ivi

v3.0.3

Published

6 months ago

Lightweight Embeddable Web UI Library.

Downloads

292

0High
0Medium
0Low

localvoid

ivi ui web-components custom-elements

ivi is a JavaScript library that enables you to create dynamic user interfaces using template literals with embedded JavaScript expressions. It identifies static and dynamic parts of your templates and efficiently updates only changed parts. ivi has a small memory footprint, supports lightweight stateful components, Server-Side Rendering, and Client-Side Hydration, all of which can help improve performance and efficiency of your web application.

f(state) => UI
The basic example is just 2.7KB.
Vite / Astro / Rollup plugins.
Precompiled templates optimized for size and performance.
Small memory footprint.
Embeddable.
Server-Side Rendering and Client-Side Hydration.

Examples

import { createRoot, update, component, useState } from "ivi";
import { htm } from "@ivi/htm";

const Example = component((c) => {
  const [count, setCount] = useState(c, 0);
  const inc = () => { setCount(count() + 1); };

  return () => htm`
    <div class="app">
      <div>${count()}</div>
      <button @click=${inc}>Increment</button>
    </div>
  `;
});

update(
  createRoot(document.body),
  Example(),
);

The size of the precompiled example above is just 2.7KB (minified+brotli). It includes entire runtime for declarative UI rendering. Precompiled templates are optimized for code size, cold-start performance and doesn't generate any additional code for hydration.

Setup

ivi templates will work without any precompilation for Client-Side Rendering,but it is highly recommended to use precompilation to improve performance and reduce code size.

Vite

"@ivi/vite-plugin" package provides Vite plugin that supports Client-Side Rendering and Server-Side Rendering.

// vite.config.mjs
import { defineConfig } from "vite";
import { ivi } from "@ivi/vite-plugin";

export default defineConfig({
  plugins: [ivi()],
});

Astro

"@ivi/astro" package provides Astro integration plugin that supports Server-Side Rendering and Client-Side Hydration.

// astro.config.mjs
import { defineConfig } from "astro/config";
import ivi from "@ivi/astro";

// https://astro.build/config
export default defineConfig({
  integrations: [ivi()],
});

Rollup

"@ivi/rollup-plugin" package provides Rollup plugin that supports Client-Side Rendering only.

IMPORTANT: Rollup plugin doesn't work with Vite in DEV mode.

// rollup.config.mjs
import { ivi } from "@ivi/rollup-plugin";

export default {
  input: "src/main.js",
  output: {
    file: "bundle.js",
  },
  plugins: [ivi()]
};

Babel Plugin

"@ivi/babel-plugin" package provides babel plugins for precompiling and optimizing templates.

"@ivi/babel-plugin/client" plugin precompiles templates and should be applied in a module transformation pass for Client-Side Rendering.
"@ivi/babel-plugin/client-optimizer" deduplicates shared data and should be applied in a chunk transformation pass for Client-Side Rendering.
"@ivi/babel-plugin/server" plugin precompiles templates for Server-Side Rendering.

Declarative UI

Templates

The ivi library allows you to create HTML templates using JavaScript template literals. These literals can be tagged with either htm or svg, and are comprised of static HTML and dynamic expessions that will be evaluated at runtime, allowing for dynamic content in templates.

When you write an ivi template expression, it does not immediately create or update the DOM. Instead, it describes the DOM structure as a VTemplate object. To actually create or update the DOM based on this description, you will need to pass VTemplate object to the update() function, along with a root node which will contain the rendered template.

import { createRoot, update } from "ivi";
import { htm } from "@ivi/htm";

const dynamicExpr = "World";
const example = htm`<h1>Hello ${dynamicExpr}</h1>`

update(
  createRoot(document.body),
  example,
);

The ivi library provides support for various template languages, and it's easy to create a new one. The complexity involved in compiling these templates is abstracted away in "ivi/template/..." modules.

HTML Template Language

"@ivi/htm" package provides HTML Template Language.

The ivi HTML Template Language has an HTML-like syntax with additional syntax for DOM properties, events and whitespace removal.

import { htm } from "@ivi/htm";
const Example = component((c) => {
  // ...
  return () => htm`
    <div class="app">
      <div>${count()}</div>
      <button @click=${inc}>Increment</button>
    </div>
  `;
});

ivi Template Language

"@ivi/tpl" package provides ivi Template Language.

The ivi Template Language was designed as a concise language for defining DOM tree structures and uses indentation for nesting DOM nodes.

import { htm } from "@ivi/tpl";
const Example = component((c) => {
  // ...
  return () => htm`
    div.app
      div ${count()}
      button @click=${inc} "Increment"
  `;
});

Expressions

In ivi templates, you can include dynamic content called expressions. An expression is just a piece of JavaScript code that gets evaluated when template is rendered. Whatever value an expression produces at that time will be included in the final rendered template.

htm`
<div attr=${attributeValueExpr}>
  ${childExpr}
</div>`;

HTML Template Language Syntax

HTML Template Language supports additional syntax to work with DOM properties, events, etc.

