Riccardo Cardin
18 min read •
Elevate Your Tests: Testing Functional Kotlin with Arrow and Raise
Share on:
Introduction
At Rock the JVM, we deeply understand the power of the Kotlin Arrow library and the Raise DSL, and we’ve previously shared our insights in our article on Functional Error Handling in Kotlin, Part 3: The Raise DSL. Now, we’re ready to introduce the crucial topic of testing applications that use the Raise DSL. To fully grasp the concepts we’ll be discussing, it’s essential to have a solid understanding of Kotlin, Arrow library, and the Raise DSL. We recommend revisiting our previous posts or resources if you need to refresh your knowledge. Get ready to explore the testing world of Arrow Raise with confidence and understanding.
Setting Up the project
First things first, let’s set up a Kotlin project. We can use Gradle or Maven, but for this article, we’ll focus on Gradle. As always, use the gradle init command from a shell to create a Gradle application from scratch. We’ll then add the following dependencies to the build.gradle.kts file, providing you with practical examples and clear explanations along the way:
dependencies {
implementation("io.arrow-kt:arrow-core:1.2.4")
testImplementation("org.jetbrains.kotlin:kotlin-test-junit5")
testImplementation("org.junit.jupiter:junit-jupiter-engine:5.10.0")
testImplementation("org.assertj:assertj-core:3.26.3")
testImplementation("io.kotest:kotest-runner-junit5:5.9.0")
testImplementation("io.kotest.extensions:kotest-assertions-arrow:1.4.0")
testImplementation("org.mockito.kotlin:mockito-kotlin:5.4.0")
}
Once we set up the project, we need some helpful use cases to test since this article will focus on testing the Raise DSL. We’ll take a typical example. We want to create some business logic to create a stock portfolio for a user. Initially, we want to make an empty portfolio for the user, adding some initial amount of money. We can model the use case as follows:
interface CreatePortfolioUseCase {
fun Raise<DomainError>.createPortfolio(model: CreatePortfolio): PortfolioId
}
data class CreatePortfolio(
val userId: UserId,
val amount: Money,
)
@JvmInline
value class UserId(val id: String)
@JvmInline
value class PortfolioId(val id: String)
sealed interface DomainError
Let’s analyze the above code briefly since the function’s signature tells us a lot of helpful information. We can start from the name. The function creates a portfolio. It would persist in some databases. If so, we should have used the suspend modifier for the function. However, having a suspended function would have added much complexity unrelated to Arrow to our code. So, we will not use a suspended function. Once the new portfolio is created, the function returns its identifier, representing the happy path. Finally, the function can raise a DomainError in case of failure since it’s declared using the type Raise<DomainError> as a receiver.
In this article, we’ll take a different approach than the one in the article Functional Error Handling in Kotlin, Part 3: The Raise DSL. In fact, context receivers have been deprecated in the meantime, and they are eligible to be deleted in future versions in favor of context parameters. So, it’s better to use simple receivers and wait for context parameters to be released soon.
We can start implementing the use case now that we have set up. Since we are diligent and well-behaved developers, we want to write tests before implementing the use case, following the Test-Driven Development (TDD) approach.
Testing the Use Case
First, we want to test the happy path, meaning the use case creates a new portfolio for the user. We need to implement our use case interface with a concrete class, which we usually call a service:
fun createPortfolioUseCase(): CreatePortfolioUseCase =
object : CreatePortfolioUseCase {
override fun Raise<DomainError>.createPortfolio(model: CreatePortfolio): PortfolioId = TODO()
}
As you might have noticed, the createPortfolioUseCase method is nothing more than a factory method.
We’ll use different testing frameworks. Let’s begin with a setup that should be familiar to developers addicted to Kotlin and Spring: JUnit 5 as the test runtime and AssertJ for assertions.
internal class CreatePortfolioUseCaseJUnit5Test {
private val underTest = createPortfolioUseCase()
@Test
internal fun `given a userId and an initial amount, when executed, then it create the portfolio`() =
TODO("Not yet implemented")
}
For convention, we call the unit we want to test underTest. Moreover, we used a typical given-when-then structure for the test’s name.
Now, we have to test that, given some inputs, the function will return the expected output. However, here we have a complication. The function is declared in the Raise<DomainError> context, which means that the code that calls it should be able to handle a possible error of type DomainError. Moreover, we need to use some scope function to create a scope with an instance of the CreatePortfolioUseCase type. Usually, we use the with scope function for this purpose.
