A couple of years ago, I developed an application in Kotlin based on Camunda BPMN to help me manage my conference submission workflow. It tracks my submissions in Trello and synchronizes them on Google Calendar and in a Google Sheet. Google Calendar offers a REST API. As REST APIs go, it's cluttered with String
everywhere. Here's an excerpt of the code:
fun execute(color: String, availability: String) { findCalendarEntry(client, google, execution.conference)?.let { it.colorId = color // 1 it.transparency = availability // 2 client.events() .update(google.calendarId, it.id, it).execute() } }
- Set the event's color. Valid values are "0", "1", ... to "11"
- Set the event's availability. Valid values are
"transparent"
and"opaque"
However, my experience has taught me to favor strong typing. I also want to avoid typos. I want to list some alternatives to using String
in this post.
Constants
The oldest trick in the book, available in most languages, is to define constants. Before Java 5, developers used this alternative a lot as it was the only one available. It would look like this:
const val Default = "0" const val Blue = "1" const val Green = "2" const val Free = "transparent" const val Busy = "opaque"
We can now call the function accordingly:
execute(Blue, Busy)
Constants help with typos. The flip side is that they cannot enforce strong typing:
execute(Blue, Red) // 1 execute(Free, Red) // 2
- Pass two colors, but the compiler is fine
- Invert the arguments; the compiler is still fine
Type aliases
The idea behind type aliases is to alias the name of an existing type to something more meaningful.
typealias Color = String typealias Availability = String
With this, we can change the signature of the function:
fun execute(color: Color, availability: Availability) { // ... }
Unfortunately, type aliases are just cosmetic. Whatever the alias, a String
stays a String
. We can still write incorrect code:
execute(Blue, Red) // 1 execute(Free, Red) // 1
- Nothing has improved
Enumerations
Whether in Java or Kotlin, enumerations are the first step toward strong typing. I believe most developers know about them. Let's change our code to use enums:
enum class Color(val id: String) { Default("0"), Blue("1"), Green("2"), } enum class Availability(val value: String) { Free("transparent"), Busy("opaque"), }
We need to change the function accordingly, both the signature and the implementation:
fun execute(color: Color, availability: Availability) { findCalendarEntry(client, google, execution.conference)?.let { it.colorId = color.id // 1 it.transparency = availability.value // 1 client.events() .update(google.calendarId, it.id, it).execute() } }
- Extract the value wrapped by the
enum
The usage of enumerations enforces strong-typing:
execute(Color.Blue, Availability.Busy) // 1 execute(Color.Blue, Color.Red) // 2 execute(Availability.Free, Color.Blue) // 2
- Compile
- Doesn't compile!
Inline classes
A recent Kotlin feature is fully dedicated to strong typing: inline classes. An inline class wraps a single "primitive" value, such as Int
or String
. Picture the following class:
data class Person(givenName: String, familyName: String)
Callers of this class would have to remember whether the first parameter is the given name or the family name. Kotlin already helps by allowing named parameters:
val p = Person(givenName = "John", familyName = "Doe")
However, we can improve the snippet above by wrapping the String
in two different value types, one for each role.
@JvmInline value class GivenName(value: String) @JvmInline value class FamilyName(value: String) val p = Person(GivenName("John"), FamilyName("Doe"))
At this point, one cannot swap a given name for a family name, or vice versa. Likewise, we can use value classes in our example and define possible values in a companion object.
@JvmInline value class Color(val id: String) { companion object { val Default = Color("0") val Blue = Color("1") val Green = Color("2") } } @JvmInline value class Availability(val value: String) { companion object { val Free = Availability("transparent") val Busy = Availability("opaque") } } execute(Color.Blue, Availability.Busy) // 1 execute(Color.Blue, Color.Red) // 2 execute(Availability.Free, Color.Blue) // 2
- Compile
- Doesn't compile!
Sealed classes
Sealed classes are another possible way to enforce strong typing. The limitation is we need to define all subclasses of a sealed class in the same file. There can't be any inheritance by third parties. In effect, it makes the class open
for your code and final
for client code.
Instead of defining a type and several instances of it as in value classes, we define the different types directly.
sealed class Color(val id: String) { object Default: Color("0") object Blue: Color("1") object Green: Color("2") } sealed class Availability(val value: String) { object Free : Availability("transparent") object Busy : Availability("opaque") } execute(Color.Blue, Availability.Busy) // 1 execute(Color.Blue, Color.Red) // 2 execute(Availability.Free, Color.Blue) // 2
- Compile
- Doesn't compile!
Note that I defined the objects in their respective parent classes. Depending on your context, you may want to make them top-level instead.
sealed class Color(val id: String) object Default: Color("0") object Blue: Color("1") object Green: Color("2") sealed class Availability(val value: String) object Free : Availability("transparent") object Busy : Availability("opaque") execute(Blue, Busy)
Conclusion
Kotlin offers several options to enforce strong typing on one's APIs: enumerations, value classes, and sealed classes.
While most developers are pretty comfortable with enumerations, I'd advise considering value and sealed classes as they bring additional benefits to the table.
To go further:
Originally published at A Java Geek on February 20nd, 2022