My current data model uses nullables and a state property to handle the objects state ```data class...

# announcements

[Kurt Renzo Acosta]

My current data model uses nullables and a state property to handle the objects state

data class Model(
    val a: String,
    val b: String,
    val nullableOnlyForStateA: String?,
    val state: State,
    val propForStateA: String?,
    val propForStateB: String?
)

I’m trying to use sealed classes to make it more idiomatic but it seems too verbose:

sealed class Model {
    abstract val a: String
    abstract val b: String
    abstract val nullableOnlyForStateA: String?

    data class A(
        override val a: String,
        override val b: String,
        override val nullableOnlyForStateA: String?,
        val propForStateA: String
    ) : Model()

    data class B(
        override val a: String,
        override val b: String,
        override val nullableOnlyForStateA: String,
        val propForStateB: String
    ) : Model()
}

This seems okay but my original class has much more properties and more states making a 16-line data class turn into a more-than-100 line sealed class. Any ideas on how to make this more ergonomic?

[Michael de Kaste]

not really, it's the cost of making sure you will never encounter a problem with it once it's made vs. encountering potential checks everytime once you use it.

[Kurt Renzo Acosta]

Fair enough. There’s just a lot of repeated code. I wish I can just say that every subclass has these properties without having to put all those override statements

[Nir]

IMHO this is a major shortcoming in Kotlin, plain and simple

[Nir]

No way to avoid repeating fields between data classes

[Nir]

The ideal approach here is to have a dataclass that contains all the common fields

[Nir]

And then you have a member of the sealed class type that contains only the things that differ

[Nir]

But that obviously changes your schema

[Michael de Kaste]

Well you don't HAVE to if you don't need the constructor, but then who would be responsible for setting such data? At least all your sealedsubclasses can't be data classes than anymore.

You could then technically use a builder and all, but at what cost?

[Kurt Renzo Acosta]

The model is what’s in the database but I wanted to map it to a sealed class as the domain model.

When I receive it, I map it out via:

fun DataModel.toDomainModel() = when(state) {
    Case1 -> Case1DomainModel(...)
    Case2 -> Case2DomainModel(...)
    ...
}

[Kurt Renzo Acosta]

Currently, I just have a copy of it as a domain model and it’s working so I don’t necessarily have to do this. Just trying to find ways to make it more expressive

[Nir]

What's in the ... ?

[Nir]

Are you performing some direct deserializing?

[Kurt Renzo Acosta]

The fields Case1DomainModel(field1 = field1, field2 = field2, … and extra cases omitted for brevity

[Nir]

If you have to forward fields around like this anyway

[Nir]

Then it's not strictly necessary that your class structure match the database exactly, so you can do what I suggested

[Nir]

So you would leave Model as a data class, and leave in all the fields that are common to all states

[Nir]

And then have a sealed class, StateData

[Kurt Renzo Acosta]

Yup. I totally get your idea and I’m leaning towards it

[Nir]

Cool sorry wasn't sure if I was clear :-)

[Kurt Renzo Acosta]

It was and I think it’s clever. I’m glad I asked here.