However, the below implementation will not even compile:
@Test
internal fun `given a userId and an initial amount, when executed, then it create the portfolio`() {
val actualResult: PortfolioId =
with(underTest) {
createPortfolio(CreatePortfolio(UserId("bob"), Money(1000.0)))
}
}
The compiler gives us the following error:
No context receiver for 'arrow.core.raise.Raise<in.rcard.arrow.raise.testing.DomainError>' found.
Nothing more true. Unfortunately, we can’t build our instance of the Raise<E> type. So, we need to use what the Arrow library gives us. One handful approach is to transform the Raise<E>.() -> A function in a more manageable () -> Either<E, A> function. Arrow gives us all the builders to create wrapped types from a function with a Raise<E> context, so let’s use them:
val actualResult: Either<DomainError, PortfolioId> =
either {
with(underTest) {
createPortfolio(CreatePortfolio(UserId("bob"), Money(1000.0)))
}
}
Now that we have a manageable function, we can add the assertion to our test, which becomes the following:
@Test
internal fun `given a userId and an initial amount, when executed, then it create the portfolio`() {
val actualResult: Either<DomainError, PortfolioId> =
either {
with(underTest) {
createPortfolio(CreatePortfolio(UserId("bob"), Money(1000.0)))
}
}
Assertions.assertThat(actualResult.getOrNull()).isEqualTo(PortfolioId("1"))
}
The assertThat function is a static method of the org.assertj.core.api.Assertions, and we take advantage of the getOrNull method of the Either type to get the value of the Right side of the Either type. The AssertJ library has no built-in support for Arrow types. However, we can use the assertions imported by the assertj-arrow-core library (if you’re asking, the library’s author is me). The library has a lot of fluent assertions tailored to the Arrow library. For example, we can use the assertions on the Either<E, A> type directly, changing the above test as follows:
@Test
internal fun `given a userId and an initial amount, when executed, then it create the portfolio (using AssertJ-Arrow-Core)`() {
val actualResult: Either<DomainError, PortfolioId> =
either {
with(underTest) {
createPortfolio(CreatePortfolio(UserId("bob"), Money(1000.0)))
}
}
EitherAssert.assertThat(actualResult).containsOnRight(PortfolioId("1"))
}
As we can see, we don’t need to extract the value from the Either<E, A> anymore.
Now, we can implement the function createPortfolio to make the test pass. Let’s do it in a dumb way, returning a fixed value:
fun createPortfolioUseCase(): CreatePortfolioUseCase =
object : CreatePortfolioUseCase {
override fun Raise<DomainError>.createPortfolio(model: CreatePortfolio): PortfolioId = PortfolioId("1")
}
If we run our test, it should be green.
Instead of transforming the Raise<E>.() -> A function in a () -> Either<E, A> function, we can use the fold function provided by the Arrow library:
@Test
internal fun `given a userId and an initial amount, when executed, then it create the portfolio (using fold)`() {
fold(
block = {
with(underTest) {
createPortfolio(CreatePortfolio(UserId("bob"), Money(1000.0)))
},
},
recover = { Assertions.fail("The use case should not fail") },
transform = { Assertions.assertThat(it).isEqualTo(PortfolioId("1")) },
)
}
However, the above code is cumbersome and less readable than the previous one. Moreover, we must apply a fold function whenever we want to test a function declared in a Raise<E> context. Fortunately, the assertj-arrow-core does it for us, defining some handful of assertions that use the fold function under the hood:
@Test
internal fun `given a userId and an initial amount, when executed, then it create the portfolio (using RaiseAssert)`() {
RaiseAssert
.assertThat {
with(underTest) {
createPortfolio(
CreatePortfolio(
UserId("bob"),
Money(1000.0),
),
)
}
}.succeedsWith(PortfolioId("1"))
}
The test is less readable than the one with the either builder because the library doesn’t support suspending functions (at least for now). However, the RaiseAssert class is a powerful tool to test functions declared in a Raise<E> context. As we said, the succeedWith uses the fold function under the hood, so we don’t need to worry about it.
We have used JUnit 5 until now. However, we can switch to Kotest. Kotest is a robust testing framework for Kotlin, which is very close to ScalaTest for the Scala language. Kotest also has a set of tailored assertions for some of the available types in the Arrow library (see the documentation for further details).