<div name="value" /> - Static attribute.
<div name /> - Static attribute.
<div name=${expr} /> - Dynamic attribute element.setAttribute(name, expr).
<div .name=${expr} /> - Property element[name] = expr.
<div *name=${expr} /> - Property element[name] = expr, diffs against a DOM value.
<div ~name="value" /> - Static style <div style="name:value;">.
<div ~name=${expr} /> - Dynamic style element.style.setProperty(name, expr).
<div @name=${expr} /> - Event element.addEventListener(name, expr).
<div ${directive} /> - Client-Side Element Directive directive(element).
<div &=${directive} /> - Client-Side Element Directive directive(element).
<div &:ssr=${directive} /> - Element Directive that works during Client-Side and Server-Side Rendering directive(element, hydrate).

Conditionals

You can use regular JavaScript expressions in your templates, which means you can use any javascript control flow constructs like conditional operators, function calls, and if or switch statements to generate dynamic content based on runtime conditions.

This means you can create templates with complex logic that conditionally renders different content based on what's happening in your application. You can nest template expressions inside one another to build up more complex templates, and you can store the results of templates in variables to use them later in your code.

const Example = component((c) => {
  // ...
  return (show) => htm`
    <div>
      ${show && htm`<span>Show</span>`}
    </div>
  `;
});

Arrays

If an expression is used in the child position of an HTML element and it returns an array, ivi will render all of the items in that array as separate nodes.

const Example = () => htm`
  <div>
    ${[
      "Text Node 1",
      "Text Node 2",
    ]}
  </div>
`;

ivi allows components to return arrays of elements as their root nodes. This means that a component can return multiple top-level elements instead of just one.

For example, a component could return an array of <li> elements that make up a list. When this component is rendered, ivi will treat the array of <li> elements as a set of top-level elements, just like it would with a single root element.

This feature provides more flexibility when building complex UI components, as it allows you to create components that generate a dynamic number of top-level elements depending on their input.

const Example = component((c) => {
  return (entries) => entries.map((e) => htm`
    <li>${e}</li>
  `);
);
// Example([1, 2, 3])

When arrays are updated, stateless tree nodes are mapped onto their stateful nodes by their position in the array.

When array contains a conditional expression that returns a "hole" value (null, undefined or false), the hole will occupy a slot in a stateful tree, so that all nodes will be correclty mapped onto their stateful nodes.

[
  conditional ? "text" : null,
  StatefulComponent(),
]

In the example above, when conditional expression goes from a text to a "hole" and vice versa, StatefulComponent will preserve its internal state.

When array grows or shrinks in size, stateful nodes will be created or removed at the end of an array.

Dynamic Lists

In ivi, you can render lists of items using the map() function that loops through an array of data and returns a list of elements. However, when list is updated, it is important to correctly map rendered items onto their stateful views. This means that if an item is rendered as a component that has internal state that could change as a result of user actions or external events, it should be mapped onto the same component instance.

To render dynamic lists, ivi provides the List() function.

function List<E, K>(
  // Input Entries.
  entries: E[],
  // Function that retrieves unique key from an entry.
  getKey: (entry: E, index: number) => K,
  // Function that renders an entry.
  render: (entry: E) => VAny,
): VList;

It creates a dynamic list with an array of keys that uniquely identify each item in the list. When list is updated, ivi uses keys to map items onto their stateful nodes.

It's important to note that when rendering a dynamic list, you should always use a unique identifier as a key. This helps ivi identify each element in a list and avoid rendering errors. If you use an index or a random value as a key, ivi may not be able to identify correct elements in a list, which can cause errors.

interface DataEntry {
  key: number;
  text: string;
}
const getEntryKey = (entry: DataEntry) => entry.key;
const EntryView = (entry: DataEntry) => (
  htm`<li>${entry.text}</li>`
);

const ListView = (data: DataEntry[]) => htm`
  <ul>${List(data, getEntryKey, EntryView)}</ul>
`;

ivi is using an optimal algorithm for dynamic lists that uses the minimum number of Node.insertBefore() operations to rearrange DOM nodes.

Almost all popular libraries are using naive algorithms and can handle efficiently only basic use cases. And some libraries optimizing their list diffing algorithms only for use cases that are used in benchmarks.

Reducing Node.insertBefore() operations is important not just because it invalidates internal DOM state, but also because each time one of the DOM nodes attached to the document is moved, it may produce a MutationObserver notification. And a lot of popular extensions are using Mutation Observers to observe entire document subtree, so each insertBefore operation can become quite costly when it is used outside of benchmarking sandboxes.

Components

Components can be either stateful or stateless. Stateful components are used when you need to manage state that changes over time, such as user input, network requests, or animations.

Stateful Components

Stateful components are declared with component() function. It creates a factory function that produces component nodes.

// `component()` function creates a factory function for component
// nodes of this type.
const Example = component((c) => {
  // When component state is initialized, it should return a render
  // function.
  return (props) => (
    htm`<div>${props.value}</div>`
  );
});

update(
  document.body,
  Example({ value: "Hello World" }),
);

Stateful components are using JavaScript closures to store internal state.

const Example = component((c) => {
  // Internal state.
  let _counter = 0;

  // Event handler.
  const increment = () => {
    // Mutate internal state.
    _counter++;
    // Invalidate component and schedule an update.
    invalidate(c);
  };

  // Render function.
  return () => htm`
    <div>
      <p>Count: ${_counter}</p>
      <button @click=${increment}>Increment</button>
    </div>
  `;
});

When internal state is mutated, it doesn't trigger component updates automatically and it should be manually invalidated with invalidate() function.

