Feature request: Add CaseIterable implementation to generated enum from a kotlin sealed class that contains objects

[corrado4eyes]

When I try to loop over an enum using Skie.KotlinModule.SealedClassEnumName.__Sealed.allCases allCases won't be found since the generated enum is not implementing CaseIterable protocol

[TadeasKriz]

Hey @corrado4eyes, thanks for the idea. In 0.5.0 we've added Hashable support for the sealed class enums, but we didn't think about adding CaseIterable. It'd have the same limitations as the Hashable (see https://skie.touchlab.co/features/sealed#hashable-conformance), would that still make sense for your use case?

Could you also share your usecase for CaseIterable for sealed classes? How do you currently deal with it, and on Android? Thanks!

[corrado4eyes]

Hi @TadeasKriz I forked the repo and gave it a try by adding superType list the sirBuiltins.Swift.CaseIterable.defaultType, but I found out that probably is not possible.
On iOS when I build the target that uses SKIE I get a an error on the generated __Sealed class type 'Skie.KotlinModule.SealedClassEnumName.__Sealed' does not conform to protocol 'CaseIterable'. That is because each case in the generated enum has an associated value (which is the respective Kotlin sealed class type) and therefore in these cases you need to implement the allCases property yourself.

My use case is to be able to loop over a sealed class and when I put the value inside a switch statement I want it to be exhaustive.
In my case I wanted to expose to swift N "TestCases" and implement them on a UITest target by using a switch statement inside a forEach loop. Currently I had some duplication on Kotlin where I have an enum class and a sealed class that match 1:1 with each other.

At the moment from kotlin I expose an allCases property which is a List<MyEnum> from the enum companion, that allows me to go through the enum cases exhaustively and I do what I want to do with the sealed classes inside that specific case.

To be honest at the moment I didn't put that much thought on Android because I could use reflekt or kotlin-reflect to get a list of the subclasses of a sealed class. Also on android the sealed class inside a when is already exhaustive

[FilipDolnik]

Hi! If I understand the problem correctly, we can add support for CaseIterable, but only in cases where all children of the sealed class are objects. Would that be enough for your use case?

The problem with regular classes is that you can have multiple different instances for each of those classes (with different values), which makes it impossible to implement the allCases property.

[corrado4eyes]

Hi @FilipDolnik, even in case you have only objects that won't be working. Also that doesn't fit my use case because some of those TestCase that I expose have different parameters

If I understand the problem correctly, we can add support for CaseIterable, but only in cases where all children of the sealed class are objects.

For the following kotlin sealed class, this is the generate swift enum

sealed class OnlyObjectsSealedClass {
    object ObjectOne : OnlyObjectsSealedClass()
    object ObjectTwo : OnlyObjectsSealedClass()
    object ObjectThree : OnlyObjectsSealedClass()
}

enum OnlyObjectsSealedClass {

    public typealias __Kotlin = shared.OnlyObjectsSealedClass

    @frozen
    public enum __Sealed : Swift.Hashable, Swift.CaseIterable {

      case objectOne(Skie.KotlinModule.OnlyObjectsSealedClass.ObjectOne.__Kotlin)
      case objectThree(Skie.KotlinModule.OnlyObjectsSealedClass.ObjectThree.__Kotlin)
      case objectTwo(Skie.KotlinModule.OnlyObjectsSealedClass.ObjectTwo.__Kotlin)

    }

  }

Note that Swift.CaseIterable is there because I'm using my forked version of SKIE and it still gives the error
type 'Skie.KotlinModule.OnlyObjectsSealedClass.__Sealed' does not conform to protocol 'CaseIterable'

Edit: I see what you meant in your comment above. Yes, even though the generated enum for only objects still generates cases that expect a parameter, you only have one possible instance to pass there and not multiple.

[FilipDolnik]

Yeah, it's unfortunately not as easy as adding the CaseIterable conformance. We would still need to generate the custom implementation for the allCases because all these sealed enums have associated values. That's possible, but I think only for the case with objects (for the above reason). However, I think that even this can be a valuable feature.

Thanks for the idea; we will put it in the "possible new features" pile.

[corrado4eyes]

Okay @FilipDolnik thanks for the quick responses! I will rename the issue to reflect what we discussed.