So, let’s translate the above tests in Kotest notation.
internal class CreatePortfolioUseCaseKotestTest : ShouldSpec({
val underTest = createPortfolioUseCase()
context("The create portfolio use case") {
should("create a portfolio for a user") {
val actualResult: Either<DomainError, PortfolioId> =
either {
with(underTest) {
createPortfolio(CreatePortfolio(UserId("bob"), Money(1000.0)))
}
}
actualResult.shouldBeRight(PortfolioId("1"))
}
}
})
This article is not a tutorial on how to use Kotest. However, we can notice that tests are not methods of the class, as in JUnit 5. Tests are lambdas with receiver given to the ShouldSpec class, one of the available contexts. If you extend the ShouldSpec class, you can access the context and should methods, declared functions in the ShouldSpecRootScope. The lambda you give to the ShouldSpec constructor is defined with the ShouldSpecRootScope context. As you might notice, we didn’t use the runTest or runBlocking functions since the should method accepts a suspending function itself.
In this test, we used the approach of converting the Raise<E>.() -> A context into an () -> Either<E, A>, and then we took advantage of the shouldBeRight assertions given by the kotest-assertions-arrow library.
Unfortunately, Kotest has no extension to test a Raise<E>.() -> A function directly. So, if we don’t want to convert it into an Either<E, A>, we need to use the approach with the fold method:
should("create a portfolio for a user (using fold)") {
fold(
block = {
with(underTest) {
createPortfolio(CreatePortfolio(UserId("bob"), Money(1000.0)))
},
},
recover = { fail("The use case should not fail") },
transform = { it.shouldBe(PortfolioId("1")) },
)
}
So, we briefly introduce how to test a function that can raise an error of type E. What if the function calls another function that can raise an error, too? We should not depend on code outside the unit we’re testing. At the end of the day, we focus on unit testing. Let’s see what the available options are.
Mocking the Raise DSL
When dealing with unit tests and depending on a component outside our unit under test, we have at least two choices: fake objects and mocks. But before going deep into the topic, let’s prepare the playground.
Let’s say we only have one portfolio per user. So, we need to check if the user already has a portfolio before creating a new one. If we want to use a hexagonal architecture pattern, the use case should retrieve the information through an interface we call a port. A port is a way for our business logic (or application) to be driven by or to drive external systems without depending directly on them. Okay, too much theory for today. Let’s jump into the code.
First, we now have a way for our use case to fail: A portfolio for a user may already exist. So, we need to add a new error:
sealed interface DomainError {
data class PortfolioAlreadyExists(val userId: UserId) : DomainError
}
Then, we can define the new port interface. Given a userId, we can count the user’s portfolios.
interface CountUserPortfoliosPort {
fun Raise<DomainError>.countByUserId(userId: UserId): Int
}
Finally, let’s wire all the things together, starting using our port into the use case:
fun createPortfolioUseCase(countUserPortfolios: CountUserPortfoliosPort): CreatePortfolioUseCase =
object : CreatePortfolioUseCase {
override fun Raise<DomainError>.createPortfolio(model: CreatePortfolio): PortfolioId {
ensure(countUserPortfolios.countByUserId(model.userId) == 0) {
raise(DomainError.PortfolioAlreadyExists(model.userId))
}
return PortfolioId("1")
}
}
We’re passing the port directly as an input parameter of the factory function createPortfolioUseCase.
As you might expect, we need to change our tests too. When creating the use case instance under test, we must understand what to pass to the createPortfolioUseCase function. As we said, we have at least two choices. The first one is called fake objects. Martin Fowler defines a fake object as follows (see the articles Test Double for further details):
Fake objects actually have working implementations, but usually take some shortcut which makes them not suitable for production.
In our case, we need to implement the port for our test. For example, we want to implement a test verifying the use case when the user already has a portfolio. We can implement the port as follows:
private val fakeCountUserPortfolios: CountUserPortfoliosPort =
object : CountUserPortfoliosPort {
override fun Raise<DomainError>.countByUserId(userId: UserId): Int =
if (userId == UserId("bob")) 0 else 1
}
So, if the given userId is not bob, the fake implementation will show the user has a portfolio. Otherwise, it returns that the user has no portfolio. We can use the fakeCountUserPortfolios object to create the use case under test and implement the test we need using Kotest:
should("raise a PortfolioAlreadyExists") {
val alice = UserId("alice")
val actualResult: Either<DomainError, PortfolioId> =
either {
with(underTest) {
createPortfolio(CreatePortfolio(alice, Money(1000.0)))
}
}
actualResult.shouldBeLeft(PortfolioAlreadyExists(alice))
}
Implementing the test in JUnit 5 is the same, so we’ll omit it.
