UNPKG

lexical

Version:

Lexical is an extensible text editor framework that provides excellent reliability, accessible and performance.

462 lines (461 loc) 19.3 kB
/** * Copyright (c) Meta Platforms, Inc. and affiliates. * * This source code is licensed under the MIT license found in the * LICENSE file in the root directory of this source tree. * */ import type { EditorConfig, Klass, KlassConstructor, LexicalEditor } from './LexicalEditor'; import type { BaseSelection, RangeSelection } from './LexicalSelection'; import { type DecoratorNode, ElementNode, NODE_STATE_KEY } from '.'; import { type NodeState } from './LexicalNodeState'; export type NodeMap = Map<NodeKey, LexicalNode>; /** * The base type for all serialized nodes */ export type SerializedLexicalNode = { /** The type string used by the Node class */ type: string; /** A numeric version for this schema, defaulting to 1, but not generally recommended for use */ version: number; [NODE_STATE_KEY]?: Record<string, unknown>; }; /** * Omit the children, type, and version properties from the given SerializedLexicalNode definition. */ export type LexicalUpdateJSON<T extends SerializedLexicalNode> = Omit<T, 'children' | 'type' | 'version'>; /** @internal */ export interface LexicalPrivateDOM { __lexicalTextContent?: string | undefined | null; __lexicalLineBreak?: HTMLBRElement | HTMLImageElement | undefined | null; __lexicalDirTextContent?: string | undefined | null; __lexicalDir?: 'ltr' | 'rtl' | null | undefined; __lexicalUnmanaged?: boolean | undefined; } export declare function $removeNode(nodeToRemove: LexicalNode, restoreSelection: boolean, preserveEmptyParent?: boolean): void; export type DOMConversion<T extends HTMLElement = HTMLElement> = { conversion: DOMConversionFn<T>; priority?: 0 | 1 | 2 | 3 | 4; }; export type DOMConversionFn<T extends HTMLElement = HTMLElement> = (element: T) => DOMConversionOutput | null; export type DOMChildConversion = (lexicalNode: LexicalNode, parentLexicalNode: LexicalNode | null | undefined) => LexicalNode | null | undefined; export type DOMConversionMap<T extends HTMLElement = HTMLElement> = Record<NodeName, (node: T) => DOMConversion<T> | null>; type NodeName = string; export type DOMConversionOutput = { after?: (childLexicalNodes: Array<LexicalNode>) => Array<LexicalNode>; forChild?: DOMChildConversion; node: null | LexicalNode | Array<LexicalNode>; }; export type DOMExportOutputMap = Map<Klass<LexicalNode>, (editor: LexicalEditor, target: LexicalNode) => DOMExportOutput>; export type DOMExportOutput = { after?: (generatedElement: HTMLElement | DocumentFragment | Text | null | undefined) => HTMLElement | Text | null | undefined; element: HTMLElement | DocumentFragment | Text | null; }; export type NodeKey = string; export declare class LexicalNode { ['constructor']: KlassConstructor<typeof LexicalNode>; /** @internal */ __type: string; /** @internal */ __key: string; /** @internal */ __parent: null | NodeKey; /** @internal */ __prev: null | NodeKey; /** @internal */ __next: null | NodeKey; /** @internal */ __state?: NodeState<this>; /** * Returns the string type of this node. Every node must * implement this and it MUST BE UNIQUE amongst nodes registered * on the editor. * */ static getType(): string; /** * Clones this node, creating a new node with a different key * and adding it to the EditorState (but not attaching it anywhere!). All nodes must * implement this method. * */ static clone(_data: unknown): LexicalNode; /** * Perform any state updates on the clone of prevNode that are not already * handled by the constructor call in the static clone method. If you have * state to update in your clone that is not handled directly by the * constructor, it is advisable to override this method but it is required * to include a call to `super.afterCloneFrom(prevNode)` in your * implementation. This is only intended to be called by * {@link $cloneWithProperties} function or via a super call. * * @example * ```ts * class ClassesTextNode extends TextNode { * // Not shown: static getType, static importJSON, exportJSON, createDOM, updateDOM * __classes = new Set<string>(); * static clone(node: ClassesTextNode): ClassesTextNode { * // The inherited TextNode constructor is used here, so * // classes is not set by this method. * return new ClassesTextNode(node.__text, node.