There are high-level APIs like useState() or useReducer() that use the low-level invalidate() function behind the scenes to automatically invalidate components when internal state is mutated.

const Example = component((c) => {
  // Internal state.
  const [counter, setCounter] = useState(c, 0);

  const increment = () => {
    // Automatically invalidates component when counter value is mutated.
    setCounter(counter() + 1);
  };

  // Render function.
  return () => (
    htm`
    <div>
      <p>Count: ${counter()}</p>
      <button @click=${increment}>Increment</button>
    </div>`
  );
});

Stateless Components

Stateless components in ivi are just basic JavaScript functions. They are faster and more lightweight than stateful components, which makes them a good choice for simple, reusable components that don't need to manage state.

const Button = (text, onClick) => htm`
  <button @click=${onClick}>${text}</button>
`;

API

Opaque Types

Opaque type hides its internal representation at boundaries between a module and code that works with the module.

It is useful in use cases like Server-Side Rendering. When code is executed in the SSR context, some ivi types will have completely different internal representation.

Stateful Tree

type SNode = Opaque;
type Root<State> = Opaque<State>;
type Component<Props> = Opaque<Props>;

Stateless Tree

type VAny =
  | null       // Hole
  | undefined  // Hole
  | false      // Hole
  | string     // Text
  | number     // Text
  | VRoot      // Root
  | VTemplate  // Template
  | VComponent // Component
  | VContext   // Context Provider
  | VList      // Dynamic List with track by key algo
  | VAny[]     // Dynamic List with track by index algo
  ;

type VRoot = Opaque;
type VTemplate = Opaque;
type VComponent = Opaque;
type VContext = Opaque;
type VList = Opaque;

Root Nodes

A root node is the topmost node in a stateful tree, from which all other nodes are rendered. It represents an entry point for the ivi rendering algorithm and stores a position in the DOM tree.

`createRoot()`

createRoot creates a root node that uses microtask queue for scheduling updates.

function createRoot(
  parentElement: Element,
  nextNode: Node | null = null,
): Root;

parentElement - Parent DOM Element.
nextNode - Next DOM Node.

`dirtyCheck()`

dirtyCheck performs the dirty checking algorithm in a root subtree and updates all dirty components.

function dirtyCheck(
  root: Root,
  forceUpdate: boolean = false,
): void;

root - Root node.
forceUpdate - Force all components to update, even when they are using optimization hints to reduce updates.

`update()`

update updates a root subtree with a new representation.

function update(
  root: Root,
  v: VAny,
  forceUpdate: boolean = false,
): void;

root - Root node.
v - New representation.
forceUpdate - Force all components to update, even when they are using optimization hints to reduce updates.

`unmount()`

unmount unmounts a root subtree from the DOM and triggers unmount hooks in components.

function unmount(
  root: Root,
  detach: boolean,
): void;

root - Root node.
detach - Detach the topmost DOM nodes from the DOM subtree.

`hydrate()`

hydrate hydrates a Server-Side Rendered root subtree.

function hydrate(
  root: Root,
  v: VAny,
): void;

root - Root node.
v - Current representation.

`defineRoot()`

defineRoot creates a root node factory that uses a custom OnRootInvalidated hook.

function defineRoot(
  onInvalidate: (root: Root<undefined>) => void,
) : (parentElement: Element, nextNode: Node | null) => Root<undefined>;

function defineRoot<S>(
  onInvalidate: (root: Root<S>, state: S) => void,
) : (parentElement: Element, nextNode: Node | null, state: S) => Root<S>;

onInvalidate - OnRootInvalidated hook that receives a root node and custom state associated with that root node.

Components

`component()`

component creates a factory that produces component nodes.

function component(
  factory: (c: Component) => () => VComponent<undefined>,
  areEqual?: () => boolean
): () => VComponent<undefined>;
function component<P>(
  factory: (c: Component) => (props: P) => VAny,
  areEqual?: (prev: P, next: P) => boolean
): (props: P) => VComponent<P>;

factory - Function that produces stateful component render functions.
areEqual - Optional function that checks input properties for changes and is used as an optimization hint to reduce unnecessary updates when properties didn't change.

When root subtree is updated with forceUpdate option, areEqual hint is ignored and all components are updated.

`getProps()`

getProps gets current component props from component instance.

function getProps = <P>(component: Component<P>): P;

component - Component instance.

`invalidate()`

invalidate invalidates component and schedules an update.

function invalidate(component: Component): void;

component - Component instance.

`useUnmount()`

Adds an unmount hook.

function useUnmount(
  component: Component,
  hook: () => void,
): void;

component - Component instance.
hook - Unmount hook.

Component State

`useMemo()`

useMemo creates a memoized function.

function useMemo<T, U>(
  areEqual: (prev: T, next: T) => boolean,
  fn: (props: T) => U,
): (props: T) => U;

areEqual - Checks input properties for changes to avoid recomputations.
fn - Function to memoize.

