Blocking code is not always properly handled

[ghik]

Reproduction

import monix.eval.Task
import monix.execution.Scheduler

object Bug extends App {
  implicit val scheduler: Scheduler = Scheduler.forkJoin(1, 128)

  val task = Task.async[Unit](_.onSuccess(())).map { _ =>
    Task.evalAsync(println("foo")).runSyncUnsafe()
  }
  task.executeAsync.runSyncUnsafe()
}

Current behaviour

Deadlock

Expectation

No deadlock. The blocking .runSyncUnsafe() call should prompt the underlying thread pool to allocate a new thread in place of the blocked one.

Analysis

The culprit is monix.execution.internal.Trampoline.trampolineContext which overrides the parent BlockContext that is responsible for wrapping the blocking code in ForkJoinPool.managedBlock

[alexandru]

Thanks for the analysis @ghik, needs to be investigated.

[ghik]

Hi, IMO this bug should target 3.x series - we needed to build and maintain our own version with the fix from #1544 included.

[alexandru]

@ghik thanks. I merged the PR. Will plan for a new 3.x release.