Hacker News Reader

Comment by Klathmon | Hacker News Reader

If `example.myVar = 500` can throw errors, then is the private field really private? Since now it's existence can break the public interface, and changing it to `myOtherVar` during a refactor can break existing code.

That removes a major benefit of private fields (which is to be able to change implementation details without breaking the public interface)

I don't think anyone really loves the syntax, but there doesn't seem to be much of a better way of doing it.

View on HN

What happens if you do `example.#myVar = 500`? Does it throw an error? Wouldn't it be the same situation?

as of right now, `example.#myVar` isn't valid syntax, which is the reason that the # sigil was chosen.

It's not possible to have current code that works like that, so defining new properties for that code (that it can only be used inside the class as a private field) isn't breaking anything.

In other words, outside of the class implementation, `example.#myVar` is always an error, regardless of the existence of a private field or not, and it will always be an error with or without this proposal. And inside the class it's only legal if the private field was defined in the class.

The difference is that when you type example.myVar = 400 you are setting a property names "myVar" on the object to 400. When you type example.#myVar you are trying to set example's private field to 400. The intentions are distinct and it is not costly to determine at the call site what you're trying to do.

What if I have private values I want to access programmatically? Ie. fields = ['#foo', '#baz', '#bar']; fields.forEach(field => { this[field] = frobnicate(field) }); ?

I don't know what the spec's solution for that would be, but what you've typed would not work. Specifically because when I say `this.#foo` it is not a property on the object named "#foo". It is just the syntax for looking up a private field referenced by the literal #foo.

The spec's solution is to not solve that problem ;)

Dynamic property access isn't allowed for private fields, so it kind of side steps that whole issue.

Correction: What happens if you do `example['#myVar'] = 500`? Does it throw an error? Wouldn't it be the same situation?

dynamic access to private fields is considered a "non-goal" of sorts, so `example['#myVar']` will act the same as it does now, it will always refer to the public property of that name.

https://github.com/tc39/proposal-class-fields/blob/master/PR...

Then what about inside the class? What about `this['#myVar']`? Is it different from `this.#myVar`? If they are different it's very inconsistent and IMHO a lot worse. If they are the same, it's more confusing and still worse than the alternative suggested... what a mess.

On one hand I follow a more functional programming style so this won't affect me but on the other I worry about trying to understand other people's code (especially very serious programming [tm]).

`this['#myVar']` will access the public field named `#myVar`, so it's the same as the `example` method.

>If they are different it's very inconsistent and IMHO a lot worse.

It is called out in the link I posted above as a "downside", and I agree that it sucks, but there aren't any other good options for this without breaking compatibility with current code.

>I worry about trying to understand other people's code

Luckily there is one big benefit to this that makes the impact of the weirdness surrounding the naming much less awful. Because of the "hard encapsulation" that the private fields have, they are literally and explicitly un-viewable, un-accessible, and un-usable outside of the class body.

So the amount of complex meta-programming and many of the reasons you'd want dynamic property access in the first place (iterating, monkey-patching, using arbitrary strings as keys, computed keys, etc...) aren't applicable, and if you really want those features but also with privacy, you can nest a plain object as a private field and work with it that way just the same.

So it may have some gotchas, but because the proposal is intentionally restricted as much as possible, and it explicitly disallows a lot of the metaprogramming that can be common and it only allows a very simple lookup syntax, the weirdness is very well contained, and at worst will be confined to between the opening and closing brace of a `class` statement.

That's the thing, current code is not using the `private` keyword since it's not valid, so it would not break anything and would be totally retro-compatible. `example.myVar` will still work whether or not it's possible to create a private prop called `myVar`, because it's just not created now since you cannot do `private myVar` now. Now, I do agree that it might be confusing in the future as well and might be not future-compatible, but it is for sure retro-compatible...

But then what happens if you have `private myVar` and you need to access the public `myVar` from within the class?

That's not an extreme hypothetical, it was a real example from someone trying to use this feature. They had tried to emulate private fields with an underscore before, people started using the private variables, and now they want to refactor while keeping the current "private-turned-public" variable working the same.

I'm sure you can come up with answers for most of those things, but will they really be easier to understand than the current proposal? I'd hate if I have to look around in the class any time `this.anything` happens to see if it's private, public, shadowed, a getter/setter, etc...

`this.#myVar` is explicit, it's obvious, and even if it's ugly as a sin, you know it's not doing something "normal" right away.

If `myVar` is defined as private then it cannot be re-defined as public... It seems simple. It's the same as normal variables: what happens if you define myVar as `const` and then you want to redefine it as `var`? Well, you cannot, you'd have to work around it instead.