`useState()`

useState creates a reactive component state.

function useState<S>(
  component: Component,
  state: S,
): [
  get: () => S,
  set: (s: S) => void,
];

component - Component instance.
state - Initial state.

Returns state getter and state setter functions.

`useReducer()`

useReducer creates a reactive component state reducer.

type Dispatch<A> = (action: A) => void;

function useReducer<S, A>(
  component: Component,
  state: S,
  reducer: (state: S, action: A) => S,
): [
  get: () => S,
  dispatch: Dispatch<A>,
];

component - Component instance.
state - Initial state.
reducer - State reducer function.

Returns state getter and action dispatcher functions.

Side Effects

Side effects lets you specify how your components should behave with external systems such as imperative API calls, timer manipulations, or direct DOM interactions.

You can think of it as a combination of mount, update and unmount lifecycles hooks.

`useEffect()`

useEffect creates a side effect that is executed immediately after root node finishes an update.

function useEffect(
  component: Component,
  effect: () => (() => void) | void,
): () => void;
function useEffect<P>(
  component: Component,
  effect: (props: P) => (() => void) | void,
  areEqual?: (prev: P, next: P) => boolean
): (props: P) => void;

component - Component instance.
effect - Effect hook.
areEqual - Optional function that checks input properties for changes and is used to control when an effect should be updated.

Returns a side effect function that should be invoked in a render function.

`useLayoutEffect()`

useLayoutEffect creates a side effect that is executed before animation frame.

function useLayoutEffect(
  component: Component,
  effect: () => (() => void) | void,
): () => void;
function useLayoutEffect<P>(
  component: Component,
  effect: (props: P) => (() => void) | void,
  areEqual?: (prev: P, next: P) => boolean
): (props: P) => void;

component - Component instance.
effect - Effect hook.
areEqual - Optional function that checks input properties for changes and is used to control when an effect should be updated.

Returns a side effect function that should be invoked in a render function.

`useIdleEffect()`

useIdleEffect creates a side effect that is executed when browser is idle.

function useIdleEffect(
  ccomponent: Component,
  effect: () => (() => void) | void,
): () => void;
function useIdleEffect<P>(
  ccomponent: Component,
  effect: (props: P) => (() => void) | void,
  areEqual?: (prev: P, next: P) => boolean
): (props: P) => void;

component - Component instance.
effect - Effect hook.
areEqual - Optional function that checks input properties for changes and is used to control when an effect should be updated.

Returns a side effect function that should be invoked in a render function.

List

`List()`

List creates a dynamic lists.

function List<E, K>(
  entries: E[],
  getKey: (entry: E, index: number) => K,
  render: (entry: E) => VAny,
): VList;

entries - Input data.
getKey - Function that should return an unique key for each data entry.
render - Function that renders an entry.

Context

`context()`

context creates context getter and context provider functions.

function context = <T>(): [
  get: (component: Component) => T | undefined,
  provider: (value: T, children: VAny) => VContext<T>,
]

Returns a get function that finds the closest context value, and a provider function that creates context nodes.

// Creates a getter and provider functions.
const [getContextValue, contextValueProvider] = context();

const Example = component((c) => {
  return () => htm`
    <h1>Hello ${getContextValue(c)}</h1>
  `;
});

update(
  createRoot(document.body),
  contextValueProvider(
    "World",
    Example(),
  ),
);

Element Directive

ElementDirective is an escape hatch that allows extending ivi rendering algorithm.

type ElementDirective = <E extends Element>(
  element: E,
  hydrate?: boolean,
) => void | string | { a?: string, c?: string; };

DOM Utilities

`eventDispatcher()`

eventDispatcher creates an event dispatcher that finds the closest child DOM node and emits a CustomEvent with EventTarget.dispatchEvent() method.

interface DispatchEventOptions {
  // Option indicating whether the event bubbles. The default
  // is `true`.
  bubbles?: boolean;
  // Option indicating whether the event can be cancelled. The
  // default is `false`.
  cancelable?: boolean;
  // Option indicating whether the event will trigger listeners
  // outside of a shadow root. The default is `false`.
  composed?: boolean;
}

type EventDispatcher = {
  (component: Component): boolean;
  <T>(component: Component, value: T): boolean;
};

function eventDispatcher = <T>(
  eventType: string,
  options?: DispatchEventOptions,
): EventDispatcher;

eventType - Event type.
options - Event options that will be used when event is dispatched.

Event dispatcher invokes event handlers synchronously. All event handlers are invoked before event dispatcher returns.

`findDOMNode()`

findDOMNode finds the closest DOM node child that belongs to a stateful node subtree.

function findDOMNode<T extends Node | Text>(
  node: SNode | null,
): T | null;

node - Stateful node.

`containsDOMElement()`

containsDOMElement checks if a stateful node contains a DOM elements in its subtree.

function containsDOMElement(
  node: SNode,
  element: Element,
): boolean;

node - Stateful node.
element - DOM Element.

`hasDOMElement()`

hasDOMElement checks if a stateful node has a DOM element as its child.

function hasDOMElement(
  node: SNode,
  child: Element,
): boolean;

node - Stateful node.
child - DOM Element.

