Typescript: MethodDecorator gets TS2322/TS2315

[Norgerman]

TypeScript Version: 2.4.2
Code

export function route(method: string, path: string): MethodDecorator {
    return (target: any, name: string, descriptor: TypedPropertyDescriptor<Function>) => {
        routeManager.regeisterRoute({
            constructor: target.constructor,
            function: descriptor.value,
            method,
            path
        });
    }
}

export function route(method: string, path: string): MethodDecorator<Function> {
    return (target: any, name: string, descriptor: TypedPropertyDescriptor<Function>) => {
        routeManager.regeisterRoute({
            constructor: target.constructor,
            function: descriptor.value,
            method,
            path
        });
    }
}



md5-f96297b63fb35a729f69dab9d2abc379



error TS2322: Type '(target: any, name: string, descriptor: TypedPropertyDescriptor<Function>) => void' is not assignable to type 'MethodDecorate
or'.
  Types of parameters 'descriptor' and 'descriptor' are incompatible.
    Type 'TypedPropertyDescriptor<T>' is not assignable to type 'TypedPropertyDescriptor<Function>'.
      Type 'T' is not assignable to type 'Function'.



md5-263ea59b36b5a390ecb20970d96a9929



error TS2315: Type 'MethodDecorator' is not generic.

[ikatyang]

For the first one, it is caused by stricter generic checks (#16368).

It seems you have to define a CustomMethodDecorator first, since there is no generic on MethodDecorator, and its declaration requires a non-constraint generic.

Or just enable the --noStrictGenericChecks option.

---

For the second one, there is indeed no generic on MethodDecorator, the error message seems a little weird.

(lib.d.ts)

declare type MethodDecorator = <T>(target: Object, propertyKey: string | symbol, descriptor: TypedPropertyDescriptor<T>) => TypedPropertyDescriptor<T> | void;

[Norgerman]

CustomMethodDecorator or --noStrictGenericChecks will works, but I think it should infer T as Function in case 1, or make MethodDecorator generic in case 2.

[ikatyang]

MethodDecorator is defined as <T>(...something: SomeType[]): SomeOutput<T>;

So that it'll accept all kinds of T without any constraint, but in your case 1 it only accept Function, it is indeed not compatible from stricter perspective.

NOTE: With --noStrictGenericChecks enabled, all generics will be considered any before comparing.

---

make MethodDecorator generic in case 2.

I'm not familiar with that MethodDecorator type, it'd be better to raise another issue for that things.

[aluanhaddad]

There is no requirements that decorators return one of the predefined decorator types in lib.d.ts.

I would strongly advise against enabling --noStrictGenericChecks and instead using the following (which actually has a richer type but passes the checker thanks to inference.

declare const routeManager: {
  regeisterRoute: (config: {
    constructor?: Function,
    function?: Function,
    method: string,
    path: string
  }) => void;
};

export function route(method: string, path: string) {
  return <T extends object,
        K extends keyof T,
        R extends Promise<any>>
      (target: T, name: keyof T, descriptor: TypedPropertyDescriptor<(...args: any[]) => R>) => {
        routeManager.regeisterRoute({
        constructor: target.constructor,
        function: descriptor.value,
        method,
        path
      });
  };
}

class A {
  @route('GET', 'api/values') m() {
    return Promise.resolve(['a', 'b']);
  }
}

Not only will this pass the type checker since it is a valid decorator type, but the decorator will validate that the decorated method actually returns a Promise for a resource (just an example).

Trying to exactly match the predefined decorator types would preclude this very useful typechecking behavior.

[Norgerman]

Well, it works. Just like what @ikatyang says to define a CustomMethodDecorator.

[aluanhaddad]

@Norgerman Ah, I missed that comment. Sorry for being redundant.

[petispaespea]

@Norgerman Could you please share your way of implementing CustomMethodDecorator?

[Norgerman]

@petispaespea see this comment