">
2021-01-02 23:30:59 +08:00
<p>Why?</p>
2021-01-02 23:30:59 +08:00
<p>Avoid
<a href="https://en.wikipedia.org/wiki/Defensive_programming#Offensive_programming">overly defensive programming</a>.
Repeating the same defenses against a problem that will not exist in most code
leads to boiler-plate code that is not useful.</p>
2021-01-02 23:30:59 +08:00
</section>
2021-01-02 23:30:59 +08:00
<p><a id="features-empty-catch-blocks"></a></p>
2021-01-02 23:30:59 +08:00
<h5 id="empty-catch-blocks" class="numbered">Empty catch blocks</h5>
2021-01-02 23:30:59 +08:00
<p>It is very rarely correct to do nothing in response to a caught exception. When
it truly is appropriate to take no action whatsoever in a catch block, the
reason this is justified is explained in a comment.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good"> try {
return handleNumericResponse(response);
} catch (e: unknown) {
// Response is not numeric. Continue to handle as text.
2021-01-02 23:30:59 +08:00
}
return handleTextResponse(response);
2021-01-02 23:30:59 +08:00
</code></pre>
<p>Disallowed:</p>
<pre><code class="language-ts bad"> try {
shouldFail();
fail('expected an error');
} catch (expected: unknown) {
2021-01-02 23:30:59 +08:00
}
</code></pre>
<p>Tip: Unlike in some other languages, patterns like the above simply dont work
since this will catch the error thrown by <code>fail</code>. Use <code>assertThrows()</code> instead.</p>
<p><a id="features-switch-statements"></a>
<a id="features-switch-default-case"></a></p>
<h4 id="switch-statements" class="numbered">Switch statements</h4>
<p>All <code>switch</code> statements <em>must</em> contain a <code>default</code> statement group, even if it
contains no code. The <code>default</code> statement group must be last.</p>
<pre><code class="language-ts good">switch (x) {
case Y:
doSomethingElse();
break;
default:
// nothing to do.
2021-01-02 23:30:59 +08:00
}
</code></pre>
<p>Within a switch block, each statement group either terminates abruptly with a
<code>break</code>, a <code>return</code> statement, or by throwing an exception. Non-empty statement
groups (<code>case ...</code>) <em>must not</em> fall through (enforced by the compiler):</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">switch (x) {
case X:
doSomething();
// fall through - not allowed!
case Y:
// ...
2021-01-02 23:30:59 +08:00
}
</code></pre>
<p>Empty statement groups are allowed to fall through:</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">switch (x) {
case X:
case Y:
doSomething();
break;
default: // nothing to do.
2021-01-02 23:30:59 +08:00
}
</code></pre>
2021-01-02 23:30:59 +08:00
<p><a id="features-equality-checks"></a>
<a id="features-equality-checks-exceptions"></a></p>
<h4 id="equality-checks" class="numbered">Equality checks</h4>
<p>Always use triple equals (<code>===</code>) and not equals (<code>!==</code>). The double equality
operators cause error prone type coercions that are hard to understand and
slower to implement for JavaScript Virtual Machines. See also the
<a href="https://dorey.github.io/JavaScript-Equality-Table/">JavaScript equality table</a>.</p>
<pre><code class="language-ts bad">if (foo == 'bar' || baz != bam) {
// Hard to understand behaviour due to type coercion.
}
</code></pre>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">if (foo === 'bar' || baz !== bam) {
// All good here.
}
</code></pre>
2021-01-02 23:30:59 +08:00
<p><strong>Exception:</strong> Comparisons to the literal <code>null</code> value <em>may</em> use the <code>==</code> and
<code>!=</code> operators to cover both <code>null</code> and <code>undefined</code> values.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">if (foo == null) {
// Will trigger when foo is null or undefined.
}
</code></pre>
2021-01-02 23:30:59 +08:00
<h4 id="type-and-non-nullability-assertions" class="numbered">Type and non-nullability assertions</h4>
2021-01-02 23:30:59 +08:00
<p>Type assertions (<code>x as SomeType</code>) and non-nullability assertions (<code>y!</code>) are
unsafe. Both only silence the TypeScript compiler, but do not insert any runtime
checks to match these assertions, so they can cause your program to crash at
runtime.</p>
<p>Because of this, you <em>should not</em> use type and non-nullability assertions
without an obvious or explicit reason for doing so.</p>
<p>Instead of the following:</p>
<pre><code class="language-ts bad">(x as Foo).foo();
y!.bar();
</code></pre>
<p>When you want to assert a type or non-nullability the best answer is to
explicitly write a runtime check that performs that check.</p>
<pre><code class="language-ts good">// assuming Foo is a class.
if (x instanceof Foo) {
x.foo();
}
if (y) {
y.bar();
}
</code></pre>
<p>Sometimes due to some local property of your code you can be sure that the
assertion form is safe. In those situations, you <em>should</em> add clarification to
explain why you are ok with the unsafe behavior:</p>
<pre><code class="language-ts good">// x is a Foo, because ...
(x as Foo).foo();
// y cannot be null, because ...
y!.bar();
</code></pre>
<p>If the reasoning behind a type or non-nullability assertion is obvious, the
comments <em>may</em> not be necessary. For example, generated proto code is always
2021-01-02 23:30:59 +08:00
nullable, but perhaps it is well-known in the context of the code that certain
fields are always provided by the backend. Use your judgement.</p>
<h5 id="type-assertions-syntax" class="numbered">Type assertion syntax</h5>
2021-01-02 23:30:59 +08:00
<p>Type assertions <em>must</em> use the <code>as</code> syntax (as opposed to the angle brackets
2021-01-02 23:30:59 +08:00
syntax). This enforces parentheses around the assertion when accessing a member.</p>
<pre><code class="language-ts bad">const x = (&lt;Foo&gt;z).length;
const y = &lt;Foo&gt;z.length;
</code></pre>
<pre><code class="language-ts good">// z must be Foo because ...
const x = (z as Foo).length;
2021-01-02 23:30:59 +08:00
</code></pre>
<h5 id="double-assertions" class="numbered">Double assertions</h5>
<p>From the
<a href="https://www.typescriptlang.org/docs/handbook/2/everyday-types.html#type-assertions">TypeScript handbook</a>,
TypeScript only allows type assertions which convert to a <em>more specific</em> or
<em>less specific</em> version of a type. Adding a type assertion (<code>x as Foo</code>) which
does not meet this criteria will give the error: <q>Conversion of type 'X' to type
'Y' may be a mistake because neither type sufficiently overlaps with the other.</q></p>
<p>If you are sure an assertion is safe, you can perform a <em>double assertion</em>. This
involves casting through <code>unknown</code> since it is less specific than all types.</p>
<pre><code class="language-ts good">// x is a Foo here, because...
(x as unknown as Foo).fooMethod();
</code></pre>
<p>Use <code>unknown</code> (instead of <code>any</code> or <code>{}</code>) as the intermediate type.</p>
<h5 id="type-assertions-and-object-literals" class="numbered">Type assertions and object literals</h5>
2021-01-02 23:30:59 +08:00
<p>Use type annotations (<code>: Foo</code>) instead of type assertions (<code>as Foo</code>) to specify
the type of an object literal. This allows detecting refactoring bugs when the
fields of an interface change over time.</p>
<pre><code class="language-ts bad">interface Foo {
bar: number;
baz?: string; // was "bam", but later renamed to "baz".
}
const foo = {
bar: 123,
bam: 'abc', // no error!
} as Foo;
function func() {
return {
bar: 123,
bam: 'abc', // no error!
} as Foo;
}
</code></pre>
<pre><code class="language-ts good">interface Foo {
bar: number;
baz?: string;
}
const foo: Foo = {
bar: 123,
bam: 'abc', // complains about "bam" not being defined on Foo.
};
function func(): Foo {
return {
bar: 123,
bam: 'abc', // complains about "bam" not being defined on Foo.
};
}
</code></pre>
<h4 id="keep-try-blocks-focused" class="numbered">Keep try blocks focused</h4>
2021-01-02 23:30:59 +08:00
<p>Limit the amount of code inside a try block, if this can be done without hurting
readability.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">try {
const result = methodThatMayThrow();
use(result);
} catch (error: unknown) {
// ...
}
</code></pre>
<pre><code class="language-ts good">let result;
try {
result = methodThatMayThrow();
} catch (error: unknown) {
// ...
2021-01-02 23:30:59 +08:00
}
use(result);
2021-01-02 23:30:59 +08:00
</code></pre>
<p>Moving the non-throwable lines out of the try/catch block helps the reader learn
which method throws exceptions. Some inline calls that do not throw exceptions
could stay inside because they might not be worth the extra complication of a
temporary variable.</p>
2021-01-02 23:30:59 +08:00
<p><strong>Exception:</strong> There may be performance issues if try blocks are inside a loop.
