Using this package? Please consider donating to support my open source work ❤️
Help @kitajs/html grow! Star and share this amazing repository with your friends and co-workers!
@kitajs/html
is a super fast JSX runtime to generate HTML strings that works everywhere.
Express? Fastify? Hono? Bun? Htmx?
If your code works with strings, we got you covered.
- Preview
- Installing
- Getting Started
- Sanitization
- Editor Intellisense and CLI tool
- Async Components
- Migrating from HTML
- Compiling HTML
- Fragments
- Supported HTML
- Extending types
- Performance
- How it works
- Format HTML output
- Fork credits
To use the @kitajs/html
package, follow these steps:
-
Install the required npm packages,
@kitajs/html
and@kitajs/ts-html-plugin
, to enable editor intellisense. Open your terminal and run:npm install @kitajs/html @kitajs/ts-html-plugin
-
Configure your TypeScript project to transpile TSX/JSX into JavaScript and using our LSP Plugin. Update your
tsconfig.json
file with the following settings:// tsconfig.json { "compilerOptions": { "jsx": "react", "jsxFactory": "Html.createElement", "jsxFragmentFactory": "Html.Fragment", "plugins": [{ "name": "@kitajs/ts-html-plugin" }] } }
-
Append the
xss-scan
command into your test script. This CLI comes from @kitajs/ts-html-plugin, which catches XSS vulnerabilities that may be introduced into your codebase, automating the xss scanning process to run everytime you test your code, like inside your CI/CD environment. Open yourpackage.json
file and add the following script:// package.json { "scripts": { "test": "xss-scan" } }
-
Ensure that your code editor is using the TypeScript version from your project's
node_modules
instead of the globally installed TypeScript. For Visual Studio Code, you can configure this in your workspace settings:// .vscode/settings.json { "typescript.tsdk": "node_modules/typescript/lib" }
Caution
Try writing console.log(<div>{String.name}</div>);
in your editor. If it THROWS a
XSS
error, then your setup is correct. Refer to the
@kitajs/ts-html-plugin repository for more
details on setting up editor intellisense. (It should throw, as String.name
has a
type of string
, type which may or may not have special caracters)
After successfully installing and configuring your project, you can start using JSX syntax
to generate HTML. Here are two options for importing the @kitajs/html
package:
-
Import it manually: Import the package in your TypeScript files where you need it, avoiding global scope pollution.
// my-file.tsx console.log(<div>Html import needs to be in scope!</div>);
-
Use the register to add a global namespace: Import the
register
to globally register all necessary functions for convenience.// Import the register to globally register all needed functions import '@kitajs/html/register'; // another-file.tsx console.log(<div>It works without importing!</div>);
Now you can use JSX to generate HTML throughout your project. Always use the safe
attribute or manually call Html.escapeHtml
to protect against XSS vulnerabilities when
rendering user input.
Ensuring XSS prevention is vital to guarantee your application's security. You can employ
the @kitajs/ts-html-plugin
to catch XSS
issues in your code editor and enhance your code's security.
Important
Please utilize our @kitajs/ts-html-plugin
to emit TypeScript errors wherever you are
exposed to XSS. Refer to Getting Started for installation
instructions.
This package sanitizes every attribute by default. However, since the resulting element is
always a string, it's impossible to differentiate an HTML element created by a <tag>
or
from user input. This necessitates the use of the provided safe
or manual invocation of Html.escapeHtml
. Or you can also use the Html.escape
or its
alias e
template function.
<div>⚠️ This will NOT be escaped and WILL expose you to XSS</div>
<div attr="This WILL be escaped"></div>
<div safe>This WILL be escaped</div>
<div>{Html.escapeHtml('This WILL be escaped')}</div>
or using the Html.escape
or its alias e
template function:
import { e } from '@kitajs/html';
<div>⚠️ This will NOT be escaped and WILL expose you to XSS</div>
<div attr="This WILL be escaped"></div>
<div safe>This WILL be escaped</div>
<div>{Html.escape`This WILL be escaped`}</div>
<div>{e`This WILL be escaped`}</div>
Here's an example of how this is DANGEROUS for your application:
user = {
name: 'Bad guy',
description: '</div><script>getStoredPasswordAndSentToBadGuysServer()</script>'
}
<div class="user-card">{user.description}</div>
// Renders this HTML, which will execute malicious code:
<div class="user-card">
<script>getStoredPasswordAndSentToBadGuysServer()</script>
</div>
<div class="user-card" safe>{user.description}</div>
// Renders this safe HTML, which will NOT execute any malicious code:
<div class="user-card">
</div><script>getStoredPasswordAndSentToBadGuysServer()</script>
</div>
Always use the safe
attribute when rendering uncontrolled user input. This will sanitize
the contents and prevent XSS attacks.
