slate源码解析（三）- 定位

接口定义

能够对于文字、段落乃至任何元素的精准定位 并做出增删改查，都是在开发一款富文本编辑器时一项最基本也是最重要的功能之一。让我们先来看看Slate中对于如何在文档树中定位元素是怎么定义的[源码]：

/**
 * The `Location` interface is a union of the ways to refer to a specific
 * location in a Slate document: paths, points or ranges.
 *
 * Methods will often accept a `Location` instead of requiring only a `Path`,
 * `Point` or `Range`. This eliminates the need for developers to manage
 * converting between the different interfaces in their own code base.
 */
export type Location = Path | Point | Range

Location是一个包含了Path、Point及Range的联合类型，代指了Slate中所有关于“定位”的定位，同时也方便了开发。例如在几乎所有的Transforms方法中，都可以通过传递Location参数来决定Transforms方法需要应用到文档中的哪些位置上[链接]。

All transforms support a parameter options. This includes options specific to the transform, and general NodeOptions to specify which Nodes in the document that the transform is applied to.

interface NodeOptions {
  at?: Location
  match?: (node: Node, path: Location) => boolean
  mode?: 'highest' | 'lowest'
  voids?: boolean
}

Path

Path是三个中最基本的概念，也是唯一一个不可拓展的类型。

/**
 * `Path` arrays are a list of indexes that describe a node's exact position in
 * a Slate node tree. Although they are usually relative to the root `Editor`
 * object, they can be relative to any `Node` object.
 */
export type Path = number[]

Path类型就是一个数组，用来表示Slate文档树中自根节点root到指定node的绝对路径。我们以下边的示例来演示下各个node所代表的路径：

const initialValue: Descendant[] = [
    // path: [0]
    {
        type: 'paragraph',
        children: [
            { text: 'This is editable ' },		// path: [0, 0]
            { text: 'rich text!', bold: true }	// path: [0, 1]
        ]
    },
    // path: [1]
    {
        type: 'paragraph',
        children: [
            { text: 'It\' so cool.' }	// path: [1, 0]
        ]
    }
]

虽然Path所代表的路径通常是以顶层Editor作为root节点的，但也会有其他情况，比如由Node提供的get方法中传入的Path参数则代表的是相对路径[源码]：

/**
 * Get the descendant node referred to by a specific path. If the path is an
 * empty array, it refers to the root node itself.
 */
get(root: Node, path: Path): Node {
    let node = root
    for (let i = 0; i < path.length; i++) {
        const p = path[i]
        if (Text.isText(node) || !node.children[p]) {
            throw new Error(
                `Cannot find a descendant at path [${path}] in node: ${Scrubber.stringify(
                    root
                )}`
            )
        }
        node = node.children[p]
    }
    return node
}

Point

Point是在Path的基础上封装而来的概念：

/**
 * `Point` objects refer to a specific location in a text node in a Slate
 * document. Its path refers to the location of the node in the tree, and its
 * offset refers to the distance into the node's string of text. Points can
 * only refer to `Text` nodes.
 */
export interface BasePoint {
  path: Path
  offset: number
}
export type Point = ExtendedType<'Point', BasePoint>

用于定位单个字符在文档中的位置；先用Path定位到字符所属的Node，再根据offset字段信息精确到字符是在该Node的text文本中的偏移量。

我们仍然以前面的示例做说明，如果想要将光标位置定位到第一句话"This is editable rich text!"的感叹号之后，其Point值为：

const initialValue: Descendant[] = [
    // path: [0]
    {
        type: 'paragraph',
        children: [
            { text: 'This is editable ' },		// path: [0, 0]
            { text: 'rich text!', bold: true }	// { path: [0, 1], offset: 10 }
        ]
    },
    // path: [1]
    {
        type: 'paragraph',
        children: [
            { text: 'It\' so cool.' }	// path: [1, 0]
        ]
    }
]

Range

最后一个Range则是再在Point基础上延伸封装而来的概念：

/**
 * `Range` objects are a set of points that refer to a specific span of a Slate
 * document. They can define a span inside a single node or a can span across
 * multiple nodes.
 */
export interface BaseRange {
  anchor: Point
  focus: Point
}
export type Range = ExtendedType<'Range', BaseRange>

它代表的是一段文本的集合；包含有两个Point类型的字段anchor和focus。看到这，应该能发现Slate中Range的概念其实与DOM中的Selection对象是一样的，anchor和focus分别对应原生Selection中的锚点anchorNode和焦点focusNode；这正是Slate对于原生Selection做的抽象，使之在自身API中更方便地通过光标选区来获取文档树中的内容。

我们在上一篇文章中有提到过的Editor.selection是一个Selections类型，其本身就是一个Range类型，专门用来指定编辑区域中的光标位置[源码]：

export type BaseSelection = Range | null
export type Selection = ExtendedType<'Selection', BaseSelection>

Refs

当我们需要长期追踪某些Node时，可以通过获取对应Node的Path/Point/Range值并保存下来以达到目的。但这种方式存在的问题是，在Slate文档树经过insert、remove等操作后，原先的Path/Point/Range可能会产生变动或者直接作废掉。