Widening try blocks to cover a whole loop is ok.</p>
2021-01-02 23:30:59 +08:00
<h3 id="decorators" class="numbered">Decorators</h3>
2021-01-02 23:30:59 +08:00
<p>Decorators are syntax with an <code>@</code> prefix, like <code>@MyDecorator</code>.</p>
<p>Do not define new decorators. Only use the decorators defined by
frameworks:</p>
<ul>
<li>Angular (e.g. <code>@Component</code>, <code>@NgModule</code>, etc.)</li>
<li>Polymer (e.g. <code>@property</code>)</li>
</ul>
2021-01-02 23:30:59 +08:00
<section>
<p>Why?</p>
<p>We generally want to avoid decorators, because they were an experimental feature
that have since diverged from the TC39 proposal and have known bugs that won't
be fixed.</p>
</section>
2021-01-02 23:30:59 +08:00
<p>When using decorators, the decorator <em>must</em> immediately precede the symbol it
2021-01-02 23:30:59 +08:00
decorates, with no empty lines between:</p>
<pre><code class="language-ts">/** JSDoc comments go before decorators */
@Component({...}) // Note: no empty line after the decorator.
class MyComp {
@Input() myField: string; // Decorators on fields may be on the same line...
@Input()
myOtherField: string; // ... or wrap.
}
</code></pre>
<h3 id="disallowed-features" class="numbered">Disallowed features</h3>
2021-01-02 23:30:59 +08:00
<p><a id="disallowed-features-wrapper-objects"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="primitive-types-wrapper-classes" class="numbered">Wrapper objects for primitive types</h4>
2021-01-02 23:30:59 +08:00
<p>TypeScript code <em>must not</em> instantiate the wrapper classes for the primitive
types <code>String</code>, <code>Boolean</code>, and <code>Number</code>. Wrapper classes have surprising
behavior, such as <code>new Boolean(false)</code> evaluating to <code>true</code>.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">const s = new String('hello');
const b = new Boolean(false);
const n = new Number(5);
</code></pre>
2021-01-02 23:30:59 +08:00
<p>The wrappers may be called as functions for coercing (which is preferred over
using <code>+</code> or concatenating the empty string) or creating symbols. See
<a href="#type-coercion">type coercion</a> for more information.</p>
2021-01-02 23:30:59 +08:00
<p><a id="formatting-semicolons-are-required"></a>
<a id="disallowed-features-automatic-semicolon-insertion"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="automatic-semicolon-insertion" class="numbered">Automatic Semicolon Insertion</h4>
2021-01-02 23:30:59 +08:00
<p>Do not rely on Automatic Semicolon Insertion (ASI). Explicitly end all
statements using a semicolon. This prevents bugs due to incorrect semicolon
insertions and ensures compatibility with tools with limited ASI support (e.g.
clang-format).</p>
2021-01-02 23:30:59 +08:00
<h4 id="enums" class="numbered">Const enums</h4>
2021-01-02 23:30:59 +08:00
<p>Code <em>must not</em> use <code>const enum</code>; use plain <code>enum</code> instead.</p>
2021-01-02 23:30:59 +08:00
<section class="zippy">
2021-01-02 23:30:59 +08:00
<p>Why?</p>
2021-01-02 23:30:59 +08:00
<p>TypeScript enums already cannot be mutated; <code>const enum</code> is a separate language
feature related to optimization that makes the enum invisible to
JavaScript users of the module.</p>
2021-01-02 23:30:59 +08:00
</section>
2021-01-02 23:30:59 +08:00
<h4 id="debugger-statements" class="numbered">Debugger statements</h4>
2021-01-02 23:30:59 +08:00
<p>Debugger statements <em>must not</em> be included in production code.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts bad">function debugMe() {
debugger;
}
2021-01-02 23:30:59 +08:00
</code></pre>
<p><a id="disallowed-features-with"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="diallowed-features-with" class="numbered"><code>with</code></h4>
2021-01-02 23:30:59 +08:00
<p>Do not use the <code>with</code> keyword. It makes your code harder to understand and
<a href="https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Statements/with">has been banned in strict mode since ES5</a>.</p>
2021-01-02 23:30:59 +08:00
<p><a id="disallowed-features-dynamic-code-evaluation"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="dynamic-code-evaluation" class="numbered">Dynamic code evaluation</h4>
2021-01-02 23:30:59 +08:00
<p>Do not use <code>eval</code> or the <code>Function(...string)</code> constructor (except for code
loaders). These features are potentially dangerous and simply do not work in
environments using strict
<a href="https://developer.mozilla.org/en-US/docs/Web/HTTP/CSP">Content Security Policies</a>.</p>
2021-01-02 23:30:59 +08:00
<p><a id="disallowed-features-non-standard-features"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="non-standard-features" class="numbered">Non-standard features</h4>
2021-01-02 23:30:59 +08:00
<p>Do not use non-standard ECMAScript or Web Platform features.</p>
2021-01-02 23:30:59 +08:00
<p>This includes:</p>
2021-01-02 23:30:59 +08:00
<ul>
<li>Old features that have been marked deprecated or removed entirely from
ECMAScript / the Web Platform (see
<a href="https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Deprecated_and_obsolete_features">MDN</a>)</li>
<li>New ECMAScript features that are not yet standardized
<ul>
<li>Avoid using features that are in current TC39 working draft or currently
in the <a href="https://tc39.es/process-document/">proposal process</a></li>
<li>Use only ECMAScript features defined in the current ECMA-262
specification</li>
</ul></li>
<li>Proposed but not-yet-complete web standards:
<ul>
<li>WHATWG proposals that have not completed the
<a href="https://whatwg.org/faq#adding-new-features">proposal process</a>.</li>
</ul></li>
<li>Non-standard language “extensions” (such as those provided by some external
transpilers)</li>
</ul>
2021-01-02 23:30:59 +08:00
<p>Projects targeting specific JavaScript runtimes, such as latest-Chrome-only,
Chrome extensions, Node.JS, Electron, can obviously use those APIs. Use caution
when considering an API surface that is proprietary and only implemented in some
browsers; consider whether there is a common library that can abstract this API
surface away for you.</p>
2021-01-02 23:30:59 +08:00
<p><a id="disallowed-features-modifying-builtin-objects"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="modifying-builtin-objects" class="numbered">Modifying builtin objects</h4>
2021-01-02 23:30:59 +08:00
<p>Never modify builtin types, either by adding methods to their constructors or to
their prototypes. Avoid depending on libraries that do
this.</p>
2021-01-02 23:30:59 +08:00
<p>Do not add symbols to the global object unless absolutely necessary (e.g.
required by a third-party API).</p>
2021-01-02 23:30:59 +08:00
<p><a id="syntax"></a>
<a id="naming-rules-common-to-all-identifiers"></a></p>
2021-01-02 23:30:59 +08:00
<h2 id="naming" class="numbered">Naming</h2>
2021-01-02 23:30:59 +08:00
<h3 id="identifiers" class="numbered">Identifiers</h3>
2021-01-02 23:30:59 +08:00
<p>Identifiers <em>must</em> use only ASCII letters, digits, underscores (for constants
and structured test method names), and (rarely) the '$' sign.</p>
2021-01-02 23:30:59 +08:00
<h4 id="naming-style" class="numbered">Naming style</h4>
2021-01-02 23:30:59 +08:00
<p>TypeScript expresses information in types, so names <em>should not</em> be decorated
with information that is included in the type. (See also
<a href="https://testing.googleblog.com/2017/10/code-health-identifiernamingpostforworl.html">Testing Blog</a>
for more about what
not to include.)</p>
<p>Some concrete examples of this rule:</p>
<ul>
<li>Do not use trailing or leading underscores for private properties or
methods.</li>
<li>Do not use the <code>opt_</code> prefix for optional parameters.
<ul>
<li>For accessors, see <a href="#getters-and-setters-accessors">accessor rules</a>
below.</li>
</ul></li>
<li>Do not mark interfaces specially (<del><code>IMyInterface</code></del> or
<del><code>MyFooInterface</code></del>) unless it's idiomatic in its
environment. When
introducing an interface for a class, give it a name that expresses why the
interface exists in the first place (e.g. <code>class TodoItem</code> and <code>interface
TodoItemStorage</code> if the interface expresses the format used for
storage/serialization in JSON).</li>
<li>Suffixing <code>Observable</code>s with <code>$</code> is a common external convention and can
help resolve confusion regarding observable values vs concrete values.
