DEV Community

loading...

Narrowing the Type of Class Accessors

Benny Powers 🇮🇱🇨🇦
Coding is as much a matter of personal growth as it is of logic and control-flow. I keep patience, curiosity, & exuberance in the same toolbox as vim and git. *Opinions posted are my own*
・3 min read

Javascript class accessors let you define a function to get or set a particular value on an object.

Why would you want to do that? One popular reason is to run "side effects" in a setter. So for example say you want to make an HTTP request every time a user sets the options property on your object:

class Fetcher {
  constructor(public url = '/api', options?: RequestInit) {
    this.#options = options;
  }

  #options?: RequestInit;

  get options(): RequestInit {
    return this.#options;
  }

  set options(val: RequestInit) {
    this.#options = val;
    this.fetch();
  }

  fetch() {
    return fetch(this.url, this.options);
  }
}
Enter fullscreen mode Exit fullscreen mode

By defining the options property with an accessor pair, we can run our side effects (in this case, calling fetch) in the setter.

But now let's say we want to write a Poster class that only makes POST requests. It makes sense to extend Fetcher so that we don't duplicate our work. We want to narrow the type of options, however, to only allow options where the method is POST:

type PostInit = RequestInit & { method: 'POST' };

class Poster extends Fetcher {
  declare options: PostInit;
}
Enter fullscreen mode Exit fullscreen mode

What we're essentially saying with the declare keyword is "This class is exactly the same as it's parent, except that TypeScript should limit the options property to only accept PostInit objects". This should work, but...

'options' is defined as an accessor in class 'Fetcher', but is overridden here in 'Poster' as an instance property.

Oops! TypeScript complains. This change was introduced in 2020 in order to make TypeScript behave more like plain JavaScript. Indeed, TC39 decreed that class fields should have "define semantics" instead of "set semantics", which means that if we merely stripped these files of typescript syntax, our Poster class would break.

The debate over TC39's decision to use define semantics continues (even though the decision was made already), so we won't get deeper into it here, but when using typescript and "ambient declarations" like we did above with the declare keyword, we don't have the problem of our class field overriding the accessors, particularly if we don't have useDefineForClassFields turned on (which is probably a good choice anyways).

In fact, one of the three proposals which aim to fix this behaviour in typescript calls to allow fields to override accessors if they use the declare keyword.

But until that proposal, or a similar fix, is accepted, what can we do?

A Workaround Using Decorators

TypeScript field decorators have access to the class prototype. We can use one then to define our property with get and set functions. Since we're using a class private field for storage, and those are only available within a class body, let's define our decorator as a static method:

class Fetcher {
  /** @internal */
  static private o(proto: Fetcher, _: string) {
    Object.defineProperty(proto, 'options', {
      get() {
        return this.#options;
      },

      set(val) {
        this.#options = val;
        this.fetch();
      },
    });
  }

  #options?: RequestInit;

  @Fetcher.o options: RequestInit;

  constructor(public url = '/api', options?: RequestInit) {
    this.#options = options;
  }

  fetch() {
    return fetch(this.url, this.options);
  }
}
Enter fullscreen mode Exit fullscreen mode

If you squint, you can still see the original outline of the class, and this version does not error when we declare our narrow type on the subclass.

Really, this is a (kind of ugly) loophole to tide us over until TypeScript decides which proposal (if any) to adopt for this pattern. Here's hoping they don't close it before then.

Footnotes

Eagle-eyed readers might object to my example:

Poster should implement it's own setter to only allow POST at run time!

They'd be right to do so. We could come up with other examples where the options object doesn't have runtime validation, or where that validation is done in other methods. In those cases, it would be right-handy to declare the narrower type on the subclass.

Discussion (0)