So, we didn’t find many concerns about using fake objects. The only annoying thing is that we must implement our port twice: one for production execution and one for tests. Many developers prefer something other than this trade-off and use stubs and mocks.
We start again from the definition of mocks given by Martin Fowler to introduce stubs and mocks:
Stubs provide canned answers to calls made during the test, usually not responding at all to anything outside what’s programmed in for the test.
Mocks are pre-programmed with expectations which form a specification of the calls they are expected to receive. They can throw an exception if they receive a call they don’t expect and are checked during verification to ensure they got all the calls they were expecting.
For the sake of simplicity, we’ll call both test doubles as mocks. So, mocks are partial implementations of an interface that we can program to respond to a given set of inputs (signals), and we can verify that they received the expected signals.
There are a lot of libraries that can help us to create mocks. The most famous in Java is Mockito. In the Kotlin ecosystem, its counterpart is called MockK instead. MockK is a mocking library that is idiomatic to Kotlin. It’s mighty and easy to use. Again, this article is intended to be something other than a tutorial on how to use MockK. So, we’ll focus solely on how to use it to mock a function defined with a Raise<E> context.
Mocking a dependency is a three-step process. First, you need to retrieve from the library an empty mock of the dependency:
val countUserPortfoliosMock: CountUserPortfoliosPort = mockk()
The mockk() factory function provides a proxy to the port we can use to instrument our needs. The second step is the instrumentation of the mock indeed, and we should instrument the countUserPortfoliosMock in the following way:
every {
with(countUserPortfoliosMock) {
countUserPortfoliosMock.countByUserId(UserId("bob"))
}
} returns 0
The above code translates to the following sentence: “Every time we call the method countByUserId of the instance countUserPortfoliosMock with input equals to UserId("bob"), we’ll get the value 0 as a result.”. Despite that, we get an error if we try to compile it:
No context receiver for 'arrow.core.raise.Raise<in.rcard.arrow.raise.testing.DomainError>' found.
The compiler tells the truth. We defined the countByUserId function using the Raise<DomainError> context. We should remember that declaring a context receiver is like adding an implicit input parameter to the list of explicitly declared input parameters. So, the compiler tells us we’re not giving enough parameters for the function to be mocked.
We need to add the missing parameter with a matcher, as with any other input parameter. The only difference is that we need the Raise<DomainError> at the scope level. Then, we can use the with scope function once again:
every {
with(any<Raise<DomainError>>()) {
with(countUserPortfoliosMock) {
countByUserId(UserId("bob"))
}
}
} returns 0
Now, the compiler is happier, and we can proceed with the rest of the test code:
should("create a portfolio for a user (using mockk") {
val countUserPortfoliosMock: CountUserPortfoliosPort = mockk()
val underTestWithMock = createPortfolioUseCase(countUserPortfoliosMock)
every {
with(any<Raise<DomainError>>()) {
with(countUserPortfoliosMock) {
countByUserId(UserId("bob"))
}
}
} returns 0
val actualResult: Either<DomainError, PortfolioId> =
either {
with(underTestWithMock) {
createPortfolio(CreatePortfolio(UserId("bob"), Money(1000.0)))
}
}
actualResult.shouldBeRight(PortfolioId("1"))
}
There is another way to achieve the same result. We need a Raise<E> instance to mock our function. Using the matchers inside the scoped function with is one way. Another way is to provide an actual instance of a Raise<E>. As we might know, the only way to retrieve one is using the Raise.fold function or one of the builder methods, like the Raise.either we saw. In our case, we were already using the either function. So, it’s enough to extend its scope:
should("create a portfolio for a user (using mockk and either") {
val countUserPortfoliosMock: CountUserPortfoliosPort = mockk()
val underTestWithMock = createPortfolioUseCase(countUserPortfoliosMock)
val actualResult: Either<DomainError, PortfolioId> =
either {
every {
with(countUserPortfoliosMock) {
countByUserId(UserId("bob"))
}
} returns 0
with(underTestWithMock) {
createPortfolio(CreatePortfolio(UserId("bob"), Money(1000.0)))
}
}
actualResult.shouldBeRight(PortfolioId("1"))
}
We left out only the case in which we want to mock the function raising an error. As you can imagine, it’s close to what we’ve just done. The main difference is that we cannot use the returns function on the MockK scope. We don’t want to throw an exception. So, the throws function is not suitable too. We need to use the more generic answers function indeed.