Judgement on whether this is a useful convention is left up to individual
teams, but <em>should</em> be consistent within projects.</li>
</ul>
2021-01-02 23:30:59 +08:00
<p><a id="identifiers-abbreviations"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="descriptive-names" class="numbered">Descriptive names</h4>
2021-01-02 23:30:59 +08:00
<p>Names <em>must</em> be descriptive and clear to a new reader. Do not use abbreviations
that are ambiguous or unfamiliar to readers outside your project, and do not
abbreviate by deleting letters within a word.</p>
2021-01-02 23:30:59 +08:00
<ul>
<li><strong>Exception:</strong> Variables that are in scope for 10 lines or fewer, including
arguments that are <em>not</em> part of an exported API, <em>may</em> use short (e.g.
single letter) variable names.</li>
</ul>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">// Good identifiers:
errorCount // No abbreviation.
dnsConnectionIndex // Most people know what "DNS" stands for.
referrerUrl // Ditto for "URL".
customerId // "Id" is both ubiquitous and unlikely to be misunderstood.
2021-01-02 23:30:59 +08:00
</code></pre>
<pre><code class="language-ts bad">// Disallowed identifiers:
n // Meaningless.
nErr // Ambiguous abbreviation.
nCompConns // Ambiguous abbreviation.
wgcConnections // Only your group knows what this stands for.
pcReader // Lots of things can be abbreviated "pc".
cstmrId // Deletes internal letters.
kSecondsPerDay // Do not use Hungarian notation.
customerID // Incorrect camelcase of "ID".
2021-01-02 23:30:59 +08:00
</code></pre>
<p><a id="naming-camel-case-defined"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="camel-case" class="numbered">Camel case</h4>
2021-01-02 23:30:59 +08:00
<p>
Treat abbreviations like acronyms in names as whole words, i.e. use
<code>loadHttpUrl</code>, not <del><code>loadHTTPURL</code></del>, unless required by a platform name (e.g.
<code>XMLHttpRequest</code>).</p>
2021-01-02 23:30:59 +08:00
<h4 id="identifiers-dollar-sign" class="numbered">Dollar sign</h4>
2021-01-02 23:30:59 +08:00
<p>Identifiers <em>should not</em> generally use <code>$</code>, except when required by naming
conventions for third party frameworks. <a href="#naming-style">See above</a> for more on
using <code>$</code> with <code>Observable</code> values.</p>
2021-01-02 23:30:59 +08:00
<p><a id="naming-class-names"></a>
<a id="naming-method-names"></a>
<a id="naming-enum-names"></a>
<a id="naming-constant-names"></a>
<a id="naming-local-variable-names"></a>
<a id="naming-non-constant-field-names"></a>
<a id="naming-parameter-names"></a></p>
2021-01-02 23:30:59 +08:00
<h3 id="naming-rules-by-identifier-type" class="numbered">Rules by identifier type</h3>
2021-01-02 23:30:59 +08:00
<p>Most identifier names should follow the casing in the table below, based on the
identifier's type.</p>
2021-01-02 23:30:59 +08:00
<table>
<thead>
<tr>
<th>Style</th>
<th>Category</th>
2021-01-02 23:30:59 +08:00
</tr>
</thead>
<tbody>
<tr>
<td><code>UpperCamelCase</code>
2021-01-02 23:30:59 +08:00
</td>
<td>class / interface / type / enum / decorator / type
parameters / component functions in TSX / JSXElement type
parameter</td>
2021-01-02 23:30:59 +08:00
</tr>
<tr>
<td><code>lowerCamelCase</code>
2021-01-02 23:30:59 +08:00
</td>
<td>variable / parameter / function / method / property /
module alias</td>
2021-01-02 23:30:59 +08:00
</tr>
<tr>
<td><code>CONSTANT_CASE</code>
2021-01-02 23:30:59 +08:00
</td>
<td>global constant values, including enum values. See
<a href="#identifiers-constants">Constants</a> below.</td>
2021-01-02 23:30:59 +08:00
</tr>
<tr>
<td><code>#ident</code></td>
<td>private identifiers are never used.</td>
2021-01-02 23:30:59 +08:00
</tr>
</tbody>
</table>
<p><a id="naming-template-parameter-names"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="identifiers-type-parameters" class="numbered">Type parameters</h4>
2021-01-02 23:30:59 +08:00
<p>Type parameters, like in <code>Array&lt;T&gt;</code>, <em>may</em> use a single upper case character
(<code>T</code>) or <code>UpperCamelCase</code>.</p>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<h4 id="identifiers-test-names" class="numbered">Test names</h4>
2021-01-02 23:30:59 +08:00
<p>Test method names inxUnit-style test frameworks <em>may</em> be structured with <code>_</code> separators, e.g.
<code>testX_whenY_doesZ()</code>.</p>
2021-01-02 23:30:59 +08:00
<h4 id="identifiers-underscore-prefix-suffix" class="numbered"><code>_</code> prefix/suffix</h4>
2021-01-02 23:30:59 +08:00
<p>Identifiers must not use <code>_</code> as a prefix or suffix.</p>
2021-01-02 23:30:59 +08:00
<p>This also means that <code>_</code> <em>must not</em> be used as an identifier by itself (e.g. to
indicate a parameter is unused).</p>
2021-01-02 23:30:59 +08:00
<blockquote>
<p>Tip: If you only need some of the elements from an array (or TypeScript
tuple), you can insert extra commas in a destructuring statement to ignore
in-between elements:</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts">const [a, , b] = [1, 5, 10]; // a &lt;- 1, b &lt;- 10
2021-01-02 23:30:59 +08:00
</code></pre>
</blockquote>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<h4 id="identifiers-imports" class="numbered">Imports</h4>
2021-01-02 23:30:59 +08:00
<p>Module namespace imports are <code>lowerCamelCase</code> while files are <code>snake_case</code>,
which means that imports correctly will not match in casing style, such as</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">import * as fooBar from './foo_bar';
</code></pre>
2021-01-02 23:30:59 +08:00
<p>Some libraries might commonly use a namespace import prefix that violates this
naming scheme, but overbearingly common open source use makes the violating
style more readable. The only libraries that currently fall under this exception
are:</p>
2021-01-02 23:30:59 +08:00
<ul>
<li><a href="https://jquery.com/">jquery</a>, using the <code>$</code> prefix</li>
<li><a href="https://threejs.org/">threejs</a>, using the <code>THREE</code> prefix</li>
</ul>
2021-01-02 23:30:59 +08:00
<h4 id="identifiers-constants" class="numbered">Constants</h4>
2021-01-02 23:30:59 +08:00
<p><strong>Immutable</strong>: <code>CONSTANT_CASE</code> indicates that a value is <em>intended</em> to not be
changed, and <em>may</em> be used for values that can technically be modified (i.e.
values that are not deeply frozen) to indicate to users that they must not be
modified.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">const UNIT_SUFFIXES = {
'milliseconds': 'ms',
'seconds': 's',
};
// Even though per the rules of JavaScript UNIT_SUFFIXES is
// mutable, the uppercase shows users to not modify it.
2021-01-02 23:30:59 +08:00
</code></pre>
<p>A constant can also be a <code>static readonly</code> property of a class.</p>
<pre><code class="language-ts good">class Foo {
private static readonly MY_SPECIAL_NUMBER = 5;
2021-01-02 23:30:59 +08:00
bar() {
return 2 * Foo.MY_SPECIAL_NUMBER;
}
}
2021-01-02 23:30:59 +08:00
</code></pre>
<p><strong>Global</strong>: Only symbols declared on the module level, static fields of module
level classes, and values of module level enums, <em>may</em> use <code>CONST_CASE</code>. If a
value can be instantiated more than once over the lifetime of the program (e.g.
a local variable declared within a function, or a static field on a class nested
in a function) then it <em>must</em> use <code>lowerCamelCase</code>.</p>
<p>If a value is an arrow function that implements an interface, then it <em>may</em> be
declared <code>lowerCamelCase</code>.</p>
2021-01-02 23:30:59 +08:00
2021-01-02 23:30:59 +08:00
<p><a id="naming-local-aliases"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="aliases" class="numbered">Aliases</h4>
2021-01-02 23:30:59 +08:00
<p>When creating a local-scope alias of an existing symbol, use the format of the
existing identifier. The local alias <em>must</em> match the existing naming and format
of the source. For variables use <code>const</code> for your local aliases, and for class
fields use the <code>readonly</code> attribute.</p>
2021-01-02 23:30:59 +08:00
<blockquote>
<p>Note: If you're creating an alias just to expose it to a template in your
framework of choice, remember to also apply the proper
<a href="#properties-used-outside-of-class-lexical-scope">access modifiers</a>.</p>
</blockquote>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">const {BrewStateEnum} = SomeType;
const CAPACITY = 5;
2021-01-02 23:30:59 +08:00
class Teapot {
readonly BrewStateEnum = BrewStateEnum;
readonly CAPACITY = CAPACITY;
}
</code></pre>
2021-01-02 23:30:59 +08:00
<h2 id="type-system" class="numbered">Type system</h2>
2021-01-02 23:30:59 +08:00
<h3 id="type-inference" class="numbered">Type inference</h3>
2021-01-02 23:30:59 +08:00
<p>Code <em>may</em> rely on type inference as implemented by the TypeScript compiler for
all type expressions (variables, fields, return types, etc).
