I have a coroutine dispatcher `d` that I have obta...

# coroutines

[Ulrik Rasmussen]

I have a coroutine dispatcher d that I have obtained by calling .asCoroutineDispatcher() on a single-threaded executor service. I want to launch long-running coroutines on this dispatcher from synchronous code. The obvious choice seems to be runBlocking(d) { launch { ... } }, but that blocks the thread until the job launched by launch { } completes. Can I submit a long-running job to d from synchronous code without blocking the current thread?

[uli]

You need to create a val scope : CoroutineScope

class XXX {
  val scop = CoroutineScope(SupervisorJob() + d)

  fun launchJob() { 
    scope.launch { ... }
  }

  fun dispose() {
    scope.cancel()
  }
}

The CoroutineScope will tell it’s coroutines, which dispatcher to use, how to handle exceptions and most noteably to cancel the potentially whole tree of child coroutines.

Disclaimer: Might not even compile.

[Ulrik Rasmussen]

Thank you, I'll try that :)

[uli]

Please read about scopes. I can not say if above setup of the scope is correct for you.

[Ulrik Rasmussen]

Me neither, I'll read up. Thanks for the pointer!

[uli]

oops, i think ther is no SupervisorScope. I’ll change it above

[Ulrik Rasmussen]

Why the SupervisorJob, and why the d?

[Ulrik Rasmussen]

I assume SubervisorJob is necessary because launch will fail if there is no parent job in the scope?

[uli]

not sure if it will fail. but `scope.cancel()`will actually call cancel on scope’s Job which will then cancel all it’s children. So if you do not set a job, cancelation aka structured concurrency will not work.

[Ulrik Rasmussen]

Nevermind, I should just read the docs. Thanks!

[Ulrik Rasmussen]

Ahh, I see. Thanks!

[Ulrik Rasmussen]

Thanks, this was exactly what I needed, and it turned out to solve another problem I have had for some time! I can now accomplish what I want without launching any threads at all - I even succeeded with replacing my dispatcher d with a custom serial dispatcher that just runs all its jobs on the current thread when a custom .run() method is called on it. I am using coroutines to simulate nodes in a network that pass messages back and forth, and this allows me to make the simulation completely deterministic and allows the detection of deadlocks immediately without the use of timeouts.