Proposal: Ignore private/protected members when doing inferrence

[chrisnicola]

TypeScript Version: 2.0.3 / nightly (2.1.0-dev.201xxxxx)

Code

A trivial example of the problem I'm referring to is in the Gist linked below.

https://gist.github.com/chrisnicola/918f1f0cf3ddf35df2a354514edbf15a

Expected behavior:

I actually expected that this already worked this way, but I was surprised to find out that twoj types are incompatible even if they differ in their private members, not just their public members. I'm not sure if there are valid issues with inferring two objects of different types to be compatible based solely on their public methods but I can't really see why they wouldn't be.

The lack of compatibility here leads to just declaring the object as any to get around the compiler error, which feels like a suboptimal solution.

Actual behavior:

Two types are only compatible if the requested type is a subset of the provided type inclusive of all the requested types private members. My expectation would be that only public members are actually considered.

Resoning:

There are many cases where inference is preferred to declaring and using interfaces or using inheritance. Test substitutes are one example. We often use concrete test substitutes rather than mocking or spying libraries in our tests. These substitutes have identical public interfaces to the concrete implementations but are, obviously, of a different type.

While declaring a common interface would be one solution, as we are testing services in Angular 2 interfaces are an inconvenience with Angular 2's @Injectable behaviour. Also I personally feel it should be unnecessary, it is again just extra ceremony and given the choice I would just stick with just declaring the substitute instance as any in the test instead.

[RyanCavanaugh]

This seems like a bad idea?

class Alpha {
  x: number = 0;
  private y: string = "me";
  equals(other: Alpha) {
    return this.y.toLowerCase() === other.y.toLowerCase();
  }
}

class Beta {
  x: number = 0;
  private y: number = 0;
  equals(other: Alpha) { return false; }
}

let a = new Alpha();
let b = new Beta();
a.equals(b); // OK according to types, but throws

[chrisnicola]

Yeah we discussed this problem amongst our group and I agree there are more edge cases than I originally considered. That being said I feel like there may be ways for the compiler to manage this while providing the usability/productivity benefit of not always needing to define a common interface class just to make this work.

In the case above this is only possible because you've passed an object of a type to another object of the same type. In that case could the compiler not flag that as a potential error, or even be so specific as to warn based on the fact that you have an actual usage of a private variable.

In other words, could the compiler, as a default, infer that a type is represented by it's public interface, but if in usage a part of the private interface is accessed, then it would infer it as being represented by its whole?

Another way might be some sort of annotation or way of using a type that implies you intend only its public interface.

My main motivation is that 90% of all usages involve only the public interface.

[appsforartists]

Would this make sense as a generic type, e.g. Public<Alpha> or ShapeOf<Alpha>, that would give you an interface that includes only the public aspects of Alpha?

In midicast, I've got a Lerna-esque monorepo. If I let each package have its own copy of RxJS, I get errors to the effect of "Observable is not compatible with Observable because they have different protected implementations of source". If I hoist the rxjs dependency to a single node_modules folder that is shared amongst them all, TypeScript stops finding types in sibling packages, and starts treating them as any.

[mhegazy]

If I hoist the rxjs dependency to a single node_modules folder that is shared amongst them all, TypeScript stops finding types in sibling packages, and starts treating them as any.

do you have path mapping entry for rxjs? can you share a repro?

[appsforartists]

Try:

git clone https://github.com/appsforartists/midicast.git
lerna run build
lerna bootstrap

and look at the type of messages in SongScanner.

[mhegazy]

I believe this is now addressed by #6496, can you give typescript@2.6 a try and let us know if you are still running into issues?

[appsforartists]

Worked. Thanks.

[typescript-bot]

Automatically closing this issue for housekeeping purposes. The issue labels indicate that it is unactionable at the moment or has already been addressed.