</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">const x = 15; // Type inferred.
</code></pre>
<p>Leave out type annotations for trivially inferred types: variables or parameters
initialized to a <code>string</code>, <code>number</code>, <code>boolean</code>, <code>RegExp</code> literal or <code>new</code>
expression.</p>
<pre><code class="language-ts bad">const x: boolean = true; // Bad: 'boolean' here does not aid readability
</code></pre>
<pre><code class="language-ts bad">// Bad: 'Set' is trivially inferred from the initialization
const x: Set&lt;string&gt; = new Set();
</code></pre>
<p>Explicitly specifying types may be required to prevent generic type parameters
from being inferred as <code>unknown</code>. For example, initializing generic types with
no values (e.g. empty arrays, objects, <code>Map</code>s, or <code>Set</code>s).</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">const x = new Set&lt;string&gt;();
</code></pre>
<p>For more complex expressions, type annotations can help with readability of the
program:</p>
<pre><code class="language-ts bad">// Hard to reason about the type of 'value' without an annotation.
const value = await rpc.getSomeValue().transform();
</code></pre>
<pre><code class="language-ts good">// Can tell the type of 'value' at a glance.
const value: string[] = await rpc.getSomeValue().transform();
</code></pre>
<p>Whether an annotation is required is decided by the code reviewer.</p>
2021-01-02 23:30:59 +08:00
<h4 id="return-types" class="numbered">Return types</h4>
2021-01-02 23:30:59 +08:00
<p>Whether to include return type annotations for functions and methods is up to
the code author. Reviewers <em>may</em> ask for annotations to clarify complex return
types that are hard to understand. Projects <em>may</em> have a local policy to always
require return types, but this is not a general TypeScript style requirement.</p>
<p>There are two benefits to explicitly typing out the implicit return values of
functions and methods:</p>
<ul>
<li>More precise documentation to benefit readers of the code.</li>
<li>Surface potential type errors faster in the future if there are code changes
that change the return type of the function.</li>
</ul>
<p><a id="null-vs-undefined"></a></p>
<h3 id="undefined-and-null" class="numbered">Undefined and null</h3>
2021-01-02 23:30:59 +08:00
<p>TypeScript supports <code>undefined</code> and <code>null</code> types. Nullable types can be
2021-01-02 23:30:59 +08:00
constructed as a union type (<code>string|null</code>); similarly with <code>undefined</code>. There
is no special syntax for unions of <code>undefined</code> and <code>null</code>.</p>
2021-01-02 23:30:59 +08:00
<p>TypeScript code can use either <code>undefined</code> or <code>null</code> to denote absence of a
value, there is no general guidance to prefer one over the other. Many
JavaScript APIs use <code>undefined</code> (e.g. <code>Map.get</code>), while many DOM and Google APIs
use <code>null</code> (e.g. <code>Element.getAttribute</code>), so the appropriate absent value
depends on the context.</p>
<h4 id="nullableundefined-type-aliases" class="numbered">Nullable/undefined type aliases</h4>
2021-01-02 23:30:59 +08:00
<p>Type aliases <em>must not</em> include <code>|null</code> or <code>|undefined</code> in a union type.
Nullable aliases typically indicate that null values are being passed around
through too many layers of an application, and this clouds the source of the
original issue that resulted in <code>null</code>. They also make it unclear when specific
values on a class or interface might be absent.</p>
<p>Instead, code <em>must</em> only add <code>|null</code> or <code>|undefined</code> when the alias is actually
used. Code <em>should</em> deal with null values close to where they arise, using the
above techniques.</p>
<pre><code class="language-ts bad">// Bad
type CoffeeResponse = Latte|Americano|undefined;
class CoffeeService {
getLatte(): CoffeeResponse { ... };
}
</code></pre>
<pre><code class="language-ts good">// Better
type CoffeeResponse = Latte|Americano;
class CoffeeService {
getLatte(): CoffeeResponse|undefined { ... };
}
</code></pre>
<p><a id="optionals-vs-undefined-type"></a></p>
2021-01-02 23:30:59 +08:00
<h4 id="prefer-optional-over-undefined" class="numbered">Prefer optional over <code>|undefined</code></h4>
2021-01-02 23:30:59 +08:00
<p>In addition, TypeScript supports a special construct for optional parameters and
fields, using <code>?</code>:</p>
<pre><code class="language-ts good">interface CoffeeOrder {
sugarCubes: number;
milk?: Whole|LowFat|HalfHalf;
}
function pourCoffee(volume?: Milliliter) { ... }
</code></pre>
<p>Optional parameters implicitly include <code>|undefined</code> in their type. However, they
are different in that they can be left out when constructing a value or calling
a method. For example, <code>{sugarCubes: 1}</code> is a valid <code>CoffeeOrder</code> because <code>milk</code>
is optional.</p>
<p>Use optional fields (on interfaces or classes) and parameters rather than a
<code>|undefined</code> type.</p>
<p>For classes preferably avoid this pattern altogether and initialize as many
fields as possible.</p>
<pre><code class="language-ts good">class MyClass {
field = '';
}
</code></pre>
<p><a id="structural-types-vs-nominal-types"></a></p>
<h3 id="use-structural-types" class="numbered">Use structural types</h3>
2021-01-02 23:30:59 +08:00
<p>TypeScript's type system is structural, not nominal. That is, a value matches a
type if it has at least all the properties the type requires and the properties'
types match, recursively.</p>
<p>When providing a structural-based implementation, explicitly include the type at
the declaration of the symbol (this allows more precise type checking and error
reporting).</p>
<pre><code class="language-ts good">const foo: Foo = {
a: 123,
b: 'abc',
}
</code></pre>
<pre><code class="language-ts bad">const badFoo = {
a: 123,
b: 'abc',
}
</code></pre>
<p>Use interfaces to define structural types, not classes</p>
<pre><code class="language-ts good">interface Foo {
a: number;
b: string;
}
const foo: Foo = {
a: 123,
b: 'abc',
}
</code></pre>
<pre><code class="language-ts bad">class Foo {
readonly a: number;
readonly b: number;
}
const foo: Foo = {
a: 123,
b: 'abc',
}
</code></pre>
<section class="zippy">
<p>Why?</p>
2021-01-02 23:30:59 +08:00
<p>The <q>badFoo</q> object above relies on type inference. Additional fields could be
added to <q>badFoo</q> and the type is inferred based on the object itself.</p>
<p>When passing a <q>badFoo</q> to a function that takes a <q>Foo</q>, the error will be at
the function call site, rather than at the object declaration site. This is also
useful when changing the surface of an interface across broad codebases.</p>
<pre><code class="language-ts">interface Animal {
sound: string;
name: string;
}
function makeSound(animal: Animal) {}
/**
* 'cat' has an inferred type of '{sound: string}'
*/
const cat = {
sound: 'meow',
};
/**
* 'cat' does not meet the type contract required for the function, so the
* TypeScript compiler errors here, which may be very far from where 'cat' is
* defined.
*/
makeSound(cat);
/**
* Horse has a structural type and the type error shows here rather than the
* function call. 'horse' does not meet the type contract of 'Animal'.
