Recently studied the mechanism of reflect, basically achieved all the api, but also passed all the test cases, so I want to write an article to record it. Reflect Metadata is a proposal of ES7 that is mainly used to add and read metadata at declaration time. Using reflection in Typescript requires the installation of a dependent reflect-metadata. Typescript has complete object-oriented support, and dependency injection technology (DI) has been widely used in frameworks such as Angular, Nest, and so on, like this.
class Service {}
class Controller {
constructor(private Service: Service) {}
}
The IOC framework takes the class that the Controller depends on and injects an instance or the class itself, where obtaining the dependency type leverages Reflect. Use metadata Design-time type annotations to get dependency types
function Injectable(): ClassDecorator {
return target => {
const metadata = Reflect.getMetadata("design:paramtypes", target)
console.log(metadata)
}
}
class Service {
constructor() {}
}
@Injectable() // 输出 [ Service() ]
class Controller {
constructor(private Service: Service) {}
}
Here we need to use the decorator to get it. You may have some questions here, metadata needs to be obtained first by defineMetadata and then in getMetadata, why can it be obtained directly here? The answer is that if tsconfig is enabled and emitDecoratorMetadata is true, the compiler will issue design metadata information. Let's take a look at the compiled js of ts (emitDecoratorMetadata enabled)
// 判断一下Reflect上是否有metadata函数,并赋给__metadata
var __metadata =
(this && this.__metadata) ||
function(k, v) {
if (typeof Reflect === "object" && typeof Reflect.metadata === "function")
return Reflect.metadata(k, v)
}
var Service = /** @class */ (function() {
function Service() {}
return Service
})()
var Controller = /** @class */ (function() {
function Controller(Service) {
this.Service = Service
}
Controller = __decorate(
[
Injectable(),
// 注意这里,使用design-paramtypes键定义了元数据,值是[Service]
__metadata("design:paramtypes", [Service])
],
Controller
)
return Controller
})()
function Injectable() {
return function(target) {
Reflect.getMetadata("design:paramtypes", target)
}
}
When emitDecoratorMetadata is turned on, the compiler automatically generates Design-time type annotations. From the previous example, it is easy to find that a Map should be maintained inside Reflect, and it should be WeakMap Sure enough, WeakMap is used inside the reflect-metadata library, and it comes with Polyfill (not only WeakMap, but also HashMap…). [[Metadata]] internal slot Reflect.ts#L685 Naive WeakMap shim Reflect.ts#L1725 Let's start to implement. Implement Reflect.defineMetadata First, create a WeakMap that will be responsible for maintaining the metadata for all classes and class instances, as well as solving the problem of automatic recycling.
const Metadata = new WeakMap<Object, Map<PropertyKey, MetadataMap>>()
This is a high-dimensional Map, and for each Object, i.e. class or class instance, it is associated with a Map. This Map is associated with the attribute key and the Map corresponding to the attribute, so it is a high-dimensional Map. The Metadata is a three-dimensional Map.
export function defineMetadata(
metadataKey: MetadataKey,
metadataValue: MetadataValue,
target: Object,
propertyKey: PropertyKey = DEFAULTKAY
) {
// 判断target类型,因为target将作为WeakMap的键,必须是对象类型
if (typeof target !== "object" && typeof target !== "function") {
throw new TypeError()
}
// 如果传入propertyKey,要求类型为string或symbol
if (propertyKey && !["string", "symbol"].includes(typeof propertyKey)) {
throw new TypeError()
}
// 从Metadata中获取target关联的Map,若没有就创建一个新的Map
const targetMetadata =
Metadata.get(target) || new Map<PropertyKey, MetadataMap>()
// 将targetMetadata再保存回Metadata中
Metadata.set(target, targetMetadata)
// 从targetMetadata中获取propertyKey关联的Map,若没有就创建一个新的Map
const metadataMap: MetadataMap = targetMetadata.get(propertyKey) || new Map()
// 将metadataMap再保存回targetMetadata中
targetMetadata.set(propertyKey, metadataMap)
// 设置元数据到metadataMap,键为metadataKey,值metadataValue
metadataMap.set(metadataKey, metadataValue)
}
Implement Reflect.getMetadata This api depends on getMetadataMap and getOwnMetadataMap. So do it backwards first.
