Skip to content

Little utilities for more pleasant immutable data in Kotlin

License

Notifications You must be signed in to change notification settings

kopykat-kt/kopykat

Repository files navigation

This project is currently on maintenance mode. We recommend to try Kopy if you need a similar feature for Kotlin 2.x.

One of the great features of Kotlin data classes is their copy method. But using it can become cumbersome very quickly, because you need to repeat the name of the field before and after.

data class Person(val name: String, val age: Int)

val p1 = Person("Alex", 1)
val p2 = p1.copy(age = p1.age + 1)  // too many 'age'!

This plug-in generates a few new methods that make working with immutable (read-only) types, like data classes and value classes, more convenient.

IntelliJ showing the methods

Those methods can be divided in two big groups:

  • Same-type transformations, which take a value of a certain type and produce a copy of the same type by changing a few fields. Those methods are closer to the copy method available in Kotlin.
  • Isomorphic copy constructors, which produce a value of a different type from the given one, provided that they both contain the same fields. Those methods are very useful when interfacing across different layers of the application.

Same-type transformations

KopyKat extends Kotlin's built-in copy with a version based on mutable copies, and a version based on maps (that is, you state the changes to be done to the values based on the old ones instead of the new values themselves.) In addition, the default copy is extended to work on sealed hierarchies and value classes.


Mutable copy

This new version of copy takes a block as a parameter. Within that block, mutability is simulated; the final assignment of each (mutable) variable becomes the value of the new copy. These are generated for both data classes and value classes.

val p1 = Person("Alex", 1)
val p2 = p1.copy { 
  age++
}

You can use old to access the previous (immutable) value, before any changes.

val p3 = p1.copy { 
  age++
  if (notTheirBirthday) {
    age = old.age  // get the previous value
  }
}

Nested mutation

If you have a data class that contains another data class (or value class) as a property, you can also make changes to inner types. Let's say we have these types:

data class Person(val name: String, val job: Job)
data class Job(val title: String, val teams: List<String>)

val p1 = Person(name = "John", job = Job("Developer", listOf("Kotlin", "Training")))

Currently, to do mutate inner types you have to do the following:

val p2 = p1.copy(job = p1.job.copy(title = "Señor Developer"))

With KopyKat you can do this in a more readable way:

val p2 = p1.copy { job.title = "Señor Developer" }

Warning For now, this doesn't work with types that are external to the source code (i.e. dependencies). We are working on supporting this in the future.

Nested collections

The nested mutation also extends to collections, which are turned into their mutable counterparts, if they exist.

val p3 = p1.copy { job.teams.add("Compiler") }

To avoid unnecessary copies, we recommend to mutate the collections in-place as much as possible. This means that forEach functions and mutation should be preferred over map.

val p4 = p1.copy { // needs an additional toMutableList at the end
  job.teams = job.teams.map { it.capitalize() }.toMutableList()
}
val p5 = p1.copy { // mutates the job.teams collection in-place
  job.teams.forEachIndexed { i, team -> job.teams[i] = team.capitalize() }
}

The at.kopyk:mutable-utils library (documentation) contains versions of the main collection functions which reuse the same structure.

val p6 = p1.copy { // mutates the job.teams collection in-place
  job.teams.mutateAll { it.capitalize() }
}

Mapping copyMap

Instead of new values, copyMap takes as arguments the transformations that ought to be applied to each argument. The "old" value of each field is given as argument to each of the functions, so you can refer to it using it or introduce an explicit name.

val p1 = Person("Alex", 1)
val p2 = p1.copyMap(age = { it + 1 })
val p3 = p1.copyMap(name = { nm -> nm.capitalize() })

The whole "old" value (the Person in the example above) is given as receiver to each of the transformations. That means that you can access all the other fields in the body of each of the transformations.

val p4 = p1.copyMap(age = { name.count() })

When using value classes, given that you only have one property, you can skip the name of the property.

@JvmInline value class Age(ageValue: Int)

val a = Age(39)

val b = a.copyMap { it + 1 }

You can use copyMap to simulate copy, by making the transformation return a constant value.

val p5 = p1.copyMap(age = { 10 })

copy for sealed hierarchies

KopyKat also works with sealed hierarchies. These are both sealed classes and sealed interfaces. It generates regular copy, copyMap, and mutable copy for the common properties, which ought to be declared in the parent class.

abstract sealed class User(open val name: String)
data class Person(override val name: String, val age: Int): User(name)
data class Company(override val name: String, val address: String): User(name)

This means that the following code works directly, without requiring an intermediate when.

fun User.takeOver() = this.copy { name = "Me" }

Equally, you can use copyMap in a similar fashion:

fun User.takeOver() = this.copyMap(name = { "Me" })

Or, you can use a more familiar copy function:

fun User.takeOver() = this.copy(name = "Me")

Warning KopyKat only generates these if all the subclasses are data or value classes. We can't mutate object types without breaking the world underneath them. And cause a lot of pain.


copy for type aliases

KopyKat can also generate the different copy methods for a type alias.

@CopyExtensions
typealias Person = Pair<String, Int>

// generates the following methods
fun Person.copyMap(first: (String) -> String, second: (Int) -> Int): Person = TODO()
fun Person.copy(block: `Person$Mutable`.() -> Unit): Person = TODO()

The following must hold for the type alias to be processed:

  • It must be marked with the @CopyExtensions annotation,
  • It must refer to a data or value class, or a type hierarchy of those.