Refs的出现就是为了解决上述问题。

三者的ref的定义分别在slate/src/interfaces/下的path-ref.ts、point-ref.ts和range-ref.ts文件中：

/**
 * `PathRef` objects keep a specific path in a document synced over time as new
 * operations are applied to the editor. You can access their `current` property
 * at any time for the up-to-date path value.
 */
export interface PathRef {
  current: Path | null
  affinity: 'forward' | 'backward' | null
  unref(): Path | null
}
/**
 * `PointRef` objects keep a specific point in a document synced over time as new
 * operations are applied to the editor. You can access their `current` property
 * at any time for the up-to-date point value.
 */
export interface PointRef {
  current: Point | null
  affinity: TextDirection | null
  unref(): Point | null
}
/**
 * `RangeRef` objects keep a specific range in a document synced over time as new
 * operations are applied to the editor. You can access their `current` property
 * at any time for the up-to-date range value.
 */
export interface RangeRef {
  current: Range | null
  affinity: 'forward' | 'backward' | 'outward' | 'inward' | null
  unref(): Range | null
}

都包含以下三个字段：

• current：同React ref用法一样，用current字段存储最新值
• affinity：当前定位所代表的节点 在文档树变动时如果受到影响的话，所采取的调整策略
• unref：卸载方法；彻底删除当前的ref确保能够被引擎GC掉

另外我们先来看下各种Refs在Slate存储的方式，跳到slate/src/utils/weak-maps.ts中[源码]：

export const PATH_REFS: WeakMap<Editor, Set<PathRef>> = new WeakMap()
export const POINT_REFS: WeakMap<Editor, Set<PointRef>> = new WeakMap()
export const RANGE_REFS: WeakMap<Editor, Set<RangeRef>> = new WeakMap()

可以看到Refs的存储区在一个Set数据结构中的；而对于不同Editor下Set<xxxRef>的存储则是放在哈希表WeakMap中的，使其不会影响到GC（WeakMap同Map原理一样，都是ES6之后新出的哈希数据结构，与Map不同点在于其持有的引用算作弱引用）。

生成三种Ref以及获取相应Refs的方法定义在EditorInterface接口上[源码]：

export interface EditorInterface {
    // ...
    pathRef: (
    	editor: Editor,
    	path: Path,
    	options?: EditorPathRefOptions
  	) => PathRef
    pointRef: (
    	editor: Editor,
    	point: Point,
    	options?: EditorPointRefOptions
  	) => PointRef
    rangeRef: (
    	editor: Editor,
    	range: Range,
    	options?: EditorRangeRefOptions
  	) => RangeRef
    pathRefs: (editor: Editor) => Set<PathRef>
    pointRefs: (editor: Editor) => Set<PointRef>
    rangeRefs: (editor: Editor) => Set<RangeRef>
}

Path、Point及Range三者的实现逻辑都差不多，下面就以Path为例作介绍。

pathRef：

  /**
   * Create a mutable ref for a `Point` object, which will stay in sync as new
   * operations are applied to the editor.
   */
  pointRef(
    editor: Editor,
    point: Point,
    options: EditorPointRefOptions = {}
  ): PointRef {
    const { affinity = 'forward' } = options
    const ref: PointRef = {
      current: point,
      affinity,
      unref() {
        const { current } = ref
        const pointRefs = Editor.pointRefs(editor)
        pointRefs.delete(ref)
        ref.current = null
        return current
      },
    }
    const refs = Editor.pointRefs(editor)
    refs.add(ref)
    return ref
  }

实现逻辑非常简单，就是根据传入的参数放入ref对象中，并添加卸载方法unref。然后通过pathRefs拿到对应的Set，讲当前的ref对象添加进去。unref方法中实现的则是相反的操作：通过pathRefs拿到对应的Set后，将当前ref对象移除掉，然后再把ref.current的值置空。

pathRefs：

  /**
   * Get the set of currently tracked path refs of the editor.
   */
  pathRefs(editor: Editor): Set<PathRef> {
    let refs = PATH_REFS.get(editor)
    if (!refs) {
      refs = new Set()
      PATH_REFS.set(editor, refs)
    }
    return refs
  }

代码非常简短，类似懒加载的方式做Set的初始化，然后调用get方法获取集合后返回。

Ref同步

前一篇文章我们提到过，用于修改内容的单个Transform方法会包含有多个Operation；Operation则是Slate中的原子化操作。而在Slate文档树更新之后，解决Ref同步更新的方式就是：在执行了任意Operation之后，对所有的Ref根据执行的Operation类型做相应的调整。

看到create-editor.ts中的apply方法，该方法是所有Operation执行的入口[源码]：

apply: (op: Operation) => {
    for (const ref of Editor.pathRefs(editor)) {
        PathRef.transform(ref, op)
    }
    for (const ref of Editor.pointRefs(editor)) {
        PointRef.transform(ref, op)
    }
    for (const ref of Editor.rangeRefs(editor)) {
        RangeRef.transform(ref, op)
    }
    // ...
}

在apply方法的最开头就是三组for of循环，对所有的Ref执行对应的Ref.transform方法并传入当前执行的Operation。

同样以Path为例，看下path-ref.ts中的PathRef.transform方法[源码]：

export const PathRef: PathRefInterface = {
  /**
   * Transform the path ref's current value by an operation.
   */
  transform(ref: PathRef, op: Operation): void {
    const { current, affinity } = ref
    if (current == null) {
      return
    }
    const path = Path.transform(current, op, { affinity })
    ref.current = path
    if (path == null) {
      ref.unref()
    }
  }
}

将当前ref中的数据和Operation作为参数，传递给对应的Path.transform，返回更新后的path值并赋值给ref.current。如果path为空，则说明当前ref指代的位置已经失效，调用卸载方法unref。

至于Path.transform的细节在本篇就不展开了，我们留到后续的Transform篇再统一讲解: )