Equality Functions

`preventUpdates()`

preventUpdates is a noop function that always returns true value.

function preventUpdates<T>(a: T, b: T): true;

`strictEq()`

strictEq checks values for equality with strict equality operator ===.

function strictEq<T>(a: T, b: T): boolean;

`shallowEq()`

shallowEq checks objects with shallow equality algorithm and uses strict equality operator to check individual values for equality.

function shallowEq<T extends object>(a: T, b: T): boolean;

`shallowEqArray()`

shallowEqArray checks arrays with shallow equality algorithm and uses strict equality operator to check individual values for equality.

function shallowEqArray<T>(a: T[], b: T[]): boolean;

Escape Functions

When creating ElementDirective functions that support Server-Side Rendering, it is important to escape attribute values and children text nodes to prevent cross-site scripting (XSS) attacks.

`escapeHTMLAttribute()`

escapeHTMLAttribute escapes HTML attribute values.

function escapeHTMLAttribute(str: string): string;

`escapeHTMLText()`

escapeHTMLText escapes HTML text nodes.

function escapeHTMLText(str: string): string;

CheatSheet

Passive Event Listener

const Example = component(() => {
  const _onTouchDown = (ev) => {};

  const addPassiveTouchDown = (element) => {
    element.addEventListener(
      "touchdown",
      _onTouchDown,
      { passive: true },
    );
  };

  return () => htm`
    <div ${addPassiveTouchDown}></div>
  `;
});

Dynamic Argument Name

const useDynamicArg = () => {
  let prevKey;
  let prevValue;
  return (key, value) => (element) => {
    if (prevKey !== key) {
      if (prevKey) {
        element.removeAttribute(prevKey);
      }
      element.setAttribute(key, value);
    } else if (prevValue !== value) {
      element.setAttribute(key, value);
    }
  };
};

const Example = component(() => {
  const arg = useDynamicArg();

  return ([key, value]) => htm`
    <div ${arg(key, value)}></div>
  `;
});

Stateless Components with `areEqual` hook

const _Button = ([text, onClick]) => htm`
  <button @click=${onClick}>${text}</button>
`;
const Button = component(
  // Component factory will reuse the same hoisted render
  // function for all component instances.
  () => _Button,
  // areEqual function
  (prev, next) => (
    prev[0] === next[0] &&
    prev[1] === next[1]
  ),
);

Integrating External/Imperative Libraries

import { createRoot, update, component, findDOMNode, useEffect } from "ivi";
import { htm } from "@ivi/htm";
import { EditorView, basicSetup } from "codemirror";
import { javascript } from "@codemirror/lang-javascript";

const CodeMirror = component((c) => {
  let _editor;

  useEffect(c, () => {
    _editor = new EditorView({
      extensions: [basicSetup, javascript()],
      // findDOMNode finds the closest child DOM node.
      parent: findDOMNode(c),
    });

    // Reset function will be invoked when component is unmounted.
    return () => {
      _editor.destroy();
    };
  })();
  // ^ When effect doesn't have any dependencies, it can be executed just
  // once in the outer scope. Effect will run when its DOM tree is mounted.

  return () => htm`
    <div class="CodeMirror"></div>
  `;
});

update(
  createRoot(document.body),
  CodeMirror(),
);

Advanced

Component Invalidation and Dirty Checking

Component invalidation algorithm is implemented by marking component as dirty and marking all its parent nodes with a flag that they have a dirty subtree. When marking algorithm reaches a root node, it invokes OnRootInvalidated() hook that can be used to implement a custom scheduler.

Component invalidated and marked with Dirty flag.
Node marked with DirtySubtree flag.
Root Node marked with DirtySubtree flag, OnRootInvalidated() hook invoked.
Component invalidated and marked with Dirty flag, parents already marked with DirtySubtree flag.

When scheduler decides to update a root node with a dirty subtree, it starts a dirty checking algorithm. This algorithm goes top-down in a right-to-left order, visiting all nodes with a dirty subtree flag until it reaches a dirty component and updates it.

Starts dirty checking from the root node.
Clean node, skips visiting its subtree.
Node with DirtySubtree flag, starts checking its children.
Component with Dirty flag, triggers an update.
Component with Dirty flag, triggers an update.
Clean node, skips visiting its subtree.

Right-to-Left Updates

One of the reasons why the core library is so small is because update algorithm is implemented in RTL order. Algorithm that performs updates in RTL order simplifies a lot of complex issues with DOM updates. The main issue with DOM updates is that when we start updating a DOM tree structure, we need to have a reference to a parent and a next DOM node, so that we can use parent.insertBefore(newNode, nextNode). In most cases it is easy to retrieve a next DOM node, but there are edge cases like when we have two adjacent conditional expressions and one of their states is that it completely removes a DOM node from the tree, or two adjacent components with conditionals at their roots, etc.

Majority of libraries are dealing with this edge cases by introducing marker DOM nodes (comment or an empty text node). For example, to implement conditional expressions we can add an empty text node when conditional doesn't render any DOM node and when conditional goes into a state when it needs to add a DOM node, it will use a marker node as a next DOM node reference. In some edge case scenarios, some libraries can insert a lot of marker nodes. Update algorithm in ivi doesn't use any marker nodes.