*/
const horse: Animal = {
sound: 'niegh',
};
const dog: Animal = {
sound: 'bark',
name: 'MrPickles',
};
makeSound(dog);
makeSound(horse);
</code></pre>
</section>
2021-01-02 23:30:59 +08:00
<p><a id="interfaces-vs-type-aliases"></a></p>
<h3 id="prefer-interfaces" class="numbered">Prefer interfaces over type literal aliases</h3>
2021-01-02 23:30:59 +08:00
<p>TypeScript supports
<a href="https://www.typescriptlang.org/docs/handbook/2/everyday-types.html#type-aliases">type aliases</a>
2021-01-02 23:30:59 +08:00
for naming a type expression. This can be used to name primitives, unions,
tuples, and any other types.</p>
<p>However, when declaring types for objects, use interfaces instead of a type
alias for the object literal expression.</p>
<pre><code class="language-ts good">interface User {
firstName: string;
lastName: string;
}
</code></pre>
<pre><code class="language-ts bad">type User = {
firstName: string,
lastName: string,
}
</code></pre>
<section class="zippy">
<p>Why?</p>
2021-01-02 23:30:59 +08:00
<p>These forms are nearly equivalent, so under the principle of just choosing one
out of two forms to prevent variation, we should choose one. Additionally, there
are also
2021-01-02 23:30:59 +08:00
<a href="https://ncjamieson.com/prefer-interfaces/">interesting technical reasons to prefer interface</a>.
That page quotes the TypeScript team lead: <q>Honestly, my take is that it should
really just be interfaces for anything that they can model. There is no benefit
to type aliases when there are so many issues around display/perf.</q></p>
</section>
2021-01-02 23:30:59 +08:00
<h3 id="arrayt-type" class="numbered"><code>Array&lt;T&gt;</code> Type</h3>
2021-01-02 23:30:59 +08:00
<p>For simple types (containing just alphanumeric characters and dot), use the
syntax sugar for arrays, <code>T[]</code> or <code>readonly T[]</code>, rather than the longer form
<code>Array&lt;T&gt;</code> or <code>ReadonlyArray&lt;T&gt;</code>.</p>
<p>For multi-dimensional non-<code>readonly</code> arrays of simple types, use the syntax
sugar form (<code>T[][]</code>, <code>T[][][]</code>, and so on) rather than the longer form.</p>
2021-01-02 23:30:59 +08:00
<p>For anything more complex, use the longer form <code>Array&lt;T&gt;</code>.</p>
<p>These rules apply at each level of nesting, i.e. a simple <code>T[]</code> nested in a more
complex type would still be spelled as <code>T[]</code>, using the syntax sugar.</p>
<pre><code class="language-ts good">let a: string[];
let b: readonly string[];
let c: ns.MyObj[];
let d: string[][];
let e: Array&lt;{n: number, s: string}&gt;;
let f: Array&lt;string|number&gt;;
let g: ReadonlyArray&lt;string|number&gt;;
let h: InjectionToken&lt;string[]&gt;; // Use syntax sugar for nested types.
let i: ReadonlyArray&lt;string[]&gt;;
let j: Array&lt;readonly string[]&gt;;
2021-01-02 23:30:59 +08:00
</code></pre>
<pre><code class="language-ts bad">let a: Array&lt;string&gt;; // The syntax sugar is shorter.
let b: ReadonlyArray&lt;string&gt;;
let c: Array&lt;ns.MyObj&gt;;
let d: Array&lt;string[]&gt;;
let e: {n: number, s: string}[]; // The braces make it harder to read.
let f: (string|number)[]; // Likewise with parens.
let g: readonly (string | number)[];
let h: InjectionToken&lt;Array&lt;string&gt;&gt;;
let i: readonly string[][];
let j: (readonly string[])[];
2021-01-02 23:30:59 +08:00
</code></pre>
<h3 id="indexable-key-string-type" class="numbered">Indexable types / index signatures (<code>{[key: string]: T}</code>)</h3>
2021-01-02 23:30:59 +08:00
<p>In JavaScript, it's common to use an object as an associative array (aka <q>map</q>,
<q>hash</q>, or <q>dict</q>). Such objects can be typed using an
<a href="https://www.typescriptlang.org/docs/handbook/2/objects.html#index-signatures">index signature</a>
(<code>[k: string]: T</code>) in TypeScript:</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts">const fileSizes: {[fileName: string]: number} = {};
fileSizes['readme.txt'] = 541;
</code></pre>
<p>In TypeScript, provide a meaningful label for the key. (The label only exists
for documentation; it's unused otherwise.)</p>
<pre><code class="language-ts bad">const users: {[key: string]: number} = ...;
</code></pre>
<pre><code class="language-ts good">const users: {[userName: string]: number} = ...;
</code></pre>
<blockquote>
<p>Rather than using one of these, consider using the ES6 <code>Map</code> and <code>Set</code> types
instead. JavaScript objects have
<a href="http://2ality.com/2012/01/objects-as-maps.html">surprising undesirable behaviors</a>
and the ES6 types more explicitly convey your intent. Also, <code>Map</code>s can be
keyed by—and <code>Set</code>s can contain—types other than <code>string</code>.</p>
</blockquote>
<p>TypeScript's builtin <code>Record&lt;Keys, ValueType&gt;</code> type allows constructing types
with a defined set of keys. This is distinct from associative arrays in that the
keys are statically known. See advice on that
<a href="#mapped-conditional-types">below</a>.</p>
<h3 id="mapped-conditional-types" class="numbered">Mapped and conditional types</h3>
2021-01-02 23:30:59 +08:00
<p>TypeScript's
<a href="https://www.typescriptlang.org/docs/handbook/2/mapped-types.html">mapped types</a>
2021-01-02 23:30:59 +08:00
and
<a href="https://www.typescriptlang.org/docs/handbook/2/conditional-types.html">conditional types</a>
2021-01-02 23:30:59 +08:00
allow specifying new types based on other types. TypeScript's standard library
includes several type operators based on these (<code>Record</code>, <code>Partial</code>, <code>Readonly</code>
etc).</p>
<p>These type system features allow succinctly specifying types and constructing
powerful yet type safe abstractions. They come with a number of drawbacks
though:</p>
<ul>
<li>Compared to explicitly specifying properties and type relations (e.g. using
interfaces and extension, see below for an example), type operators require
the reader to mentally evaluate the type expression. This can make programs
substantially harder to read, in particular combined with type inference and
expressions crossing file boundaries.</li>
<li>Mapped &amp; conditional types' evaluation model, in particular when combined
with type inference, is underspecified, not always well understood, and
often subject to change in TypeScript compiler versions. Code can
<q>accidentally</q> compile or seem to give the right results. This increases
future support cost of code using type operators.</li>
<li>Mapped &amp; conditional types are most powerful when deriving types from
complex and/or inferred types. On the flip side, this is also when they are
most prone to create hard to understand and maintain programs.</li>
<li>Some language tooling does not work well with these type system features.
E.g. your IDE's find references (and thus rename property refactoring) will
not find properties in a <code>Pick&lt;T, Keys&gt;</code> type, and Code Search won't
hyperlink them.</li>
<li>
</ul>
<p>The style recommendation is:</p>
<ul>
<li>Always use the simplest type construct that can possibly express your code.</li>
<li>A little bit of repetition or verbosity is often much cheaper than the long
term cost of complex type expressions.</li>
<li>Mapped &amp; conditional types may be used, subject to these considerations.</li>
</ul>
<p>For example, TypeScript's builtin <code>Pick&lt;T, Keys&gt;</code> type allows creating a new
2021-01-02 23:30:59 +08:00
type by subsetting another type <code>T</code>, but simple interface extension can often be
easier to understand.</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts">interface User {
shoeSize: number;
favoriteIcecream: string;
favoriteChocolate: string;
}
// FoodPreferences has favoriteIcecream and favoriteChocolate, but not shoeSize.
type FoodPreferences = Pick&lt;User, 'favoriteIcecream'|'favoriteChocolate'&gt;;
</code></pre>
<p>This is equivalent to spelling out the properties on <code>FoodPreferences</code>:</p>
<pre><code class="language-ts">interface FoodPreferences {
favoriteIcecream: string;
favoriteChocolate: string;
}
</code></pre>
<p>To reduce duplication, <code>User</code> could extend <code>FoodPreferences</code>, or (possibly
better) nest a field for food preferences:</p>
<pre><code class="language-ts good">interface FoodPreferences { /* as above */ }
interface User extends FoodPreferences {
shoeSize: number;
// also includes the preferences.