function UserCard({ name, description, date, about }) {
return (
<div class="card">
<h1 safe>{name}</h1>
<br />
<p safe>{description}</p>
<br />
// Controlled input, no need to sanitize
<time datetime={date.toISOString()}>{date.toDateString()}</time>
<br />
<p safe>{about}</p>
</div>
);
}
Note that you should only use the safe
attribute at the very bottom of the HTML tree
where it's needed.
Note: This section has been relocated to the @kitajs/ts-html-plugin repository.
Please consult their "Getting Started" section for instructions on enabling editor IntelliSense and using the CLI tool.
Async components are supported. When any child or sub child of a component tree is a promise, the whole tree will return a promise of the html string.
If no async components are found, the result will be simply a string, and you can safely cast it into a string.
async function Async() {
await callApi();
return <div>Async!</div>;
}
function Sync() {
return <div>Sync!</div>;
}
const async = (
<div>
<Async />
</div>
);
async instanceof Promise;
const sync: string = (
<div>
<Sync />
</div>
);
typeof sync === 'string';
A JSX.Element
will always be a string. Once a children element is a async component, the
entire upper tree will also be async.
Learn when JSX.Element is a Promise.
The only problem when rendering templates is that you must wait for the whole template to be rendered before sending it to the client. This is not a problem for small templates, but it can be a problem for large templates.
To solve this problem, we provide a Suspense
component that combined with
renderToStream()
rendering method, will stream a fallback component while it waits for
his children to be rendered. This is a perfect combo to use with
async components.
import { Suspense, renderToStream } from '@kitajs/html/suspense';
async function MyAsyncComponent() {
const data = await database.query();
return <User name={data.username} />;
}
function renderUserPage(rid: number) {
return (
<Suspense
rid={rid}
fallback={<div>Loading username...</div>}
catch={(err) => <div>Error: {err.stack}</div>}
>
<MyAsyncComponent />
</Suspense>
);
}
// Html is a string readable stream that can be piped to the client
const html = renderToStream(renderUserPage);
Note
The renderToStream()
is returns a native node/bun stream, head over to our
suspense-server example to see how to use it with
node:http, Express or Fastify servers.
The above example would render <div>Loading username...</div>
while waiting for the
MyAsyncComponent
to be rendered.
When using Suspense
, you cannot just call the component and get the html string, you
need to use the renderToStream
function to get a stream that can be piped to the client
with updates. Otherwise, the fallback would render forever.
As the result of any JSX component is always a string, you must use the rid
provided by
renderToStream
into all your suspense components, this way we can identify which
suspense is for which request and be able to render concurrent requests.
Suspense also accepts async fallbacks, but it blocks rendering until the fallback is resolved.
function renderTemplate(rid: number) {
return (
<Suspense
rid={rid}
fallback={<MyAsyncFallback />}
catch={(err) => <div>Error: {err.stack}</div>}
>
<MyAsyncComponent />
</Suspense>
);
}
The above example would only return anything after MyAsyncFallback
is resolved. To catch
async fallback errors, you must wrap it into a ErrorBoundary
.
The same way as promises must be awaited to resolve its own html, errors must be caught. Outside of suspense components, you can use the provided error boundaries to catch errors.
import { ErrorBoundary } from '@kitajs/html/error-boundary';
async function MyAsyncComponent() {
const data = await database.query(); // this promise may reject
return <User name={data.username} />;
}
function renderTemplate() {
return (
<ErrorBoundary catch={(err) => <div>Error: {err.stack}</div>}>
<MyAsyncComponent />
</ErrorBoundary>
);
}
// If MyAsyncComponent throws an error, it will render <div>Error: ...</div>
const html = await renderTemplate();
Error boundaries will only work for errors thrown inside async components, for sync components you must use try/catch.