The RTL algorithm that is used in ivi also makes it way much easier to implement node displacements without introducing any additional code paths, fragments and pretty much everything that involves updating a DOM structure.

Template Call-Site Unique Identity

Each call-site that creates template has unique identity, so even identical templates created from different call-sites won't be able to diff against each other.

function TemplateUniqueIdentity(condition, text) {
  return (condition)
    ? htm`div ${text}`
    : htm`div ${text}`;
}

In th example above, when condition is changed, instead of updating text node, update algorithm will replace entire div element with a new one.

Forcing Component Updates

There are some use cases that require a lot of frequent reads from a reactive variable. And whenever this variable changes, it affects a lot of UI nodes, like switching between light/dark themes.

Instead of creating a lot of subscriptions to this variables, it is recommended to use simple javascript values and rerender entire UI subtree with dirtyCheck(root, true) when this values are changed.

const root = createRoot(document.getElementById("app"));

let theme = "Light";
function setTheme(t) {
  if (theme !== t) {
    theme = t;
    dirtyCheck(root, true);
  }
}

const App = component((c) => {
  const toggleTheme = () => {
    setTheme((theme === "Light") ? "Dark" : "Light");
  };
  return () => htm`
    div
      div =${theme}
      button @click=${toggleTheme} 'Toggle Theme'
  `;
});

update(root, App());

Template Cloning

Template cloning is an optimization that is used for cloning HTML templates with a Node.cloneNode() method.

By default, template cloning is enabled for all templates. But sometimes it would be wasteful to create a template for cloning and instantiate from it when this template is rendered just once on the page.

To disable cloning, template should have a leading comment /* preventClone */. E.g.

const Example = () => /* preventClone */htm`
  <div class="Title">${text}</div>
`;

Templates with just one element that doesn't have any static properties will be created with document.createElement().

htm`<div attr=${0}>${1}</div>`;

Disabling Static Template Generation for Client-Side Rendering

When templates are precompiled, they are producing static HTML template string and OpCodes for dynamic updates. In a lot of use cases with Server-Side Rendering and partial hydration, static template part is generated on the server and client just needs OpCodes to perform dynamic updates.

To reduce code size that is transfered to the client in such scenarios, it is possible to disable static template generation with /* ssr */ comment.

export default component((c) => {
  let _open = true;
  const onClick = () => {
    _open = !_open;
    invalidate(c);
  };

  return ({ children }) => {
    return /* ssr */ htm`
    <div class="toggle" class=${_open ? "toggle open" : "toggle"}>
      <a @click=${onClick}>
        ${_open ? "[-]" : "[+] comments collapsed"}
      </a>
    </div>
    <ul class="comment-children" ~display=${_open ? "block" : "none"}>
      ${children}
    </ul>
    `;
  };
});

Internal Data Structures

To get a rough estimate of memory usage it is important to understand internal data structures.

In the description below we are going to calculate memory usage in a Chromium-based engines with Pointer Compression in V8.

UI Tree

UI Tree is implemented with a stateful tree SNode and immutable stateless tree VAny.

Stateless Tree has a simple data structure:

// 20 bytes
interface VNode<D extends VDescriptor, P> {
  // Descriptors are reused for all VNodes with the same type and its memory
  // usage can be ignored during estimation.
  readonly d: D;
  // Prop value is used for storing the results of template expressions in an
  // array, prop value for Components, or VRoot and VList props.
  readonly p: P;
}

type VArray = VAny[];
type VAny =
  | null       // empty slot
  | undefined  // empty slot
  | false      // empty slot
  | string     // text
  | number     // text
  | VRoot      // VNode<RootDescriptor, RootProps>
  | VTemplate  // VNode<TemplateDescriptor, P>
  | VComponent // VNode<ComponentDescriptor, P>
  | VContext   // VNode<ContextDescriptor, ContextProps<T>>
  | VList      // VNode<ListDescriptor, ListProps<K>>
  | VArray     // VAny[]
  ;

// 20 bytes
// Root Props stores a location where its children should be rendered.
interface RootProps {
  // Parent Element
  p: Element,
  // Next Node
  n: Node | null,
}

// 20 bytes
// Context Props stores a context value and stateless child node.
interface ContextProps<T> {
  // Context value
  v: T;
  // Stateless child
  c: VAny;
}

// 20 bytes
interface ListProps<K> {
  // Keys that uniquely identify each stateless node in a dynamic list.
  k: K[],
  // Stateless nodes
  v: VAny[],
}

For each stateless node VAny there is a stateful node SNode that has an interface:

// 32 bytes
interface SNode1<V extends VAny, S1> {
  // Stateless node associated with the current state.
  v: V;
  // Bitflags
  f: Flags; // SMI value - Small Integer
  // Children nodes.
  c: SNode | (SNode | null)[] | null;
  // Parent node.
  p: SNode | null,
  // State Slot #1.
  s1: S1;
}

// 36 bytes
interface SNode2<V = VAny, S1 = any, S2 = any> extends SNode1<V, S1> {
  // State slot #2.
  s2: S2;
}

// Stateful Nodes are using two different shapes. Call-sites that accessing its
// flags to determine node type will be in a polymorphic state. In this case it
// is perfectly fine to use polymorphic call-sites to reduce memory usage.
type SNode<V = VAny> = SNode1<V> | SNode2<V>;