}
</code></pre>
<p>Using interfaces here makes the grouping of properties explicit, improves IDE
support, allows better optimization, and arguably makes the code easier to
understand.</p>
2021-01-02 23:30:59 +08:00
<h3 id="any" class="numbered"><code>any</code> Type</h3>
2021-01-02 23:30:59 +08:00
<p>TypeScript's <code>any</code> type is a super and subtype of all other types, and allows
dereferencing all properties. As such, <code>any</code> is dangerous - it can mask severe
programming errors, and its use undermines the value of having static types in
the first place.</p>
<section>
<p><strong>Consider <em>not</em> to use <code>any</code>.</strong> In circumstances where you want to use <code>any</code>,
consider one of:</p>
</section>
2021-01-02 23:30:59 +08:00
<ul>
<li><a href="#any-specific">Provide a more specific type</a></li>
<li><a href="#any-unknown">Use <code>unknown</code></a></li>
<li><a href="#any-suppress">Suppress the lint warning and document why</a></li>
</ul>
<h4 id="any-specific" class="numbered">Providing a more specific type</h4>
2021-01-02 23:30:59 +08:00
<p>Use interfaces , an
inline object type, or a type alias:</p>
2021-01-02 23:30:59 +08:00
<pre><code class="language-ts good">// Use declared interfaces to represent server-side JSON.
declare interface MyUserJson {
name: string;
email: string;
}
// Use type aliases for types that are repetitive to write.
type MyType = number|string;
// Or use inline object types for complex returns.
function getTwoThings(): {something: number, other: string} {
// ...
return {something, other};
}
// Use a generic type, where otherwise a library would say `any` to represent
// they don't care what type the user is operating on (but note "Return type
// only generics" below).
function nicestElement&lt;T&gt;(items: T[]): T {
// Find the nicest element in items.
// Code can also put constraints on T, e.g. &lt;T extends HTMLElement&gt;.
}
</code></pre>
<h4 id="any-unknown" class="numbered">Using <code>unknown</code> over <code>any</code></h4>
2021-01-02 23:30:59 +08:00
<p>The <code>any</code> type allows assignment into any other type and dereferencing any
property off it. Often this behaviour is not necessary or desirable, and code
just needs to express that a type is unknown. Use the built-in type <code>unknown</code> in
that situation — it expresses the concept and is much safer as it does not allow
dereferencing arbitrary properties.</p>
<pre><code class="language-ts good">// Can assign any value (including null or undefined) into this but cannot
// use it without narrowing the type or casting.
const val: unknown = value;
</code></pre>
<pre><code class="language-ts bad">const danger: any = value /* result of an arbitrary expression */;
danger.whoops(); // This access is completely unchecked!
</code></pre>
<section>
<p>To safely use <code>unknown</code> values, narrow the type using a
<a href="https://www.typescriptlang.org/docs/handbook/advanced-types.html#type-guards-and-differentiating-types">type guard</a></p>
2021-01-02 23:30:59 +08:00
</section>
<h4 id="any-suppress" class="numbered">Suppressing <code>any</code> lint warnings</h4>
2021-01-02 23:30:59 +08:00
<p>Sometimes using <code>any</code> is legitimate, for example in tests to construct a mock
object. In such cases, add a comment that suppresses the lint warning, and
document why it is legitimate.</p>
<pre><code class="language-ts good">// This test only needs a partial implementation of BookService, and if
// we overlooked something the test will fail in an obvious way.
// This is an intentionally unsafe partial mock
// tslint:disable-next-line:no-any
const mockBookService = ({get() { return mockBook; }} as any) as BookService;
// Shopping cart is not used in this test
// tslint:disable-next-line:no-any
const component = new MyComponent(mockBookService, /* unused ShoppingCart */ null as any);
</code></pre>
<h3 id="empty-interface-type" class="numbered"><code>{}</code> Type</h3>
<p>The <code>{}</code> type, also known as an <em>empty interface</em> type, represents a interface
with no properties. An empty interface type has no specified properties and
therefore any non-nullish value is assignable to it.</p>
<pre><code class="language-ts bad">let player: {};
player = {
health: 50,
}; // Allowed.
console.log(player.health) // Property 'health' does not exist on type '{}'.
</code></pre>
<pre><code class="language-ts bad">function takeAnything(obj:{}) {
}
takeAnything({});
takeAnything({ a: 1, b: 2 });
</code></pre>
<p>Google3 code <strong>should not</strong> use <code>{}</code> for most use cases. <code>{}</code> represents any
non-nullish primitive or object type, which is rarely appropriate. Prefer one of
the following more-descriptive types:</p>
<ul>
<li><code>unknown</code> can hold any value, including <code>null</code> or <code>undefined</code>, and is
generally more appropriate for opaque values.</li>
<li><code>Record&lt;string, T&gt;</code> is better for dictionary-like objects, and provides
better type safety by being explicit about the type <code>T</code> of contained values
(which may itself be <code>unknown</code>).</li>
<li><code>object</code> excludes primitives as well, leaving only non-nullish functions and
objects, but without any other assumptions about what properties may be
available.</li>
</ul>
<h3 id="tuple-types" class="numbered">Tuple types</h3>
2021-01-02 23:30:59 +08:00
<p>If you are tempted to create a Pair type, instead use a tuple type:</p>
<pre><code class="language-ts bad">interface Pair {
first: string;
second: string;
}
function splitInHalf(input: string): Pair {
...
return {first: x, second: y};
}
</code></pre>
<pre><code class="language-ts good">function splitInHalf(input: string): [string, string] {
...
return [x, y];
}
// Use it like:
const [leftHalf, rightHalf] = splitInHalf('my string');
</code></pre>
<p>However, often it's clearer to provide meaningful names for the properties.</p>
<p>If declaring an <code>interface</code> is too heavyweight, you can use an inline object
literal type:</p>
<pre><code class="language-ts good">function splitHostPort(address: string): {host: string, port: number} {
...
}
// Use it like:
const address = splitHostPort(userAddress);
use(address.port);
// You can also get tuple-like behavior using destructuring:
const {host, port} = splitHostPort(userAddress);
</code></pre>
<h3 id="wrapper-types" class="numbered">Wrapper types</h3>
2021-01-02 23:30:59 +08:00
<p>There are a few types related to JavaScript primitives that <em>should not</em> ever be
2021-01-02 23:30:59 +08:00
used:</p>
<ul>
<li><code>String</code>, <code>Boolean</code>, and <code>Number</code> have slightly different meaning from the
corresponding primitive types <code>string</code>, <code>boolean</code>, and <code>number</code>. Always use
the lowercase version.</li>
<li><code>Object</code> has similarities to both <code>{}</code> and <code>object</code>, but is slightly looser.
Use <code>{}</code> for a type that include everything except <code>null</code> and <code>undefined</code>,
or lowercase <code>object</code> to further exclude the other primitive types (the
three mentioned above, plus <code>symbol</code> and <code>bigint</code>).</li>
</ul>
<p>Further, never invoke the wrapper types as constructors (with <code>new</code>).</p>
<h3 id="return-type-only-generics" class="numbered">Return type only generics</h3>
2021-01-02 23:30:59 +08:00
<p>Avoid creating APIs that have return type only generics. When working with
existing APIs that have return type only generics always explicitly specify the
generics.</p>
<p><a id="appendices-style-related-tools"></a></p>
<h2 id="toolchain-requirements" class="numbered">Toolchain requirements</h2>
<p>Google style requires using a number of tools in specific ways, outlined here.</p>
<p><a id="clang-format"></a>
<a id="google-format"></a>
<a id="use-the-google3-formatter"></a>
<a id="source-code-formatting"></a>
<a id="appendices-clang-format"></a>
<a id="formatting"></a>
<a id="formatting-array-literals"></a>
<a id="formatting-block-indentation"></a>
<a id="formatting-column-limit"></a>
<a id="formatting-empty-blocks"></a>
<a id="formatting-function-expressions"></a>
<a id="formatting-nonempty-blocks"></a>
<a id="formatting-object-literals"></a>
<a id="formatting-one-statement-perline"></a>
<a id="formatting-statements"></a>
<a id="formatting-switch-statements"></a>
<a id="formatting-vertical-whitespace"></a>
<a id="formatting-whitespace"></a></p>
<h3 id="typescript-compiler" class="numbered">TypeScript compiler</h3>
<p>All TypeScript files must pass type checking using the standard
tool chain.</p>
<h4 id="ts-ignore" class="numbered">@ts-ignore</h4>
<p>Do not use <code>@ts-ignore</code> nor the variants <code>@ts-expect-error</code> or <code>@ts-nocheck</code>.</p>
<section class="zippy">
<p>Why?</p>
<p>They superficially seem to be an easy way to <q>fix</q> a compiler error, but in
practice, a specific compiler error is often caused by a larger problem that can
be fixed more directly.</p>
<p>For example, if you are using <code>@ts-ignore</code> to suppress a type error, then it's
hard to predict what types the surrounding code will end up seeing. For many
type errors, the advice in <a href="#any">how to best use <code>any</code></a> is useful.</p>
</section>
<p>You may use <code>@ts-expect-error</code> in unit tests, though you generally <em>should not</em>.
