▲ up

Migrating your project from Java to Kotlin

I have been part of a few migrations of projects from Java to Kotlin in particular as I have convinced more teams of its value over Java. However I did not want to write another comparison article but rather assuming that you are undertaking this migration how you can tackle it. One of the great benefits here is that Kotlin is entirely interoperable with Java allowing you to not make the changes in a single grand rewrite effort. Instead you can address the changes piecemeal.

This post has been split into two sections of traps to avoid when migrating and tricks to try. Not all sections may be relevant to your particular use case as some sections are focused on the tooling/frameworks that has been used, so please find the ones most applicable and hopefully it assists.

Traps

First we shall cover the areas of migrations that teams have tripped up on unexpectedly and how they were overcome.

Compilation order

When working with a polyglot project it becomes important to understand the compilation order as this defines the allowed dependencies. When using tools such as Gradle I believe it is configurable using the dependsOn argument of the task. In my experience the standardised tooling has chosen the order to be Kotlin => Java and thus I shall cover the elements that this direction of dependency introduces.

The output class files should be co-located and hence full interaction should be capable however there are many fancy tools in the JVM world that generate code or hook into the compilation cycle to alter the byte code (such as Lombok in the next section). It is important to know how these tools works in principle but also as it can lead to unexpected issues. Therefore comprehensively understand the tooling of the project prior to the migration to determine any of these such tools used in the Java toolchain as these will not be able to be used by the Kotlin classes as they are compiled prior to the tool being invoked.

Lombok

Lombok is such a tool that hooks into the compilation cycle to generate boilerplate code within Java. Commonly used to generate getters, setters, and constructors. These files are generated as part of the Java compilation step, therefore all the Lombok generated code becomes inaccessible by Kotlin code in a polyglot project. This has led to several compilation errors of the nature of "symbol not found". Now that you are aware of this trap in the migration these cryptic messages should ideally have a clearer cause.

The best way to avoid this issue is to "delombok" your code as you migrate. It is not necessary to "delombok" the entire code base or even the entire file. The issue can simply be rectified by writing the required function explicitly as Lombok skips over the generation of functions that already exist.

The following shows an object of performing just that, assume that our Kotlin code only needs access to the getName() function.

@Value
public class Person {
  private final int id;
  private final String name;
  private final int age;
}

Will be changed to the following to allow Kotlin access:

@Value
public class Person {
  private final int id;
  private final String name;
  private final int age;

  public String getName() {
    return name;
  }
}

Builder pattern

Another commonly generated pattern via Lombok is the builder pattern. This is very useful in Java in particular to aide in readability when there are several parameters to construct an object. In Kotlin such a pattern is not required as the language itself supports this with named arguments. What this does mean however is when migrating such a class from Java to Kotlin you must implement the builder pattern within Kotlin as the Java classes cannot make use of the named arguments and further it also reduces the amount that is changed in each migration step.

The following is an example of how to implement the builder pattern for a basic data class.

data class Name(val first: String, val last: String) {

  companion object {
    fun builder() = Builder()

      data class Builder(
        var first: String? = null,
        var second: String? = null
        ) {

        fun first(first: String) = apply { this.first = first }
        fun second(second: String) = apply { this.second = second }
        // You can do better handling of null than this but I have found this
        // sufficient until the I get to migrate the calling class to Kotlin
        fun build() = Name(first!!, second!!)
      }
    }
}

Tricks

Next up we will cover some tricks discovered to make the migration easier, in particular focusing on tooling that Kotlin already provides to make this simpler.

Order to migrate

Migrating a project can be a monumental task depending on the size of the code base, therefore it is important just as any other large changes that it be done in small, manageable chunks. The following is some general advice on how to tackle this migration.

Convert the value/data objects first. The principle here is that they have the fewest dependencies (ideally none), therefore moving these across should be simple. Also, typically to migrate a class it is greatly beneficial that the dependent (or imported) classes are already in Kotlin so none of the above traps occur.

Work your way outwards. Once all value/data classes have been converted choose classes that are closely related to it, ideally having only a few dependencies, moving gradually out to the entry point of the program.

Either convert or add functionality, never both. This is strongly aligned with the recommendation when refactoring (which converting a class is), that adding new functionality whilst also performing a refactoring creates two possible sources of error. In regards to migrating to Kotlin in particular this leads to a bloat in code reviews making it very difficult to see the functionality changes. Thus always perform the migration, then add the new functionality.

Kotlin compiler plugins

Kotlin chose a few differing paradigms from Java which results in certain tooling being unable to work due to their leveraging of particular Java paradigms. The most obvious example is mocking frameworks in Java make use of the fact that they can create a subclass of the class to mock This cannot be done in Kotlin as easily as for Java as Kotlin has closed classes by default. To assist in this the Kotlin team have provided compiler plugins such as all-open and no-arg to address these limitations when migrating. It also allows you to continue to make use of the more mature tooling in Java if you wish to.

In saying this however I do recommend exploring Kotlin native libraries/frameworks where possible as these do provide a much better experience.

Kotlin JVM annotations

Another major difference in the languages is that Kotlin does not have a notion of static, its companion object is similar but not the same. Therefore when interoperating with Java one may still wish to use static functions or static constants. Typically these will be placed in the companion object and need to be accessed through MyClass.Companion.MY_CONSTANT. There are annotations provided with the Kotlin standard library to make this look more like a Java class such as MyClass.MY_CONSTANT which then means that you do not need to change all call sites of constants and functions.

For constants/static fields the @JvmField annotation can be used and for static methods the @JvmStatic annotation can be used.

These annotations are not only there to provide aesthetics but also can improve your code. For example using @JvmOverloads one can get all variations of a function generated for a Java class to consume. This is incredibly powerful for creating test data for example.