// Additional state size of the root nodes depends on the implementation of
// root nodes. Default root implementation doesn't use any additional state and
// stores `null` value in the additional state slot.
type SRoot<S> = SNode1<VRoot, S>;
// Text nodes are storing a reference to a Text DOM node.
type SText = SNode1<string | number, Text>;
// Template nodes are storing a reference to a root DOM node, DOM nodes with
// dynamic properties and DOM nodes that will be used as a reference for
// `parent.insertBefore(node, nextNode)` operations. Slots for DOM nodes with
// dynamic properties that also used as a reference for insertBefore operation
// will share the same slots, there won't be any duplicated references.
type STemplate = SNode1<VTemplate, Node[]>;
// Dynamic lists doesn't have any additional state.
type SList = SNode1<VList, null>;
// Components are using State Nodes with 2 state slots.
type SComponent = SNode2<
  VComponent,
  // Render function.
  //
  // Stateless components will share the same function.
  // Stateful components will create closures and its memory usage will depend
  // on the size of the closure context.
  null | ((props: any) => VAny),
  // Unmount hooks.
  //
  // Usually components don't have any unmount hooks, or they have just one
  // unmount hook.
  //
  // When there is one hook, it will be stored without any additional arrays.
  // If we add one more hook, array will be preallocated with exactly two
  // slots `[firstHook, newHook]`. And when it grows even more, javascript
  // engine will preallocate internal storage using a growth factor[1][2].
  //
  // 1. https://en.wikipedia.org/wiki/Dynamic_array#Growth_factor
  // 2. https://github.com/v8/v8/blob/1e6775a539a3b88b25cc0ffdb52529c68aad2be8/src/objects/js-objects.h#L584-L590
  null | (() => void) | (() => void)[]
>;
// Contexts doesn't have any additional state.
type SContext = SNode1<null, null>;

This data structures were carefully designed to have small memory overhead and avoid a lot of polymorphic/megamorphic call-sites that access this data structures.

To understand why monomorphic call-sites are important for performance, it is recommended to read a great article on this topic: "What's up with monomorphism?".

Template

Templates are precompiled into a static part that is stored in a TemplateDescriptor object and an array of dynamic expressions.

const Example = (attr, child) => htm`div :attr=${attr} span ${child}`;

Gets compiled into:

// _T() creates TemplateDescriptor
const _tpl_1 = _T(
  // _h() creates a template factory that uses Node.cloneNode(true) to
  // instantiate static template structure.
  _h("<div><span></span></div>"),
  // SMI (Small Integer) value that packs several values:
  // struct Data {
  //   stateSize:6;    // The number of state slots
  //   childrenSize:6; // The number of children slots
  //   svg:1;          // Template with SVG elements
  // }
  // stateSize and childrenSize are used for preallocating arrays with
  // exact number to avoid dynamic growth and reduce memory consumption.
  1026,
  // propOpCodes is an array of SMI values that stores opCodes for updating
  // element properties.
  [2],
  // childOpCodes is an array of SMI values that stores opCodes for updating
  // children nodes.
  [7, 4],
  // stateOpCodes is an array of SMI values that stores opCodes for traversing
  // DOM nodes and saving references to DOM nodes into internal state when
  // template is instantiated.
  [4],
  // Data is an array of string values that stores keys for dynamic properties.
  ["attr"],
);
// _t() creates stateless tree node VTemplate with shared TemplateDescriptor
// and an array of dynamic expressions.
const Example = (attr, child) => _t(_tpl_1, [attr, child]);

// Descriptor with TemplateData and template factory function.
type TemplateDescriptor = VDescriptor<TemplateData, () => Element>;

interface TemplateData {
  // stateSize / childrenSize / svg flag
  f: number,
  // Prop OpCodes
  p: PropOpCode[],
  // Child OpCodes
  c: ChildOpCode[],
  // State OpCodes
  s: StateOpCode[],
  // Data
  d: any[],
}

// Stateless tree node VTemplate.
type VTemplate<P = any> = VNode<TemplateDescriptor, P>;

Template Optimizations

@ivi/babel-plugin module contains a babel plugin with a template compiler and optimizer that significantly improves start-up performance.

Template compiler doesn't just eliminate compilation step during runtime, it also hoists static expressions and deduplicates OpCodes, static data and template factory functions. E.g.

import { className } from "styles.css";

const a = (id) => htm`
<div class=${className} id=${id}></div>
`;
const b = (id) => htm`
<div class=${className} id=${id}></div>
`;

Will generate two different templates with shared data structures:

import { className } from "styles.css";
import { _h, _T, _t } from "ivi";

const EMPTY_ARRAY = [];
const SHARED_DATA = ["id"];
const ELEMENT_FACTORY_1 = _h('<div class="' + className + '"></div>');
const SHARED_OP_CODES_1 = [/*..*/];
const _tpl_a = _T(
  /* factory */ELEMENT_FACTORY_1,
  /* flags */0,
  /* propOpCodes */SHARED_OP_CODES_1,
  /* childOpCodes */EMPTY_ARRAY,
  /* stateOpCodes */EMPTY_ARRAY,
  /* data */SHARED_DATA,
);
const _tpl_b = _T(
  /* factory */ELEMENT_FACTORY_1,
  /* flags */0,
  /* propOpCodes */SHARED_OP_CODES_1,
  /* childOpCodes */EMPTY_ARRAY,
  /* stateOpCodes */EMPTY_ARRAY,
  /* data */SHARED_DATA,
);

const a = (id) => _t(_tpl_a, [id]);
const b = (id) => _t(_tpl_b, [id]);

Quite often, OpCodes that are used for different purposes (props,child,state) are going to have similar values, so when OpCodes are deduplicated they are treated as simple arrays with integers that can be used for different purposes.