<code>@ts-expect-error</code> suppresses all errors. It's easy to accidentally over-match
and suppress more serious errors. Consider one of:</p>
<ul>
<li>When testing APIs that need to deal with unchecked values at runtime, add
casts to the expected type or to <code>any</code> and add an explanatory comment. This
limits error suppression to a single expression.</li>
<li>Suppress the lint warning and document why, similar to
<a href="#any-suppress">suppressing <code>any</code> lint warnings</a>.</li>
</ul>
<h3 id="conformance" class="numbered">Conformance</h3>
<p>Google TypeScript includes several <em>conformance frameworks</em>,
<a href="https://tsetse.info">tsetse</a> and
<a href="https://github.com/google/tsec">tsec</a>.</p>
<p>These rules are commonly used to enforce critical restrictions (such as defining
globals, which could break the codebase) and security patterns (such as using
<code>eval</code> or assigning to <code>innerHTML</code>), or more loosely to improve code quality.</p>
<p>Google-style TypeScript must abide by any applicable global or framework-local
conformance rules.</p>
<p><a id="jsdoc"></a>
<a id="formatting-comments"></a></p>
<h2 id="comments-documentation" class="numbered">Comments and documentation</h2>
<h4 id="jsdoc-vs-comments" class="numbered">JSDoc versus comments</h4>
<p>There are two types of comments, JSDoc (<code>/** ... */</code>) and non-JSDoc ordinary
comments (<code>// ...</code> or <code>/* ... */</code>).</p>
<ul>
<li>Use <code>/** JSDoc */</code> comments for documentation, i.e. comments a user of the
code should read.</li>
<li>Use <code>// line comments</code> for implementation comments, i.e. comments that only
concern the implementation of the code itself.</li>
</ul>
<p>JSDoc comments are understood by tools (such as editors and documentation
generators), while ordinary comments are only for other humans.</p>
<p><a id="formatting-block-comment-style"></a></p>
<h4 id="multi-line-comments" class="numbered">Multi-line comments</h4>
<p>Multi-line comments are indented at the same level as the surrounding code. They
<em>must</em> use multiple single-line comments (<code>//</code>-style), not block comment style
(<code>/* */</code>).</p>
<pre><code class="language-ts good">// This is
// fine
</code></pre>
<pre><code class="language-ts bad">/*
* This should
* use multiple
* single-line comments
*/
/* This should use // */
</code></pre>
<p>Comments are not enclosed in boxes drawn with asterisks or other characters.</p>
<h3 id="jsdoc-general-form" class="numbered">JSDoc general form</h3>
<p>The basic formatting of JSDoc comments is as seen in this example:</p>
<pre><code class="language-ts good">/**
* Multiple lines of JSDoc text are written here,
* wrapped normally.
* @param arg A number to do something to.
*/
function doSomething(arg: number) { … }
</code></pre>
<p>or in this single-line example:</p>
<pre><code class="language-ts good">/** This short jsdoc describes the function. */
function doSomething(arg: number) { … }
</code></pre>
<p>If a single-line comment overflows into multiple lines, it <em>must</em> use the
multi-line style with <code>/**</code> and <code>*/</code> on their own lines.</p>
<p>Many tools extract metadata from JSDoc comments to perform code validation and
optimization. As such, these comments <em>must</em> be well-formed.</p>
<h3 id="jsdoc-markdown" class="numbered">Markdown</h3>
<p>JSDoc is written in Markdown, though it <em>may</em> include HTML when necessary.</p>
<p>This means that tooling parsing JSDoc will ignore plain text formatting, so if
you did this:</p>
<pre><code class="language-ts bad">/**
* Computes weight based on three factors:
* items sent
* items received
* last timestamp
*/
</code></pre>
<p>it will be rendered like this:</p>
<pre><code>Computes weight based on three factors: items sent items received last timestamp
</code></pre>
<p>Instead, write a Markdown list:</p>
<pre><code class="language-ts good">/**
* Computes weight based on three factors:
*
* - items sent
* - items received
* - last timestamp
*/
</code></pre>
<h3 id="jsdoc-tags" class="numbered">JSDoc tags</h3>
<p>Google style allows a subset of JSDoc tags. Most tags must occupy their own line, with the tag at the beginning
of the line.</p>
<pre><code class="language-ts good">/**
* The "param" tag must occupy its own line and may not be combined.
* @param left A description of the left param.
* @param right A description of the right param.
*/
function add(left: number, right: number) { ... }
</code></pre>
<pre><code class="language-ts bad">/**
* The "param" tag must occupy its own line and may not be combined.
* @param left @param right
*/
function add(left: number, right: number) { ... }
</code></pre>
<h3 id="jsdoc-line-wrapping" class="numbered">Line wrapping</h3>
<p>Line-wrapped block tags are indented four spaces. Wrapped description text <em>may</em>
be lined up with the description on previous lines, but this horizontal
alignment is discouraged.</p>
<pre><code class="language-ts good">/**
* Illustrates line wrapping for long param/return descriptions.
* @param foo This is a param with a particularly long description that just
* doesn't fit on one line.
* @return This returns something that has a lengthy description too long to fit
* in one line.
*/
exports.method = function(foo) {
return 5;
};
</code></pre>
<p>Do not indent when wrapping a <code>@desc</code> or <code>@fileoverview</code> description.</p>
<h3 id="document-all-top-level-exports-of-modules" class="numbered">Document all top-level exports of modules</h3>
<p>Use <code>/** JSDoc */</code> comments to communicate information to the users of your
code. Avoid merely restating the property or parameter name. You <em>should</em> also
document all properties and methods (exported/public or not) whose purpose is
not immediately obvious from their name, as judged by your reviewer.</p>
<p><strong>Exception:</strong> Symbols that are only exported to be consumed by tooling, such as
@NgModule classes, do not require comments.</p>
<h3 id="jsdoc-class-comments" class="numbered">Class comments</h3>
<p>JSDoc comments for classes should provide the reader with enough information to
know how and when to use the class, as well as any additional considerations
necessary to correctly use the class. Textual descriptions may be omitted on the
constructor.</p>
<h3 id="method-and-function-comments" class="numbered">Method and function comments</h3>
<p>Method, parameter, and return descriptions may be omitted if they are obvious
from the rest of the methods JSDoc or from the method name and type signature.</p>
<p>Method descriptions begin with a verb phrase that describes what the method
does. This phrase is not an imperative sentence, but instead is written in the
third person, as if there is an implied <q>This method ...</q> before it.</p>
<h3 id="parameter-property-comments" class="numbered">Parameter property comments</h3>
<p>A
<a href="https://www.typescriptlang.org/docs/handbook/2/classes.html#parameter-properties">parameter property</a>
is a constructor parameter that is prefixed by one of the modifiers <code>private</code>,
<code>protected</code>, <code>public</code>, or <code>readonly</code>. A parameter property declares both a
parameter and an instance property, and implicitly assigns into it. For example,
<code>constructor(private readonly foo: Foo)</code>, declares that the constructor takes a
parameter <code>foo</code>, but also declares a private readonly property <code>foo</code>, and
assigns the parameter into that property before executing the remainder of the
constructor.</p>
<p>To document these fields, use JSDoc's <code>@param</code> annotation. Editors display the
description on constructor calls and property accesses.</p>
<pre><code class="language-ts good">/** This class demonstrates how parameter properties are documented. */
class ParamProps {
/**
* @param percolator The percolator used for brewing.
* @param beans The beans to brew.
*/
constructor(
private readonly percolator: Percolator,
private readonly beans: CoffeeBean[]) {}
}
</code></pre>
<pre><code class="language-ts good">/** This class demonstrates how ordinary fields are documented. */
class OrdinaryClass {
/** The bean that will be used in the next call to brew(). */
nextBean: CoffeeBean;
constructor(initialBean: CoffeeBean) {
this.nextBean = initialBean;
}
}
</code></pre>
<p><a id="omit-comments-that-are-redundant-with-typescript"></a>
<a id="do-not-use-override"></a></p>
<h3 id="jsdoc-type-annotations" class="numbered">JSDoc type annotations</h3>
<p>JSDoc type annotations are redundant in TypeScript source code. Do not declare
types in <code>@param</code> or <code>@return</code> blocks, do not write <code>@implements</code>, <code>@enum</code>,
<code>@private</code>, <code>@override</code> etc. on code that uses the <code>implements</code>, <code>enum</code>,
<code>private</code>, <code>override</code> etc. keywords.</p>
<h3 id="redundant-comments" class="numbered">Make comments that actually add information</h3>
<p>For non-exported symbols, sometimes the name and type of the function or
parameter is enough. Code will <em>usually</em> benefit from more documentation than
just variable names though!</p>
<ul>
<li><p>Avoid comments that just restate the parameter name and type, e.g.</p>
<pre><code class="language-ts bad">/** @param fooBarService The Bar service for the Foo application. */
</code></pre></li>
<li><p>Because of this rule, <code>@param</code> and <code>@return</code> lines are only required when
they add information, and <em>may</em> otherwise be omitted.</p>
<pre><code class="language-ts good">/**
* POSTs the request to start coffee brewing.