function MySyncComponent() {
try {
const data = database.query(); // this may throw an error
return <User name={data.username} />;
} catch (err) {
return <div>Error: {err.stack}</div>;
}
}
Error boundaries outside suspense components will only catch errors thrown by the fallback
component. You must use the Suspense's catch
property to handle errors thrown by its
children components.
function renderTemplate(rid: number) {
return (
<ErrorBoundary catch={<div>fallback error</div>}>
<Suspense
rid={rid}
fallback={<MyAsyncFallback />}
catch={<div>Children error</div>}
>
<MyAsyncComponent />
</Suspense>
</ErrorBoundary>
);
}
const html = renderToStream(renderTemplate);
The above example would render <div>Children error</div>
if MyAsyncComponent
throws an
error, or <div>fallback error</div>
if MyAsyncFallback
throws an error. If both throws
an error, the first error will be changed to the second error as soon as the children
error is thrown.
Note
Until #14729 gets implemented,
you need to manually cast JSX.Element
into strings if you are sure there is no inner
async components in your component tree.
JSX elements are mostly strings everywhere. However, as the nature of this package, once a children element is a async component, the entire upper tree will also be async. Unless you are sure that no other component in your entire codebase is async, you should always handle both string and promise cases.
// It may or may not have inner async components.
const html = <Layout />;
if (html instanceof Promise) {
// I'm a promise, I should be awaited
console.log(await html);
} else {
// I'm a string, I can be used as is
console.log(html);
}
We choose to not provide a context
API. Here the reasons why:
This library only outputs strings (or Promises that resolve to strings), we don't manage
lifecycle or state. So the main selling point to having a context
API is to avoid
prop drilling.
A context Api would need to use a request identifier (internally called rid
) to be safe
to access asynchronously. Take for example two requests that are being processed at the
same time, both of them use some async components. If we use a global context without
scoping it behind some sort of rid
, the second request would override the context of the
first request. So for the first request, some part of the rendering (the non async part)
would use the correct context, but then the other section inside the async component would
use the context of the second request. This is bad.. Each component that uses context
would need to receive the rid
as a prop. This will defeat the purpose of having a
context in the first place. The rid
is needed to make the context async safe.
The only way to maintain data consistency across concurrent renders without attaching a
request locator (rid
), is by using
ALS. However, this
approach introduces a lot of overhead and a significant performance penalty.
Our recommendation is to use props. If you want to avoid prop drilling, you can use a composition style of writing components. This is a common pattern in other JSX-based libraries, and it works well with this library too.
// Drills only the required properties.
app.get('/', (request, response) => (
<YourDoctype title="Home">
<YourLayout loggedIn={!!request.user} lang={request.headers['accept-language']}>
<YourNavbar user={request.user} />
<YourContent />
<YourSubMenu path={request.url} username={request.user?.name} />
</YourLayout>
</YourDoctype>
));
Migrating from plain HTML to JSX can be a pain to convert it all manually, as you will find yourself hand placing quotes and closing void elements.
You can use Html To Jsx.
<!-- Hello world -->
<div class="awesome" style="border: 1px solid red">
<label for="name">Enter your name: </label>
<input type="text" id="name" />
</div>
<p>Enter your HTML here</p>
Results into:
<>
{/* Hello world */}
<div className="awesome" style={{ border: '1px solid red' }}>
<label htmlFor="name">Enter your name: </label>
<input type="text" id="name" />
</div>
<p>Enter your HTML here</p>
</>
The usage of htmx.org is super common with this project, this is why we also provide type definitions for all HTMX attributes.
You just need to add this triple slash directive to the top of your file:
/// <reference types="@kitajs/html/htmx.d.ts" />
const html = (
// Type checking and intellisense for all HTMX attributes
<div hx-get="/api" hx-trigger="click" hx-target="#target">
Click me!
</div>
);
This project supports the usage of Turbo Hotwire. We provide a separate export that you can use to provide type definitions for the elements and attributes used with Turbo Hotwire.
You just need to add this triple slash directive to the top of your file:
/// <reference types="@kitajs/html/hotwire-turbo.d.ts" />
const html = (
// Type checking and intellisense for all HTMX attributes
<turbo-frame id="messages">
<a href="/messages/expanded">Show all expanded messages in this frame.</a>
<form action="/messages">Show response from this form within this frame.</form>
</turbo-frame>
);
Often you will have a "template" html with doctype, things on the head, body and so on... Most users try to use them as a raw string and only use JSX for other components, but this is a not a good idea as you will have problems with it.
