On iOS, does the following make sense to start cor...

# multiplatform

[Marc Knaup]

On iOS, does the following make sense to start coroutines using whatever the current dispatch queue is?

private object CurrentQueueDispatcher : CoroutineDispatcher() {

	override fun dispatch(context: CoroutineContext, block: Runnable) {
		dispatch_async(dispatch_get_current_queue()) {
			block.run()
		}
	}
}

…

GlobalScope.launch(CurrentQueueDispatcher) { … }

Also, dispatch_async or dispatch_sync? The documentation doesn't state whether it's okay to block here or not.

[Marc Knaup]

So much for that:

kotlin.native.IncorrectDereferenceException: illegal attempt to access non-shared CurrentQueueDispatcher.$dispatch$lambda-0$FUNCTION_REFERENCE$1@39254a8 from other thread

[basher]

dispatch queue doesn't guarantee single thread. it only guarantees synchronization with respect to that queue

[Marc Knaup]

Yup, just noticed. Still trying to find a way to make coroutines work on iOS from a background queue or thread 🤔

[basher]

At the moment, I don't think it's possible to write a multi-threaded dispatcher in native because of native concurrency/mutability rules + coroutine internals not using frozen/thread-safe state/structures. Generally you have 2 options when trying to pull things across threads: freeze them or transfer them.

1. Transfer: You can't transfer objects passed to you in dispatcher overrides because they're owned by coroutine internals (will crash because transferring requires detaching the object from that thread, which can't happen if the objects are still owned)
2. Freeze: Can't freeze them because nearly all of them have internal mutable state, so an exception will be thrown as soon as their internals make a call that attempts a mutation

[Marc Knaup]

Thank you for the insight!

I don't need a multi-threaded dispatcher though. Just single-threaded in the background.

[Marc Knaup]

Main thread is too sensitive for JSON parsing and Kotlin/Native bridging.

[basher]

The 3rd option is thread confinement: use Worker + runBlocking to do coroutine-based work in a Worker, being careful not to allow coroutine things to leave the worker and then transfer (or freeze) the result of that work back when you're done

[Marc Knaup]

Alright, next try 😄
Compiling…

private object BackgroundWorkerDispatcher : CoroutineDispatcher() {

	private val worker = Worker.start()


	override fun dispatch(context: CoroutineContext, block: Runnable) {
		worker.execute(TransferMode.SAFE, { block }, { it.run() })
	}
}

[basher]

The block there is owned by the calling thread and coroutine internals. Passing it to the producer there will attempt to transfer it to the worker thread and crash (i expect)

[Marc Knaup]

Damn…

[Marc Knaup]

Hmm, no exception. Nothing is actually happening.

[basher]

possible it logged the exception in the console silently

[basher]

might happen if exception happens from the Worker thread

[Marc Knaup]

True. It doesn't even say anymore what object is the problem…

kotlin.IllegalStateException: Illegal transfer state

[basher]

https://github.com/JetBrains/kotlin-native/issues/2761

[Dico]

I commonly use runBlocking as an event loop

[Andy Victors]

So iOS implementation of coroutines will always run on main thread, do I get it right?

[Marc Knaup]

No, it actually works on the background. You just can't switch between threads which includes using Ktor. There are unsafe workarounds though.

[Andy Victors]

Right, that was my case - I wanted to ask "if using Ktor with coroutines on IOS will run in main thread?" and it looks like answer is still yes.

[Marc Knaup]

Well for me, with some UNSAFE involved, Ktor now runs on a background thread except for a brief timespan between NSURLSession delegate invocation and ktor coroutine invocation.

[svyatoslav.scherbina]

https://kotlinlang.slack.com/archives/C3PQML5NU/p1560273060108300?thread_ts=1560269181.105800&cid=C3PQML5NU

This is not entirely correct since you have worker.execute resulting Future leaking.

[Marc Knaup]

What do you mean by leaking here? I'm not using the Future at all so how can it leak?

[basher]

Ah well one thing to remember (which I often forget) is the final statement in the worker execute block will be returned as the result of the future returned by execute, which result in another object transfer across threads.

[basher]

Maybe that’s what is being referred to

[Marc Knaup]

Yeah, but I hold no reference to that Future so I don't see the leak 🤔
Maybe an unused Future leaks the result internally?

[svyatoslav.scherbina]

Future is a resource. It leaks if the result is not consumed.

[basher]

Even if the result is Unit?

[Marc Knaup]

There should be a warning in the IDE about that 🤔

So how do I free a `Future`'s resources without blocking the thread?

[svyatoslav.scherbina]

Even if the result is Unit?

AFAIK yes.

So how do I free a `Future`’s resources without blocking the thread?

Once future.state is COMPUTED you can free it with .consume {} or .value.
So you can add future to some registry which gets polled for finished futures.

See also https://github.com/JetBrains/kotlin-native/pull/2971 (will be included to 1.3.50).

[Marc Knaup]

So I would basically use executeAfter with a time interval of zero and don't need to take care of consuming the future anymore?

[basher]

That appears to be the case. Dang. I have some cleanup to do then 😕

[svyatoslav.scherbina]

So I would basically use executeAfter with a time interval of zero and don’t need to take care of consuming the future anymore?

Exactly.