* @param amountLitres The amount to brew. Must fit the pot size!
*/
brew(amountLitres: number, logger: Logger) {
// ...
}
</code></pre></li>
</ul>
<p><a id="formatting-param-name-comments"></a></p>
<h4 id="comments-when-calling-a-function" class="numbered">Comments when calling a function</h4>
<p>“Parameter name” comments should be used whenever the method name and parameter
value do not sufficiently convey the meaning of the parameter.</p>
<p>Before adding these comments, consider refactoring the method to instead accept
an interface and destructure it to greatly improve call-site
readability.</p>
<p><q>Parameter name</q> comments go before the parameter value, and include the
parameter name and a <code>=</code> suffix:</p>
<pre><code class="language-ts good">someFunction(obviousParam, /* shouldRender= */ true, /* name= */ 'hello');
</code></pre>
<p>Existing code may use a legacy parameter name comment style, which places these
comments ~after~ the parameter value and omits the <code>=</code>. Continuing to use this
style within the file for consistency is acceptable.</p>
<pre><code class="language-ts">someFunction(obviousParam, true /* shouldRender */, 'hello' /* name */);
</code></pre>
<h3 id="place-documentation-prior-to-decorators" class="numbered">Place documentation prior to decorators</h3>
<p>When a class, method, or property have both decorators like <code>@Component</code> and
JsDoc, please make sure to write the JsDoc before the decorator.</p>
<ul>
<li><p>Do not write JsDoc between the Decorator and the decorated statement.</p>
<pre><code class="language-ts bad">@Component({
selector: 'foo',
template: 'bar',
})
/** Component that prints "bar". */
export class FooComponent {}
</code></pre></li>
<li><p>Write the JsDoc block before the Decorator.</p>
<pre><code class="language-ts good">/** Component that prints "bar". */
@Component({
selector: 'foo',
template: 'bar',
})
export class FooComponent {}
</code></pre></li>
</ul>
<h2 id="policies" class="numbered">Policies</h2>
<p><a id="policies-be-consistent"></a>
<a id="policies-newly-added-code-use-google-style"></a></p>
<h3 id="consistency" class="numbered">Consistency</h3>
2021-01-02 23:30:59 +08:00
<p>For any style question that isn't settled definitively by this specification, do
what the other code in the same file is already doing (<q>be consistent</q>). If that
doesn't resolve the question, consider emulating the other files in the same
directory.</p>
<p>Brand new files <em>must</em> use Google Style, regardless of the style choices of
other files in the same package. When adding new code to a file that is not in
Google Style, reformatting the existing code first is recommended, subject to
the advice <a href="#reformatting-existing-code">below</a>. If this reformatting is not
done, then new code <em>should</em> be as consistent as possible with existing code in
the same file, but <em>must not</em> violate the style guide.</p>
<p><a id="policies-code-not-in-google-style"></a>
<a id="policies-reformatting-existing-code"></a></p>
<h4 id="reformatting-existing-code" class="numbered">Reformatting existing code</h4>
<p>You will occasionally encounter files in the codebase that are not in proper
Google Style. These may have come from an acquisition, or may have been written
before Google Style took a position on some issue, or may be in non-Google Style
for any other reason.</p>
<p>When updating the style of existing code, follow these guidelines.</p>
<ol>
<li>It is not required to change all existing code to meet current style
guidelines. Reformatting existing code is a trade-off between code churn and
consistency. Style rules evolve over time and these kinds of tweaks to
maintain compliance would create unnecessary churn. However, if significant
changes are being made to a file it is expected that the file will be in
Google Style.</li>
<li>Be careful not to allow opportunistic style fixes to muddle the focus of a
CL. If you find yourself making a lot of style changes that arent critical
to the central focus of a CL, promote those changes to a separate CL.</li>
</ol>
<p><a id="policies-deprecation"></a></p>
<h3 id="deprecation" class="numbered">Deprecation</h3>
<p>Mark deprecated methods, classes or interfaces with an <code>@deprecated</code> JSDoc
annotation. A deprecation comment must include simple, clear directions for
people to fix their call sites.</p>
<p><a id="policies-generated-code-mostly-exempt"></a></p>
<h3 id="generated-code" class="numbered">Generated code: mostly exempt</h3>
<p>Source code generated by the build process is not required to be in Google
Style. However, any generated identifiers that will be referenced from
hand-written source code must follow the naming requirements. As a special
exception, such identifiers are allowed to contain underscores, which may help
to avoid conflicts with hand-written identifiers.</p>
<p><a id="goals"></a></p>
<h4 id="style-guide-goals" class="numbered">Style guide goals</h4>
2021-01-02 23:30:59 +08:00
<p>In general, engineers usually know best about what's needed in their code, so if
there are multiple options and the choice is situation dependent, we should let
decisions be made locally. So the default answer should be <q>leave it out</q>.</p>
<p>The following points are the exceptions, which are the reasons we have some
global rules. Evaluate your style guide proposal against the following:</p>
<ol>
<li><p><strong>Code should avoid patterns that are known to cause problems, especially
for users new to the language.</strong></p>
</li>
2021-01-02 23:30:59 +08:00
<li><p><strong>Code across
projects should be consistent across
irrelevant variations.</strong></p>
<p>When there are two options that are equivalent in a superficial way, we
should consider choosing one just so we don't divergently evolve for no
reason and avoid pointless debates in code reviews.</p>
<p>Examples:</p>
<ul>
<li>The capitalization style of names.</li>
<li><code>x as T</code> syntax vs the equivalent <code>&lt;T&gt;x</code> syntax (disallowed).</li>
<li><code>Array&lt;[number, number]&gt;</code> vs <code>[number, number][]</code>.</li>
</ul></li>
<li><p><strong>Code should be maintainable in the long term.</strong></p>
<p>Code usually lives longer than the original author works on it, and the
TypeScript team must keep all of Google working into the future.</p>
<p>Examples:</p>
<ul>
<li>We use software to automate changes to code, so code is autoformatted so
it's easy for software to meet whitespace rules.</li>
<li>We require a single set of compiler flags, so a given TS library can be
written assuming a specific set of flags, and users can always safely
use a shared library.</li>
2021-01-02 23:30:59 +08:00
<li>Code must import the libraries it uses (<q>strict deps</q>) so that a
refactor in a dependency doesn't change the dependencies of its users.</li>
<li>We ask users to write tests. Without tests we cannot have confidence
that changes that we make to the language, don't break users.</li>
2021-01-02 23:30:59 +08:00
</ul></li>
<li><p><strong>Code reviewers should be focused on improving the quality of the code, not
enforcing arbitrary rules.</strong></p>
<p>If it's possible to implement your rule as an
automated check that is often a good sign.
This also supports principle 3.</p>
<p>If it really just doesn't matter that much -- if it's an obscure corner of
the language or if it avoids a bug that is unlikely to occur -- it's
probably worth leaving out.</p></li>
</ol>
<p><a id="appendices-legacy-exceptions"></a>
<a id="appendices-legacy-exceptions-declareLegacyNamespace"></a>
<a id="appendices-legacy-exceptions-forward-declare"></a>
<a id="appendices-legacy-exceptions-function"></a>
<a id="appendices-legacy-exceptions-goog-provide"></a>
<a id="appendices-legacy-exceptions-goog-provide-summary"></a>
<a id="appendices-legacy-exceptions-goog-scope"></a>
<a id="appendices-legacy-exceptions-module-get"></a>
<a id="appendices-legacy-exceptions-overview"></a>
<a id="appendices-legacy-exceptions-var"></a>
<a id="appendices-legacy-exceptions-var-const"></a>
<a id="appendices-legacy-exceptions-var-declare"></a>
<a id="appendices-legacy-exceptions-var-scope"></a>
<a id="features-classes-old-style"></a></p>
<div class="footnotes">
<hr>
<ol>
<li id="fn1">
<p>Namespace imports are often called 'module imports' <a href="#fnref1" rev="footnote"></a></p>
</li>
<li id="fn2">
<p>named imports are sometimes called 'destructuring
imports' because they use similar syntax to
destructuring assignments. <a href="#fnref2" rev="footnote"></a></p>
</li>
</ol>
</div>
2021-01-02 23:30:59 +08:00
</div>
</body>
</html>