Isomorphic copy constructors

We know, isomorphic seems like a big word. However, it just means that two things are similar in some way. In this case KopyKat can generate copy constructors between two types that have the same properties, with the same name, and the same type.

In Kotlin, a copy constructor is a top level function with the same name as the type (in PascalCase) that returns the given type. This naming pattern is described in the official Kotlin Code Conventions.

To generate these you have to annotate your types with one of the following:

  • @Copy
  • @CopyFrom
  • @CopyTo

All of them take another type to copy from/to, as a parameter. In the case of @Copy, it will generate two functions for both directions. So, if we have code like this:

data class Person(val name: String, val age: Int)

@Copy(Person::class)
data class LocalPerson(val name: String, val age: Int)

The following code is generated:

inline fun Person(from: LocalPerson): Person =
    Person(name = from.name, age = from.age)

inline fun LocalPerson(from: Person): Person =
    LocalPerson(name = from.name, age = from.age)

These allow to convert from one type to the other and vice versa. This is quite a common pattern used to cross boundaries of the different layers of an application. Often, they are called mappers.

If you don't need either of the copy constructors, you can use either @CopyFrom or @CopyTo. @CopyFrom will generate a copy constructor from the provided type to the annotated type (LocalPerson -> Person). On the other hand, if you use @CopyTo will generate the oposite (Person -> LocalPerson).


Nested copy constructors

In some cases you may want to have properties that are different on both types. To support data trees like that, you must make sure that they have copy constructors generated as well. For example:

data class Person(val name: String, val job: Job)
data class Job(val title: String)

@Copy(Person::class) data class LocalPerson(val name: String, val job: LocalJob)
@Copy(Job::class) data class LocalJob(val title: String)

val localPerson = LocalPerson("Alice", LocalJob("Developer"))
val person = Person(localPerson)
check(person.name == "Alice")
check(person.job.title == "Developer")

In this example, if LocalJob is not annotated wih @Copy (or @CopyFrom) the compiler will complain about it.


Multiple copy constructors

Annotations for copy constructors (@Copy[From|To]) can be applied more than once to the same type. This means that you can define mapping across multiple isomorphic types:

@Copy(LocalPerson::class)
@Copy(RemotePerson::class)
data class Person(val name: String, val age: Int)

data class LocalPerson(val name: String, val age: Int)

data class RemotePerson(val name: String, val age: Int)

This configuration will generate 4 different copy constructors.


Using KopyKat in your project

This demo project showcases the use of KopyKat alongside version catalogs.

KopyKat builds upon KSP, from which it inherits easy integration with Gradle. To use this plug-in, add the following in your build.gradle.kts:

  1. Add Maven Central to the list of repositories.

    repositories {
      mavenCentral()
    }
  2. Add KSP to the list of plug-ins. You can check the latest version in their releases.

    plugins {
      id("com.google.devtools.ksp") version "1.7.10-1.0.6"
    }
  3. Add a KSP dependency on KopyKat.

    dependencies {
      // other dependencies
      ksp("at.kopyk:kopykat-ksp:$kopyKatVersion")
    }

    If you are using Kotlin Multiplatform you need to choose explicitly the compilation targets which need processing. For example, this is how you instruct KSP to run on the files in commonMain.

    dependencies {
      // other dependencies
      add("kspCommonMainMetadata", "at.kopyk:kopykat-ksp:$kopyKatVersion")
    }
  4. (Optional) If you are using IntelliJ as your IDE, we recommend you to follow these steps to make it aware of the new code.


Only for selected types

By default, KopyKat generates methods for every data and value class, and sealed hierarchies of those. If you prefer to enable generation for only some classes, this is of course possible. Note that you always require a @CopyExtensions annotation to process a type alias.

All classes in given packages

Change the generate option for the plug-in, by passing options to KSP. The packages should be separated by :, and you can use wildcards, as supported by wildcardMatch.

ksp {
  arg("generate", "packages:my.example.*")
}

Using annotations

  1. Add a dependency to KopyKat's annotation package. Note that we declare it as compileOnly, which means there's no trace of it in the compiled artifact.

    dependencies {
      // other dependencies
      compileOnly("at.kopyk:kopykat-annotations:$kopyKatVersion")
    }
  2. Change the generate option for the plug-in, by passing options to KSP.

    ksp {
      arg("generate", "annotated")
    }
  3. Mark those classes you want KopyKat to process with the @CopyExtensions annotation.

    import at.kopyk.CopyExtensions
    
    @CopyExtensions data class Person(val name: String, val age: Int)

Customizing the generation

You can disable the generation of some of these methods by passing options to KSP in your Gradle file. For example, the following block disables the generation of copyMap.

ksp {
  arg("mutableCopy", "true")
  arg("copyMap", "false")
  arg("hierarchyCopy", "true")
}

By default, the three kinds of methods are generated.


What about optics?

Optics, like the ones provided by Arrow, are a much more powerful abstraction. Apart from changing fields, optics allow uniform access to collections, possibly-null values, and hierarchies of data classes. You can even define a single copy function which works for every type, instead of relying on generating an implementation for each data type.

KopyKat, on the other hand, aims to be just a tiny step further from Kotlin's built-in copy. By re-using well-known idioms, the barrier to introducing this plug-in becomes much lower. Our goal is to make it easier to work with immutable data classes.