De-Structuring in Kotlin

Recently I wrote an article about de-structuring in TypeScript, and hinted that Kotlin has a similar feature, but in more of a “strongly typed” language style. Today I want to discuss that feature.

To re-cap from the previous article, de-structuring is, at the core, a syntactic sugar to easily “lift” parameters out of an object or array and declare local all in one shot.

With ECMAScript and TypeScript, this comes in two forms: positional (for arrays and iterables) and by name (for objects). One of the things that enables this ability in ECMAScript is the fact that all objects are, in effect, a map data-structure. Every property is accessible on an object by name; e.g: person['name']. Kotlin, being both a strongly-typed language as well as being built on Java, which has objects with full encapsulation rules that notably do not act like ‘maps’ has no such ability.

Consequently, to support de-structuring, Kotlin has to come up with a different solution: the component convention.

Given an arbitrary object, in Kotlin you can try to do this:

 1class Person {
 2  var name: String = ""
 3  var age: Int = 0
 6fun main(args: Array<String>) {
 7  val people: List<Person> = /* ... */
 8  for((name, age) in people) { // compiler error!
 9    println("Name: $name age: $age")
10  }

In this case, the compiler will fail with an error like: Destructuring declaration initializer of type Person must have a 'component1()' function (and will also repeat for component2).

To make it support de-structuring, we can follow the advice and define the indexed component functions. For example:

1class Person {
2  var name: String = ""
3  var age: Int = 0
5  fun component1() = name
6  fun component2() = age

Once these functions are defined, the code above will work as expected. It’s effectively the same as this code:

1fun main(args: Array<String>) {
2  val people: List<Person> = /* ... */
3  for(person in people) {
4    val name = person.component1()
5    val age = person.component2()
6    println("Name: $name age: $age")
7  }

The name you choose for the de-structuring variables doesn’t matter, it’s strictly positional.

In practice, the primary benefit of de-structuring in Kotlin shows up with data classes, in which it’s a free feature. Data classes provide all sorts of convention benefits including: equality and hashcode definition, toString declaration, copy functions, and de-structuring components. That means we can convert our Person class to a data class and get the benefits of de-structuring for free:

1data class Person(var name: String, var age: Int)
3fun main(args: Array<String>) {
4  val people: List<Person> = // ...
5  for((name, age) in people) {
6    println("Name: $name age: $age")
7  }

Of course, built-in Kotlin types also have de-structuring components pre-declared, including:

  • Pair
  • Triple
  • Map.Entry
  • List

This means that there are a lot of “syntax shortcuts” when working with data-structures. For example, consider lists and maps:

 1// Get the first 3 people out of the list.
 2val peopleList: List<Person> = /*...*/
 3val (person1, person2, person3) = peopleList
 5// Get the Key and Value of a mpa in a loop
 6val map: Map<UUID,Person> = // ..
 7for((id,person) in map) {
 8  // id = entry.getKey()
 9  // person = entry.getValue()

Of this set of examples, List is the one that requires some care in use.

  • Unlike TypeScript (which falls back to undefined), if you exceed the list boundaries, the de-structuring will still trigger an IndexOutOfBoundsException. Therefore, de-structuring must be preceded by bounds checking in this case.
  • List has a limit in the number of components available. As of Kotlin 1.2.x, List only supports component1() to component5(). Meaning if you declared the following destructuring variable, it would fail to compile: val (p1, p2, p3, p4, p5, p6) = peopleList

Another pattern that is common with de-sructuring is the “dual return type” pattern, where a function can return two values. Often times this is used for cases where a function returns either a success result or an error, rather than using exceptions for error flow. In these cases, a Pair type (or similar) is used to hold both possible result types. Anyone that has used Go as a language or asynchronous message passing is familiar with this approach:

1fun doSomething(): Pair<SomeResponse?, Error?> {
2  // ...

The advantage of using a componentized type like Pair is that it can be de-structured at the call-site:

1val (response, error) = doSomething()
2if(error != null)
3  // handle error
5  // do something with the response

De-structuring can also be used in lambda declarations. Consider a lambda that takes a Person object. That lambda can de-structure in the variable declaration:

1val people: List<Person> = /* ... */
2people.foreach { (name, age) -> println("Name: $name age: $age") }

Finally, it’s worth noting that when de-structuring you can ignore fields that you don’t care about using underscores:

1val people: List<Person> = /* ... */
2people.foreach { (_, age) -> println("Age: $age") }

Of course, being attached to a strongly typed language, de-structuring in Kotlin has some limitations compared to the feature in ECMAScript:

  • There is no partitioning or spreading supported (like (first, - this is at least partially a byproduct of the fact that spreading to produce “the rest of” results in the definition of a new “type”. In TypeScript this is no big deal; an object minus a few properties is still an object. But in a JVM based language, what do you call a Person object that doesn’t have a name property anymore? While TypeScript can change { 'name': 'test', 'age': 15 } to {'age':15} when excluding name, Kotlin needs a strict type to use for all return values, and no such type exists.
  • Everything in Kotlin de-structuring is order-based; therefore ordering is part of the API contract! if you change your data class (or whatever class is providing the component functions) in a way that changes the order of properties, it will break de-structuring usages. This may seem obvious (constructor arg reordering is also contract breaking) but it may seem surprising if you always choose to use matching names in your de-structuring sites; remember: the names at the call-site are unrelated to the names of the components!
  • There is no variable defaulting available - in other words you can’t currently say: (name, age = 13) in case age is null.
comments powered by Disqus