I'm migrating a test to coroutines 1.6.0 which should check that a suspending function throws an exc...

# coroutines

[Lukasz Kalnik]

I'm migrating a test to coroutines 1.6.0 which should check that a suspending function throws an exception.
I'm using AssertJ for assertions.
This is what the test looked like before:

@Test
fun `when no gateway from scanning present then throw`() {
    every { state.gateway } returns null

    assertThatIllegalStateException().isThrownBy { runBlocking { useCase.execute(Params()) } }
}

This is what it looks like after the migration, according to the official migration guide:

val exceptions = mutableListOf<Throwable>()
    val customCaptor = CoroutineExceptionHandler { _, throwable ->
        exceptions.add(throwable)
    }

    @Test
    fun `when no gateway from scanning present then throw`() = runTest {
        every { state.gateway } returns null

        launch(customCaptor) {
            useCase.execute(Params())
        }
        advanceUntilIdle()

        assertThat(exceptions).hasSize(1)
            .first().isInstanceOf(IllegalStateException::class.java)
    }

However, the exception is for some reason not caught by the customCaptor, but thrown in the test and the test fails.

[Lukasz Kalnik]

I works when I simply replace runBlocking by runTest, but this is not recommended according to the migration guide (it could obscure other problems with the test detected by the runTest cleanup procedure).

[Lukasz Kalnik]

I don't use any custom dispatcher in the useCase under test. Should I inject one and replace it with the TestScope.coroutineContext for testing?

[Lukasz Kalnik]

The test still throws the exception even when I set a test dispatcher:

@BeforeTest
fun setUp() {
  Dispatchers.setMain(StandardTestDispatcher())
}

[Dmitry Khalanskiy [JB]]

Won't just

assertFailsWith<IllegalStateException> { useCase.execute(Params()) }

(without `launch`ing a new coroutine) work?
The reason the whole test fails with an exception is that the coroutine you launch fails with that exception. The uncaught exception handler is not used, because, well, the exception is not uncaught, it's propagated using structured concurrency.

[Lukasz Kalnik]

Yes, this indeed works. Thank you for your quick and helpful answer!

[Lukasz Kalnik]

But I still don't understand the explanation in the migration guide.

[Lukasz Kalnik]

It says:
"... valid uses of TestCoroutineExceptionHandler include:
Accessing uncaughtExceptions when the uncaught exceptions are actually expected. In this case, cleanupTestCoroutines will fail with an exception that is being caught later. It would be better in this case to use a custom CoroutineExceptionHandler so that actual problems that could be found by the cleanup procedure are not superseded by the exceptions that are expected. An example is shown below."

[Dmitry Khalanskiy [JB]]

I answered some of your questions under the issue you opened, but basically, the migration guide is for migration from the old test API, which, in this case, you weren't using at all.

[Lukasz Kalnik]

Yes, I understand this now.

[Lukasz Kalnik]

So the uncaught exceptions in the test would only be exceptions not thrown within a coroutine?

[Dmitry Khalanskiy [JB]]

"Uncaught exceptions" are the ones that nobody catches. Given that you managed to observe the exception being caught, it's not uncaught.

Uncaught exceptions are in general quite rare and can be considered an obscure edge case.

[Dmitry Khalanskiy [JB]]

You could create an uncaught exception like this:

launch(SupervisorJob()) {
  launch {
    throw IllegalStateException()
  }
}

[Lukasz Kalnik]

Yes, I guess I misunderstood that exceptions rethrown by coroutines are also uncaught

[Dmitry Khalanskiy [JB]]

Children of SupervisorJob don't have the right to fail the SupervisorJob, so they can't propagate the exception upwards, so they have to resort to uncaught exception handlers.

[Lukasz Kalnik]

Because for me "uncaught" meant kind of "crashing the test/program",

[Lukasz Kalnik]

Thank you very much for the explanation!

[Dmitry Khalanskiy [JB]]

You're welcome! Can the issue be closed then?

[Lukasz Kalnik]

Yes, I just closed it.