__key); * } * afterCloneFrom(node: this): void { * // This calls TextNode.afterCloneFrom and LexicalNode.afterCloneFrom * // for necessary state updates * super.afterCloneFrom(node); * this.__addClasses(node.__classes); * } * // This method is a private implementation detail, it is not * // suitable for the public API because it does not call getWritable * __addClasses(classNames: Iterable<string>): this { * for (const className of classNames) { * this.__classes.add(className); * } * return this; * } * addClass(...classNames: string[]): this { * return this.getWritable().__addClasses(classNames); * } * removeClass(...classNames: string[]): this { * const node = this.getWritable(); * for (const className of classNames) { * this.__classes.delete(className); * } * return this; * } * getClasses(): Set<string> { * return this.getLatest().__classes; * } * } * ``` * */ afterCloneFrom(prevNode: this): void; static importDOM?: () => DOMConversionMap<any> | null; constructor(key?: NodeKey); /** * Returns the string type of this node. */ getType(): string; isInline(): boolean; /** * Returns true if there is a path between this node and the RootNode, false otherwise. * This is a way of determining if the node is "attached" EditorState. Unattached nodes * won't be reconciled and will ultimately be cleaned up by the Lexical GC. */ isAttached(): boolean; /** * Returns true if this node is contained within the provided Selection., false otherwise. * Relies on the algorithms implemented in {@link BaseSelection.getNodes} to determine * what's included. * * @param selection - The selection that we want to determine if the node is in. */ isSelected(selection?: null | BaseSelection): boolean; /** * Returns this nodes key. */ getKey(): NodeKey; /** * Returns the zero-based index of this node within the parent. */ getIndexWithinParent(): number; /** * Returns the parent of this node, or null if none is found. */ getParent<T extends ElementNode>(): T | null; /** * Returns the parent of this node, or throws if none is found. */ getParentOrThrow<T extends ElementNode>(): T; /** * Returns the highest (in the EditorState tree) * non-root ancestor of this node, or null if none is found. See {@link lexical!$isRootOrShadowRoot} * for more information on which Elements comprise "roots". */ getTopLevelElement(): ElementNode | DecoratorNode<unknown> | null; /** * Returns the highest (in the EditorState tree) * non-root ancestor of this node, or throws if none is found. See {@link lexical!$isRootOrShadowRoot} * for more information on which Elements comprise "roots". */ getTopLevelElementOrThrow(): ElementNode | DecoratorNode<unknown>; /** * Returns a list of the every ancestor of this node, * all the way up to the RootNode. * */ getParents(): Array<ElementNode>; /** * Returns a list of the keys of every ancestor of this node, * all the way up to the RootNode. * */ getParentKeys(): Array<NodeKey>; /** * Returns the "previous" siblings - that is, the node that comes * before this one in the same parent. * */ getPreviousSibling<T extends LexicalNode>(): T | null; /** * Returns the "previous" siblings - that is, the nodes that come between * this one and the first child of it's parent, inclusive. * */ getPreviousSiblings<T extends LexicalNode>(): Array<T>; /** * Returns the "next" siblings - that is, the node that comes * after this one in the same parent * */ getNextSibling<T extends LexicalNode>(): T | null; /** * Returns all "next" siblings - that is, the nodes that come between this * one and the last child of it's parent, inclusive. * */ getNextSiblings<T extends LexicalNode>(): Array<T>; /** * @deprecated use {@link $getCommonAncestor} * * Returns the closest common ancestor of this node and the provided one or null * if one cannot be found. * * @param node - the other node to find the common ancestor of. */ getCommonAncestor<T extends ElementNode = ElementNode>(node: LexicalNode): T | null; /** * Returns true if the provided node is the exact same one as this node, from Lexical's perspective. * Always use this instead of referential equality. * * @param object - the node to perform the equality comparison on. */ is(object: LexicalNode | null | undefined): boolean; /** * Returns true if this node logically precedes the target node in the * editor state, false otherwise (including if there is