Kotlin DSL for Test Data

Kotlin DSLs can be used to replace a boring test data strings with correct and an easy to read a code.

That time I was working on java9c plugin for Gradle, I created integration tests. In my case all those tests were of the following pattern: create sample Gradle project, execute it, check results. I decided to run a fun experiment and replace boring string constants with a Gradle-looking DSL.

Namely, instead of (and many Gradle plugin integration tests have similar)

        "apply plugin: 'java-library'\n" +
        "repositories {\n" +
        "  mavenCentral()\n" +
        "}\n" +
        "\n" +
        "dependencies {\n" +
        "  implementation 'junit:junit:4.12'\n" +

I created a tiny DSL that looks (and parses) like a Gradle script. The DSL generates test-data files for me. Kotlin compiler and IDE helps to prevent errors before a test is executed. Code completion makes a new test authoring easier. Here is the DSL usage example:

generateDSL {
  `apply plugin`("java-library")

  repositories {

  dependencies {

More examples are on GitHub in java9c plugin test sources.

The example looks like a Gradle script. There was no goal to make it 100% same looking. Also, there is an amazing project run by Gradle to support Kotlin DSLs in Gradle, natively. That DSL is not 100% same looking to a Gradle-Groovy scripts too.

The implementation of that my magic DSL fits in one file, and it’s about 150 lines.

Next, I’ll explain how one can create similar DSLs for their own needs. With Kotlin you may target JVM, Android, JS, and Native, reusing same pure-Kotlin code.

DSL Basics

We need to create a text generator DSL. The primary decision if either to generate a bare text or to use an API of a library. You may consider XML DOM or Jackson. There are kotlinpoet or javapoet to generate Kotlin or Java code via an API. I bet there are many other libraries. Nebula Gradle Lint Plugin can read/write Gradle scripts too.

There is a trade-off. Dealing with a library may be complicated and time-consuming, but way more stable.

For the sake of java9c tests, I decided to implement the generator based on bare text output. And it’d be me who covers all risks and bugs from the implementation. It is only about 150 lines (now) of code.

Writer Interface

I started with a line writer interface:

interface LineWriter {
  operator fun String.unaryMinus()

Inside the interface, I use unary minus, e.g., -"foo" as the function to write a line. It reads better in DSLs, e.g.

lineWriter.apply {
  -"line of text"

Well, you may decide to have a fun line(text: String) instead. That does not change the rest, so, please feel free to use a function instead of an operator. Alternatively, you may use String.unaryPlus, so that you’d have +"foo".

See Operator Overloading documentation or ask me, if you need to clarify the trick.

A Trivial Writer Implementation

The implementation of the interface could be something trivial, e.g.

fun generateDSL(builder: LineWriter.() -> Unit) : String {
  val result = StringBuilder()
  object : LineWriter {
    override fun String.unaryMinis() {
  return result.toString() 

The usage could be:

val text : String = generateDSL {
  -"line of text"

The generateDSL receives a lambda with receiver and returns resulting string. It is the implementation detail to pass the instance of LineWriter to the lambda. Inside the lambda, the receiver is LineWriter, it means, that this keyword resolves to an instance of LineWriter. Of course, this. can be omitted and all methods are resolved against LineWriter instance. It follows that - "foo" calls resolves to String.unaryMinus() function of LineWriter inside the lambda scope.

For short, we may compact the generateDSL function to the following:

fun generateDSL(builder: LineWriter.() -> Unit) = buildString {
  object : LineWriter {
    override fun String.unaryMinis() {

Here I use a fun buildString(builderAction: StringBuilder.() -> Unit): String function from the Kotlin standard library. It receives yet another lambda with receiver on StringBuilder type so that appendln is a function from it.

Theoretically, you may have several different entry point functions (e.g., generateDSL) to, say, generate a string, a file or something else. The rest does not depend on a particular generate* function.

Indenting and Blocks

Text generation for languages like Gradle requires indenting. We have blocks, and it’s nice to simplify blocks generation. At first, I created an offset function for it:

fun LineWriter.offset(): LineWriter {
  val outer = this
  return object : LineWriter {
    override fun String.unaryMinus() {
      val s = this
      outer.apply { 
        -("  " + s) 

The function is an extension function. It makes no need to change the original LineWriter interface, but it still reads as a method call.

The block function is as follows:

fun LineWriter.block(name : String, builder: LineWriter.() -> Unit) {
  -"$name {"

Here I use string interpolation to simplify code of the first line.

At that point I can write the DSL snippets like that:

generateDSL {
  block("jonnyzzz") {

And it yields

jonnyzzz {

Nice, isn’t it?

Gradle Specific Constructs

With block function one can create all necessary functions to generate blocks like repositories, dependencies and so on. Now it is time to implement specific parts of the DSL and allow some constructs only inside other constructs.


Let’s consider repositories block. Inside we have pre-defined functions for mavenCentral() and jcenter().

First, we need a builder interface and implementation. It can be done as follows:

interface RepositoriesWriter : LineWriter {
  fun mavenCentral() { -"mavenCentral()" }
  fun mavenLocal() { -"mavenLocal()" }
  fun jcenter() { -"jcenter()" }

fun RepositoriesHolder.repositories(builder: RepositoriesWriter.() -> Unit) =
        block("repositories") {
          object : RepositoriesWriter, LineWriter by it {

We define an interface RepositoriesWriter to play as the scope of the generation. In the interface, we have mavenLocal and other functions with trivial implementations. Those functions can be alternatively implemented as extension functions of inside the repositories function. That is up to the author.

In the repositories function, I use class delegation aka by keyword to implement LineWriter from RepositoriesWriter to delegate to another instance of LineWriter. So short to write and powerful!

As the result, we can have

generateDSL {
  repositories {
     -"// another line"

The same way I created the whole bunch of functions to support the subset of Gradle scripts I was using in my tests. You may take a look here for more details.

Specific DSL Alternatives

It was another design decision to allow LineWriter functions and extension functions (e.g., block) of the scope of RepositoriesWriter lambda. We might have decided opposite. In the case, we would need DslMarker annotation to make sure LineWriter functions and extension functions are not resolved to the outer scope. We probably have a generateDSL function call on the top.

Wrapping Up

DSLs are nice. In the post, I presented the DSL building pattern. Use it to create your DSLs. Ask me if you have questions. You may also check this article from Kotlin documentation or a video of a talk by Hadi.

comments powered by Disqus