First, we need to add a new error to our errors’ hierarchy:
sealed interface DomainError {
data class PortfolioAlreadyExists(
val userId: UserId,
) : DomainError
data class GenericError(
val throwable: Throwable,
) : DomainError
}
The GenericError represents an unexpected error, such as a problem with the network. Then, we can implement the test:
should("return a PortfolioAlreadyExists error for an existing user") {
val countUserPortfoliosMock: CountUserPortfoliosPort = mockk()
val underTestWithMock = createPortfolioUseCase(countUserPortfoliosMock)
val exception = RuntimeException("Ooops!")
val actualResult: Either<DomainError, PortfolioId> =
either {
every {
with(countUserPortfoliosMock) {
countByUserId(UserId("bob"))
}
} answers {
raise(GenericError(exception))
}
with(underTestWithMock) {
createPortfolio(CreatePortfolio(UserId("bob"), Money(1000.0)))
}
}
actualResult.shouldBeLeft(GenericError(exception))
}
The above test verifies the behavior of the use case when there was an unexpected error during the execution of the port. We used the answers function, the most generic one available in MockK.
For completeness, we can translate the above test using Mockito to understand the differences between the two libraries. In detail, we’ll use the library mockito-kotlin on top of Mockito to have a more idiomatic look and feel:
@Test
fun `given a userId and an initial amount, when executed with error, then propagates the error properly`() {
val exception = RuntimeException("Ooops!")
val actualResult =
either {
val countUserPortfoliosPort =
mock<CountUserPortfoliosPort> {
on {
countByUserId(UserId("bob"))
} doAnswer {
raise(GenericError(exception))
}
}
with(createPortfolioUseCase(countUserPortfoliosPort)) {
createPortfolio(CreatePortfolio(UserId("bob"), Money(1000.0)))
}
}
EitherAssert.assertThat(actualResult).containsOnLeft(GenericError(exception))
}
As we can see, the only part that has changed is the DSL offered by the mocking library. The rest of the test remains more or less the same.
Conclusion
In this article, we saw how to test a function declared in a Raise<E> context. We introduced several approaches and used many libraries. We saw how to write a test using JUnit 5 and the Kotest testing frameworks. We saw how to use the assertj-arrow-core library to test functions using the Raise DSL. Then, we translated the same tests using the Kotest extension asserts for Arrow. Then, we focused on mocking. The first solution we approached was using fake objects. Despite some initial programming, the fake objects are natural and ergonomic. We also saw how to mock a function declared in a Raise<E> context using the MockK and Mockito libraries. Mocking feels less idiomatic and more complex due to the Raise<E> context, which neither libraries manage natively. As of today, we still miss some Mockito’s extensions functions to handle the Raise DSL natively, which would make the mocking process more straightforward.
Appendix
As we said, Kotlin context receivers were deprecated in more recent versions of Kotlin. For the above reason, we preferred to use something other than context receivers in this article. However, context receivers almost invariably make the code cleaner and more concise. We introduced it in the article Kotlin 101: Context Receivers Quickly Explained. Let’s see how our example changes if we use context receivers.
We need to enable the usage of context receivers since they’re still an experimental feature in Kotlin 2.0.0. We need to add the following code to the build.gradle.kts file to do so:
tasks.withType<KotlinCompile>().configureEach {
compilerOptions {
freeCompilerArgs.add("-Xcontext-receivers")
}
}
Please take care that the way we can provide options to the compiler has changed since the previous article we wrote on context receivers.
First of all, we need to change the CreatePortfolioUseCase interface to use context receivers:
interface CreatePortfolioUseCase {
context(Raise<DomainError>)
fun createPortfolio(model: CreatePortfolio): PortfolioId
}
fun createPortfolioUseCase(): CreatePortfolioUseCase =
object : CreatePortfolioUseCase {
context(Raise<DomainError>)
override fun createPortfolio(model: CreatePortfolio): PortfolioId =
PortfolioId("1")
}
Easy-peasy. Now, we need to change the code to use the new version of the use case. Now, we can call the createPortfolio function directly on the CreatePortfolioUseCase type. Here is how the JUnit 5 test changes:
internal class CreatePortfolioUseCaseJUnit5Test {
private val underTest = createPortfolioUseCase()
@Test
internal fun `given a userId and an initial amount, when executed, then it create the portfolio`() {
val actualResult: Either<DomainError, PortfolioId> =
either {
underTest.createPortfolio(CreatePortfolio(UserId("bob"), Money(1000.0)))
}
Assertions.assertThat(actualResult.getOrNull()).isEqualTo(PortfolioId("1"))
}
}
No more call to with, no more scope function!
From here on, everything should run quite smoothly. Happy coding!