no common ancestor). * * Note that this notion of isBefore is based on post-order; a descendant * node is always before its ancestors. See also * {@link $getCommonAncestor} and {@link $comparePointCaretNext} for * more flexible ways to determine the relative positions of nodes. * * @param targetNode - the node we're testing to see if it's after this one. */ isBefore(targetNode: LexicalNode): boolean; /** * Returns true if this node is an ancestor of and distinct from the target node, false otherwise. * * @param targetNode - the would-be child node. */ isParentOf(targetNode: LexicalNode): boolean; /** * Returns a list of nodes that are between this node and * the target node in the EditorState. * * @param targetNode - the node that marks the other end of the range of nodes to be returned. */ getNodesBetween(targetNode: LexicalNode): Array<LexicalNode>; /** * Returns true if this node has been marked dirty during this update cycle. * */ isDirty(): boolean; /** * Returns the latest version of the node from the active EditorState. * This is used to avoid getting values from stale node references. * */ getLatest(): this; /** * Returns a mutable version of the node using {@link $cloneWithProperties} * if necessary. Will throw an error if called outside of a Lexical Editor * {@link LexicalEditor.update} callback. * */ getWritable(): this; /** * Returns the text content of the node. Override this for * custom nodes that should have a representation in plain text * format (for copy + paste, for example) * */ getTextContent(): string; /** * Returns the length of the string produced by calling getTextContent on this node. * */ getTextContentSize(): number; /** * Called during the reconciliation process to determine which nodes * to insert into the DOM for this Lexical Node. * * This method must return exactly one HTMLElement. Nested elements are not supported. * * Do not attempt to update the Lexical EditorState during this phase of the update lifecycle. * * @param _config - allows access to things like the EditorTheme (to apply classes) during reconciliation. * @param _editor - allows access to the editor for context during reconciliation. * * */ createDOM(_config: EditorConfig, _editor: LexicalEditor): HTMLElement; /** * Called when a node changes and should update the DOM * in whatever way is necessary to make it align with any changes that might * have happened during the update. * * Returning "true" here will cause lexical to unmount and recreate the DOM node * (by calling createDOM). You would need to do this if the element tag changes, * for instance. * * */ updateDOM(_prevNode: unknown, _dom: HTMLElement, _config: EditorConfig): boolean; /** * Controls how the this node is serialized to HTML. This is important for * copy and paste between Lexical and non-Lexical editors, or Lexical editors with different namespaces, * in which case the primary transfer format is HTML. It's also important if you're serializing * to HTML for any other reason via {@link @lexical/html!$generateHtmlFromNodes}. You could * also use this method to build your own HTML renderer. * * */ exportDOM(editor: LexicalEditor): DOMExportOutput; /** * Controls how the this node is serialized to JSON. This is important for * copy and paste between Lexical editors sharing the same namespace. It's also important * if you're serializing to JSON for persistent storage somewhere. * See [Serialization & Deserialization](https://lexical.dev/docs/concepts/serialization#lexical---html). * * */ exportJSON(): SerializedLexicalNode; /** * Controls how the this node is deserialized from JSON. This is usually boilerplate, * but provides an abstraction between the node implementation and serialized interface that can * be important if you ever make breaking changes to a node schema (by adding or removing properties). * See [Serialization & Deserialization](https://lexical.dev/docs/concepts/serialization#lexical---html). * * */ static importJSON(_serializedNode: SerializedLexicalNode): LexicalNode; /** * Update this LexicalNode instance from serialized JSON. It's recommended * to implement as much logic as possible in this method instead of the * static importJSON method, so that the functionality can be inherited in subclasses. * * The LexicalUpdateJSON utility type should be used to ignore any type, version, * or