Propagate Java thread interruption in Dispatcher#unsafeRunSync

[kamilkloch]

Closes #4166.

[armanbilge]

We should also make a similar change for IO#unsafeRunSync.

[kamilkloch]

We should also make a similar change for IO#unsafeRunSync.

I got a bit lost trying to comprehend the nuanced

cats-effect/core/jvm/src/main/scala/cats/effect/IOPlatform.scala

Line 70 in 725a4cd

final def unsafeRunTimed(limit: FiniteDuration)(

[armanbilge]

I got a bit lost trying to comprehend the nuanced

Actually, maybe let's hold off on that change. I forgot that the specification for this method is a bit unusual.

cats-effect/core/jvm/src/main/scala/cats/effect/IOPlatform.scala

Lines 45 to 48 in 725a4cd

	* Similar to `unsafeRunSync`, except with a bounded blocking duration when awaiting
	* asynchronous results. As soon as an async blocking limit is hit, evaluation ''immediately''
	* aborts and `None` is returned. Note that this does not run finalizers, which makes it quite
	* different (and less safe) than other mechanisms for limiting evaluation time.

---

But if we did decide to move forward ...

You can factor out a package-private version of unsafeRunAsync that returns the Fiber.

cats-effect/core/shared/src/main/scala/cats/effect/IO.scala

Lines 979 to 992 in 725a4cd

	def unsafeRunAsync(cb: Either[Throwable, A] => Unit)(
	implicit runtime: unsafe.IORuntime): Unit = {
	unsafeRunFiber(
	cb(Left(new CancellationException("The fiber was canceled"))),
	t => {
	if (!NonFatal(t)) {
	t.printStackTrace()
	}
	cb(Left(t))
	},
	a => cb(Right(a))
	)
	()
	}

Then, you can use that for cancelation in unsafeRunTimed.

cats-effect/core/jvm/src/main/scala/cats/effect/IOPlatform.scala

Lines 79 to 85 in 725a4cd

	try {
	val result = blocking(queue.poll(limit.toNanos, TimeUnit.NANOSECONDS))
	if (result eq null) None else result.fold(throw _, Some(_))
	} catch {
	case _: InterruptedException =>
	None
	}

[kamilkloch]

Actually, maybe let's hold off on that change. I forgot that the specification for this method is a bit unusual.

If only you are ok with the resulting discrepancy in sematics between Dispatcher#unsafeRunSync and IO#unsafeRunSync, I would hold off as well. (Actually, for unsafeRunTimed the semantics is already different)

[armanbilge]

thanks!