But you can always concatenate strings, like in this required use-case for <doctype>
export function Layout(props: Html.PropsWithChildren<{ head: string; title?: string }>) {
return (
<>
{'<!doctype html>'}
<html lang="en">
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>{props.title || 'Hello World!'}</title>
{props.head}
</head>
<body>{props.children}</body>
</html>
</>
);
}
const html = (
<Layout
head={
<>
<link rel="stylesheet" href="/style.css" />
<script src="/script.js" />
</>
}
>
<div>Hello World</div>
</Layout>
);
Html.compile
interface compiles a clean component into a super fast
component. This does not support unclean components / props processing.
Warning
This feature is a special use case for rendering entire page templates like what you would do with handlebars or nunjucks.
It does not works with mostly JSX components and, for small components, it will be slower than the normal JSX syntax.
This mode works just like prepared statements in SQL. Compiled components can give up to 2000 times faster html generation. This is a opt-in feature that you may not be able to use everywhere!
Due to the nature of
Proxy
objects, the spread operator (...
) will not work with compiled components. You need to
manually pass all props to their components.
import Html from '@kitajs/html';
function Component(props: Html.PropsWithChildren<{ name: string }>) {
return <div>Hello {props.name}</div>;
}
const compiled = Html.compile<typeof Component>(Component);
compiled({ name: 'World' });
// <div>Hello World</div>
const compiled = Html.compile((p) => <div>Hello {p.name}</div>);
compiled({ name: 'World' });
// <div>Hello World</div>
Properties passed for compiled components ARE NOT what will be passed as argument to the generated function.
const compiled = Html.compile((t) => {
// THIS WILL NOT print 123, but a string used by .compile instead
console.log(t.asd);
return <div></div>;
});
compiled({ asd: 123 });
That's the reason on why you cannot compile unclean components, as they need to process the props before rendering.
A clean component is a component that does not process props before applying them to the element. This means that the props are applied to the element as is, and you need to process them before passing them to the component.
// Clean component, render as is
function Clean(props: PropsWithChildren<{ sum: number }>) {
return <div>{props.sum}</div>;
}
// Calculation is done before passing to the component
html = <Clean sum={3 * 2} />;
// Unclean component, process before render
function Unclean(props: { a: number; b: number }) {
return <div>{props.a * props.b}</div>;
}
// Calculation is done inside the component, thus cannot be used with .compile()
html = <Unclean a={3} b={2} />;
JSX does not allow multiple root elements, but you can use a fragment to group multiple elements:
const html = (
<>
<div>1</div>
<div>2</div>
</>
);
Learn more about JSX syntax here!
All HTML elements and attributes should be supported.
Missing an element or attribute? Please create an issue or a PR to add it. It's easy to add.
The <tag of="">
tag is a custom internal tag that allows you to render any runtime
selected tag you want. Possibly reasons to prefer this tag over extending types:
- You want to render a tag that is chosen at runtime.
- You don't want to mess up with extending globally available types.
- You are writing javascript with typechecking enabled.
- You are writing a library and should not extend types globally.
- You need to use kebab-case tags, which JSX syntax does not support.
<tag of="asd" />
// <asd></asd>
<tag of="my-custom-KEBAB" />
// <my-custom-KEBAB></my-custom-KEBAB>
We do recommend using extending types instead, as it will give you intellisense and type checking.
Kita supports constructing class
attributes conditionally, which is a common use case
for many applications.
<div class={['a', true && 'b', false && 'c', 'd']} />
// <div class="a b d"></div>
<div class={['class-a class-b', true && 'class-c']} />
// <div class="class-a class-b class-c"></div>
This behavior is pretty similar and inspired from clsx, but we do not support objects as input.
Just as exemplified above, you may also want to add custom properties to your elements.
You can do this by extending the JSX
namespace.
declare global {
namespace JSX {
// Adds a new element called mathPower
interface IntrinsicElements {
mathPower: HtmlTag & {
// Changes properties to the math-power element
['my-exponential']: number;
// this property becomes the <>{children}</> type
children: number;
};
}
// Adds hxBoost property to all elements native elements (those who extends HtmlTag)
interface HtmlTag {
['hx-boost']: boolean;
// TIP: We already provide HTMX types, check them out!