children properties in the JSON so that the extended JSON from subclasses * are acceptable parameters for the super call. * * If overridden, this method must call super. * * @example * ```ts * class MyTextNode extends TextNode { * // ... * static importJSON(serializedNode: SerializedMyTextNode): MyTextNode { * return $createMyTextNode() * .updateFromJSON(serializedNode); * } * updateFromJSON( * serializedNode: LexicalUpdateJSON<SerializedMyTextNode>, * ): this { * return super.updateFromJSON(serializedNode) * .setMyProperty(serializedNode.myProperty); * } * } * ``` **/ updateFromJSON(serializedNode: LexicalUpdateJSON<SerializedLexicalNode>): this; /** * @experimental * * Registers the returned function as a transform on the node during * Editor initialization. Most such use cases should be addressed via * the {@link LexicalEditor.registerNodeTransform} API. * * Experimental - use at your own risk. */ static transform(): ((node: LexicalNode) => void) | null; /** * Removes this LexicalNode from the EditorState. If the node isn't re-inserted * somewhere, the Lexical garbage collector will eventually clean it up. * * @param preserveEmptyParent - If falsy, the node's parent will be removed if * it's empty after the removal operation. This is the default behavior, subject to * other node heuristics such as {@link ElementNode#canBeEmpty} * */ remove(preserveEmptyParent?: boolean): void; /** * Replaces this LexicalNode with the provided node, optionally transferring the children * of the replaced node to the replacing node. * * @param replaceWith - The node to replace this one with. * @param includeChildren - Whether or not to transfer the children of this node to the replacing node. * */ replace<N extends LexicalNode>(replaceWith: N, includeChildren?: boolean): N; /** * Inserts a node after this LexicalNode (as the next sibling). * * @param nodeToInsert - The node to insert after this one. * @param restoreSelection - Whether or not to attempt to resolve the * selection to the appropriate place after the operation is complete. * */ insertAfter(nodeToInsert: LexicalNode, restoreSelection?: boolean): LexicalNode; /** * Inserts a node before this LexicalNode (as the previous sibling). * * @param nodeToInsert - The node to insert before this one. * @param restoreSelection - Whether or not to attempt to resolve the * selection to the appropriate place after the operation is complete. * */ insertBefore(nodeToInsert: LexicalNode, restoreSelection?: boolean): LexicalNode; /** * Whether or not this node has a required parent. Used during copy + paste operations * to normalize nodes that would otherwise be orphaned. For example, ListItemNodes without * a ListNode parent or TextNodes with a ParagraphNode parent. * * */ isParentRequired(): boolean; /** * The creation logic for any required parent. Should be implemented if {@link isParentRequired} returns true. * * */ createParentElementNode(): ElementNode; selectStart(): RangeSelection; selectEnd(): RangeSelection; /** * Moves selection to the previous sibling of this node, at the specified offsets. * * @param anchorOffset - The anchor offset for selection. * @param focusOffset - The focus offset for selection * */ selectPrevious(anchorOffset?: number, focusOffset?: number): RangeSelection; /** * Moves selection to the next sibling of this node, at the specified offsets. * * @param anchorOffset - The anchor offset for selection. * @param focusOffset - The focus offset for selection * */ selectNext(anchorOffset?: number, focusOffset?: number): RangeSelection; /** * Marks a node dirty, triggering transforms and * forcing it to be reconciled during the update cycle. * * */ markDirty(): void; /** * @internal * * When the reconciler detects that a node was mutated, this method * may be called to restore the node to a known good state. */ reconcileObservedMutation(dom: HTMLElement, editor: LexicalEditor): void; } /** * Insert a series of nodes after this LexicalNode (as next siblings) * * @param firstToInsert - The first node to insert after this one. * @param lastToInsert - The last node to insert after this one. Must be a * later sibling of FirstNode. If not provided, it will be its last sibling. */ export declare function insertRangeAfter(node: LexicalNode, firstToInsert: LexicalNode, lastToInsert?: LexicalNode): void; export {};