Override munit ExecutionContext with one from IORuntime

[milanvdm]

Closes #110

This overrides the default munit parasitic-execution-context with the one from IORuntime.global.

Not sure if tests can be added for ScalaJS on the issue that @djspiewak mentions as I have no ScalaJS knowledge.

[milanvdm]

The tests seem to be flaky. I will investigate further tomorrow as there is no clear test failing as far as I can tell.

[milanvdm]

This PR is currently blocked by a bug in cats-effect in the blocking part of the compute pool. I will follow up once fixed.

[danicheg]

LGTM. The question is for my understanding rather than any block on this.

[danicheg]

I don't get why was val changed on def here. IORuntime.global is a val though.

[milanvdm]

When offering an interface to a user, they might want to overwrite this munitIoRuntime with a def.

The same reason that munit itself has the munitExecutionContext as a def on the class we are overwriting.

[danicheg]

I get it, thanks. Seems reasonable 👍🏻

[vasilmkd]

@milanvdm @armanbilge Maybe it is best to revert this change and wait until Cats Effect 3.3 releases to incorporate it.

[vasilmkd]

On the other hand, maybe it is not wise to have this change on the JVM at all. Every time a Future is awaited on the compute pool, it would send a signal to the pool to start new threads. And in huge test suites, that could result in thousands of threads being created. I don't know what exactly to think.

[vasilmkd]

Please also don't rush to make changes, we'll continue to monitor the possible implications of this PR.

[armanbilge]

every time a Future is awaited on the compute pool, it would send a signal to the pool to start new threads. And in huge test suites, that could result in thousands of threads being created.

IIUC from our Discord discussion, there is (at least) one Future created per test, when the IO is converted.

The munit docs say that

MUnit does not support running individual test cases in parallel. However, sbt automatically parallelizes the execution of multiple test suites.

So, won't the number of these per-test Futures be bounded by the overall parallelism level that sbt is using?

[vasilmkd]

I published cats-effect and munit-cats-effect-3 local versions which print when new threads are started, and this seems to not spawn anything in coreJVM/test on fs2, so we're good. I guess I underestimated how these things work. Let's keep it for now.

[ahjohannessen]

This change seems to fail fs2-grpc server streaming tests with regards to cancellation. Not sure what is happening, but restoring the previous execution context in Fs2GrpcSuite makes tests pass again:

  private val parasiticExecutionContext = new ExecutionContext {
    def execute(runnable: Runnable): Unit = runnable.run()
    def reportFailure(cause: Throwable): Unit = cause.printStackTrace()
  }

  override val munitExecutionContext: ExecutionContext = parasiticExecutionContext

[vasilmkd]

@ahjohannessen Thanks for the pointer, I will investigate.