But then the private variables can impact code outside of the class, and and they are no longer truly private.

Imagine you are using a library and extending it adding a few properties for usage in your app, and one patch upgrade your app breaks because the `example.tag` stops working because the library decided to add `this.tag` as a private variable.

Imagine React decides to use this, and they add a private `this.internalState`, but all react components are extended from that. So now all react components can't use `this.internalState` and if the react team ever changes the name of the private variables, then it is a breaking change and can cause code to stop working.

You can say "well don't do that" all you want, but a good portion of the web uses techniques like that, and breaking code because you feel they are doing it wrong doesn't make a healthy ecosystem.

Ah I see, that makes a lot of sense, thanks for the clarification. I still think it'd have been a fair compromise anyway, but it's not ideal and not easy as I thought.

> then is the private field really private

What are some of the other languages that interpret "private" as never exposing the existence of that field to a public consumer? Most languages I've used don't expose the _value_ of the field...

I'm honestly not sure, but even taken in isolation it's a convincing argument to me.

See [0] for some info straight from the proposal, but the gist is doing it this way makes it so that you never have to worry about name collisions for private fields. Subclassing, superclassing, and object "monkey patching" all can continue to work as-is with no changes, and none of them have to worry about the existence of private fields complicating implementations, and the classes with private fields can feel free to change and rename them as much as possible, even within patch versions of libraries, since they cannot impact any code outside of the class under any circumstances. That's a pretty powerful guarantee to give which GREATLY simplifies working with them.

If I want to extend the react component class, I don't need to worry that my `#component` private field is going to break if an update to the react component class adds their own `#component` field, or if a user of my library tries to do something like this:

    const fancyObj = new FancyObj()
    fancyObj.tag = Symbol('tag')

Which I'll admit to having used in some cases where I want to tag a bunch of otherwise identical objects which will be thrown into an array and then be able to easily pluck out the ones I tagged later. Without true encapsulation, that code above would break if they changed the private interface to use the `.tag` field internally, and I would have no idea because it is consitered a private interface and wouldn't need to be documented or maintained across versions.

[0] https://github.com/tc39/proposal-class-fields/blob/master/PR...

Huh, they just took all the issues with inheritance and made them more confusing!

There is nothing addressed here that can't be solved with composition. This seems like an odd step for an aspiring functional language to take...

But JS isn't aspiring to be a functional language, it's adding new syntax and changing itself to better solve the problems that it is currently being used to solve.

Some of that is pushing it in a functional direction, some in a "classic OOP" direction.

Much like classes in the first place, this is a feature added to the language to stop the proliferation of stop-gap solutions and standardize on one implementation for most.

Sure, I won't be using this feature, just like I won't be using classes at all in most cases (React components being one exception), but many will, and if this isn't added, they will continue trying to solve problems with solutions which are more complicated, slower, and buggier than this. All the talk about how it's not the "right" solution doesn't matter.

To use an over-the-top metaphor: People are jumping out of planes, and you aren't going to stop them, so the very least you can do is give them a well tested parachute and a map of where to safely land.

This is how most object-oriented languages work, people just don't realize it. Here's some Java code:

    class Base {
        private int field;
        
        Base() {
            this.field = 1;
        }
        
        void baseMethod() {
            System.out.println("Base.field = " + field);
        }
    }
    
    class Derived extends Base {
        private int field;
        
        Derived() {
            this.field = 2;
        }
        
        void derivedMethod() {
            System.out.println("Derived.field = " + field);
        }
    }
    
    public static void main(String[ ] args) {
        Derived d = new Derived();
        d.baseMethod();
        d.derivedMethod();
    }

It prints:

    Base.field = 1
    Derived.field = 2

This is not how Ruby and Python works, I believe. I guess it's debatable which paradigm JavaScript is more similar to.

Python name-mangles properties that start with "__" to emulate privacy. Run:

    class Base:
      def __init__(self):
        self.__field = 1

      def baseMethod(self):
        print("baseMethod " + str(self.__field))

    class Derived(Base):
      def __init__(self):
        super().__init__()
        self.__field = 2

      def derivedMethod(self):
        print("barMethod " + str(self.__field))

    d = Derived()
    d.baseMethod()
    d.derivedMethod()

And it prints:

    baseMethod 1
    barMethod 2

Ruby doesn't really have fields that are private to subclasses, but Ruby also embraces monkey-patching, so that's not entirely surprising.

Reply on news.ycombinator.com