Shared data SHARED_DATA is deduplicated into one array that is shared between all templates.

Custom Scheduler

ivi is designed as an embeddable solution, so that it can be integrated into existing frameworks or web components. The basic root node instantiated with createRoot() function is using microtask queue to schedule updates. Root nodes with custom scheduling algorithm can be created by defining new root factories with defineRoot() function.

function defineRoot(onInvalidate: (root: Root<undefined>) => void)
  : (parentElement: Element, nextNode: Node | null) => Root<undefined>;
function defineRoot<S>(onInvalidate: (root: Root<S>) => void)
  : (parentElement: Element, nextNode: Node | null, state: S) => Root<S>;

As an example, to remove any batching and immediately update root subtree when it is invalidated we can define the following root node:

import { defineRoot } from "ivi";

const createSyncRoot = defineRoot((root) => {
  // Immediately triggers dirty checking.
  dirtyCheck(root);
});

Using `requestAnimationFrame()` for Scheduling UI Updates

Scheduling algorithm with rAF batching has some potential footguns with race conditions.

function formStateReducer(state, action) {
  switch (action.type) {
    case "update":
      return {
        value: action.value,
        valid: /^[a-z]+$/.test(action.value),
      };
  }
  return state;
}

const Form = component((c) => {
  const [state, dispatch] = useReducer(c,
    { value: "", valid: false },
    formStateReducer,
  );
  const onInput = (ev) => {
    dispatch({ type: "update", value: ev.target.value });
  };
  return () => htm`
    form
      input
        @input=${onInput}
        *value=${state().value}
      input
        :type='submit'
        :value='Submit'
        .disabled=${!state().valid}
  `;
});

update(
  createRoot(document.getElementById("app")!),
  Form(),
);

In this example, if the user types really fast and pushes an [enter] button, it is possible to get an execution order like this:

User types 0 into <input>.
onChange() event handler is triggered, state.valid switches into a false state.
User pushes an [enter] button.
Browser sends submit request because UI is still in the old state <input type="submit" .disabled={false} />
rAF event is triggered, submit button goes into disabled state.

The simplest way to avoid issues like this is to use microtasks for batching. But if you really want to add rAF scheduling, it is possible to solve issues like this by introducing some synchronization primitives:

import { uiReady } from "my-custom-scheduler";

const onSubmit = async (ev) => {
  await uiReady();
  submit();
};

License

MIT

Published

Vulnerabilities

Links

Maintainers

Keywords

Readme

Examples

Table of Contents

Setup

Vite

Astro

Rollup

Babel Plugin

Declarative UI

Templates

HTML Template Language

ivi Template Language

Expressions

HTML Template Language Syntax

Conditionals

Arrays

Dynamic Lists

Components

Stateful Components

Stateless Components

API

Opaque Types

Stateful Tree

Stateless Tree

Root Nodes

createRoot()

dirtyCheck()

update()

unmount()

hydrate()

defineRoot()

Components

component()

getProps()

invalidate()

useUnmount()

Component State

useMemo()

useState()

useReducer()

Side Effects

useEffect()

useLayoutEffect()

useIdleEffect()

List

List()

Context

context()

Element Directive

DOM Utilities

eventDispatcher()

findDOMNode()

containsDOMElement()

hasDOMElement()

Equality Functions

preventUpdates()

strictEq()

shallowEq()

shallowEqArray()

Escape Functions

escapeHTMLAttribute()

escapeHTMLText()

CheatSheet

Passive Event Listener

Dynamic Argument Name

Stateless Components with areEqual hook

Integrating External/Imperative Libraries

Advanced

Component Invalidation and Dirty Checking

Right-to-Left Updates

Template Call-Site Unique Identity

Forcing Component Updates

Template Cloning

Disabling Static Template Generation for Client-Side Rendering

Internal Data Structures

`createRoot()`

`dirtyCheck()`

`update()`

`unmount()`

`hydrate()`

`defineRoot()`

`component()`

`getProps()`

`invalidate()`

`useUnmount()`

`useMemo()`

`useState()`

`useReducer()`

`useEffect()`

`useLayoutEffect()`

`useIdleEffect()`

`List()`

`context()`

`eventDispatcher()`

`findDOMNode()`

`containsDOMElement()`

`hasDOMElement()`

`preventUpdates()`

`strictEq()`

`shallowEq()`

`shallowEqArray()`

`escapeHTMLAttribute()`

`escapeHTMLText()`

Stateless Components with `areEqual` hook

Using `requestAnimationFrame()` for Scheduling UI Updates