}
}
}
const element = (
<mathPower my-exponential={2} hx-boost>
{3}
</mathPower>
);
// Becomes <math-power my-exponential="2" hx-boost>3</math-power>
We also provide a way to allow any tag/attribute combination, altough we do not recommend using it.
Just add this triple slash directive to the top of your file:
/// <reference types="@kitajs/html/all-types.d.ts" />
This package is just a string builder on steroids, as you can see how this works. This means that most way to isolate performance differences is to micro benchmark.
The below benchmark compares this package with other popular HTML builders, like React, Typed Html and Common Tags.
You can run this yourself by running pnpm bench
.
cpu: 13th Gen Intel(R) Core(TM) i5-13600K
runtime: node v20.9.0 (x64-linux)
benchmark time (avg) (min … max) p75 p99 p995
--------------------------------------------------- -----------------------------
• Many Components (31.4kb)
--------------------------------------------------- -----------------------------
Typed Html 39.32 µs/iter (31.5 µs … 460.86 µs) 34.97 µs 165.08 µs 216.67 µs
KitaJS/Html 11.68 µs/iter (8.31 µs … 610.87 µs) 10.29 µs 50.48 µs 132.37 µs
Common Tags 85.33 µs/iter (67.4 µs … 1.09 ms) 77.32 µs 316.12 µs 396.12 µs
React 27.93 µs/iter (17.81 µs … 878.52 µs) 24.98 µs 190.39 µs 252.25 µs
summary for Many Components (31.4kb)
KitaJS/Html
2.39x faster than React
3.37x faster than Typed Html
7.31x faster than Common Tags
• MdnHomepage (66.7Kb)
--------------------------------------------------- -----------------------------
Typed Html 367.8 µs/iter (282.65 µs … 5.63 ms) 361.54 µs 877.96 µs 1.11 ms
KitaJS/Html 71.61 µs/iter (50.04 µs … 1.39 ms) 63.63 µs 345.35 µs 429.49 µs
Common Tags 147.11 µs/iter (104.8 µs … 3.21 ms) 124.49 µs 819.16 µs 2.07 ms
React 178.66 µs/iter (138.89 µs … 1.84 ms) 173.31 µs 501.74 µs 584.45 µs
summary for MdnHomepage (66.7Kb)
KitaJS/Html
2.05x faster than Common Tags
2.49x faster than React
5.14x faster than Typed Html
• Many Props (7.4kb)
--------------------------------------------------- -----------------------------
Typed Html 79.78 µs/iter (66.44 µs … 750.62 µs) 71.83 µs 285.45 µs 343.03 µs
KitaJS/Html 18.14 µs/iter (15.14 µs … 2.52 ms) 16.66 µs 60.15 µs 105.96 µs
Common Tags 33.3 µs/iter (27.43 µs … 693.42 µs) 30.48 µs 104.2 µs 155.66 µs
React 48.27 µs/iter (38.89 µs … 1.24 ms) 43.81 µs 221.03 µs 268.48 µs
summary for Many Props (7.4kb)
KitaJS/Html
1.84x faster than Common Tags
2.66x faster than React
4.4x faster than Typed Html
This package just aims to be a drop in replacement syntax for JSX, and it works because
you tell tsc to transpile JSX syntax to calls to our own html
namespace.
<ol start={2}>
{[1, 2].map((i) => (
<li>{i}</li>
))}
</ol>
Gets transpiled by tsc to plain javascript:
Html.createElement(
'ol',
{ start: 2 },
[1, 2].map((i) => Html.createElement('li', null, i))
);
Which, when called, returns this string:
'<ol start="2"><li>1</li><li>2</li></ol>';
This package emits HTML as a compact string, useful for over the wire environments. However, if your use case really needs the output HTML to be pretty printed, you can use an external JS library to do so, like html-prettify.
import prettify from 'html-prettify';
const html = (
<div>
<div>1</div>
<div>2</div>
</div>
);
console.log(html);
// <div><div>1</div><div>2</div></div>
console.log(prettify(html));
// <div>
// <div>1</div>
// <div>2</div>
// </div>
👉 There's an open PR to implement this feature natively, wanna work on it? Check this PR.
This repository was initially forked from typed-html and modified to add some features and increase performance.
Initial credits to nicojs and contributors for the amazing work.
Licensed under the Apache License, Version 2.0.