export function getOwnMetadataMap(
target: Object,
propertyKey: PropertyKey = DEFAULTKAY
) {
// 判断target类型,因为target将作为WeakMap的键,必须是对象类型
if (typeof target !== "object" && typeof target !== "function") {
throw new TypeError()
}
// 从Metadata中获取target关联的Map,如果没有就返回undefined
const targetMetadata = Metadata.get(target)
if (!targetMetadata) return
// 从targetMetadata中获取propertyKey关联的Map,如果没有就返回undefined
const metadataMap = targetMetadata.get(propertyKey)
if (!metadataMap) return
// 返回metadataMap
return metadataMap
}
export function getMetadataMap(
target: Object,
propertyKey: PropertyKey = DEFAULTKAY
) {
// 如果自身有了propertyKey对应的map
if (Boolean(getOwnMetadataMap(target, propertyKey))) {
return getOwnMetadataMap(target, propertyKey)
}
// 去原型上找propertyKey对应的map,如果没有就返回undefined
const targetMetadata = Metadata.get(Object.getPrototypeOf(target))
if (!targetMetadata) return
// 从targetMetadata中获取propertyKey关联的Map,如果没有就返回undefined
const metadataMap = targetMetadata.get(propertyKey)
if (!metadataMap) return
// 返回metadataMap
return metadataMap
}
Implement getMetadata Use getMetadataMap to get the metadataMap associated with target, and then get the corresponding metadataValue according to metadataKey
export function getMetadata<T>(
metadataKey: MetadataKey,
target: Object,
propertyKey?: PropertyKey
): T {
// 根据propertyKey获取target的metadataMap,如果没有就返回undefined
const metadataMap = getMetadataMap(target, propertyKey)
if (!metadataMap) return
// 返回metadataKey对应的metadataValue
return metadataMap.get(metadataKey)
}
The same is true for getOwnMetadata, just consider getOwnMetadataMap Implement Reflect.getMetadataKeys Used to get all the metadata keys on the target It depends on getOwnMetadataKeys, so implement getOwnMetadataKeys first Implement getOwnMetadataKeys Get all the metadata keys of target itself
export function getOwnMetadataKeys(
target: Object,
propertyKey?: PropertyKey
): MetadataKey[] {
// 获取target身上与propertyKey关联的metadataMap,若没有返回空数组
const metadataMap = getOwnMetadataMap(target, propertyKey)
if (!metadataMap) return []
// metadataMap转为数组并返回
return Array.from(metadataMap.keys())
}
Implement getMetadataKeys Get metadataKeys of self, get metadataKeys of prototype, merge and return
export function getMetadataKeys(
target: Object,
propertyKey?: PropertyKey
): MetadataKey[] {
// 获取target自身与propertyKey关联的metadataKeys
const ownKeys = getOwnMetadataKeys(target, propertyKey)
// 获取target原型与propertyKey关联的metadataKeys
const protoKeys = getOwnMetadataKeys(
Object.getPrototypeOf(target),
propertyKey
)
// 返回结果
return [...ownKeys, ...protoKeys]
}
Implement Reflect.hasMetadata It is used to determine whether there is a corresponding metadataKey on target. Note that it is to determine whether there is key, not value, so it cannot be equivalent to Boolean (getMetadata), because metadataValue can be the equivalent of null and undefined. The test cases do a lot of test here.
export function hasMetadata(
metadataKey: MetadataKey,
target: Object,
propertyKey?: PropertyKey
) {
const metadataKeys = getMetadataKeys(target, propertyKey)
return metadataKeys.includes(metadataKey)
}
This is very simple. There is nothing to say. There is also hasOwnMetadata, which uses getOwnMetadataKeys to ok Implement Reflect.deleteMetadata
export function deleteMetadata(
metadataKey: MetadataKey,
target: Object,
propertyKey?: PropertyKey
) {
const metadataMap = getOwnMetadataMap(target, propertyKey)
if (!metadataMap) return false
return metadataMap.delete(metadataKey)
}
Implement Reflect.decorate This can get the idea from the test case.
// reflect-decorate.test.ts
it("DecoratorCorrectTargetInPipelineForFunctionOverload", () => {
let sent: Function[] = []
let A = function A(): void {}
let B = function B(): void {}
let decorators = [
(target: Function): any => {
sent.push(target)
return undefined
},
(target: Function): any => {
sent.push(target)
return undefined
},
(target: Function): any => {
sent.push(target)
return A
},
(target: Function): any => {
sent.push(target)
return B
}
]
let target = function(): void {}
Reflect.decorate(decorators, target)
expect(sent).toEqual([target, B, A, A])
})
The decorators given in the test is an array of four arrow functions, push the target parameter into sent, and then return a result The requirement is that after Reflect.decorate is executed, sent is [target, B, A, A] Combined with the name analysis of the test case, you see the Pipeline, which should think of reduce and reduceRight. Target is initialValue,decorators and reducers. Decorate A bit of type reducers, target is state, decorators are reducers
export function decorate(
decorators: (PropertyDecorator | MethodDecorator)[],
target: Object | Function,
propertyKey?: string | symbol,
attributes?: PropertyDescriptor
): PropertyDescriptor {
// 如果decorators是空数组则抛出TypeError
if (0 === decorators.length) {
throw new TypeError()
}
// target作为initialValue,遍历decorators,将上一个decorator的结果作为target传给下一个decorator
return decorators.reduceRight(
(target, decorator) => decorator(target, propertyKey, attributes) || target,
<any>target
)
}
Determine the type of Reflect and mix with native Reflect To avoid type conflicts, you first need to give the type of the native Reflect an alias, which is called IReflect.
export type IReflect = typeof Reflect
export const reflect = Reflect
And then mixed with Reflector.
export const Reflect: typeof Reflector & IReflect = Object.assign(
reflect,
Reflector
)
Because the properties of the native Reflect object cannot be enumerated, you can only merge to the native reflect
ps: the reflect-metadata on which I wrote the IOC framework has also been replaced with the current version, and demo can also run, which feels good. @ saber2pr/ioc