Hello! I’m developing a Mobile SDK using KMM and I...

# multiplatform

[Pat Teruel]

Hello! I’m developing a Mobile SDK using KMM and I would like to confirm how “dangerous” is it to use GlobalScope.async? Or is there an alternative way to do a DispatchQueue.global().async on KMM?

My problem is, KMM doesn’t have a close-to-native way of Scheduling tasks without freezing the Suspend function. So I had to create a class that’s “invalidatable”, similar to that of the Timer in Swift.

import kotlinx.coroutines.GlobalScope
import kotlinx.coroutines.async
import kotlinx.coroutines.delay

class AsyncSchedulerCaller(
    delayTime: Long,
    suspendFunction: suspend () -> Unit
) {
    init {
        GlobalScope.async {
            delay(delayTime)
            if (invalidated) {
                println("Invalidated.")
            } else {
                isRunning = true
                suspendFunction()
                isRunning = false
            }
        }
    }
    var invalidated = false
    var isRunning = false

    fun invalidate() {
        invalidated = true
    }
}

The way I use this is I store it in an optional variable and invalidate if it exists and create a new one, e.g.

class SomeClass { 
    var scheduler: AsyncSchedulerCaller? = null 

    suspend fun someSuspendFunction() { 
        runSomeApiCallHere() 

        scheduler?.invalidate() // this will make sure that the existing scheduled task will not be run
        scheduler = AsyncSchedulerCaller( 
            delayTime = 500, // will run after 500 milliseconds
            suspendFunction = {
                runSomeOtherApiCallHere() 
            }
        )
    }
}

To be honest, it already runs properly in iOS, and the scheduled task seem to run in the background. But I want to understand why it does, else, I might encounter bugs and I wouldn’t know how to fix.

That said, it works in iOS, but I haven’t tested in Android. Would this kind of code be dangerous on the app’s lifecycle in Android?
Thanks for the answers.

[Jeff Lockhart]

GlobalScope.async (although in this case I'd recommend simply GlobalScope.launch, as you're not needing to await an async returned value) is pretty similar to iOS's DispatchQueue.global().async, except it uses Kotlin's coroutines instead of GCD queues. It's only "dangerous" in that it doesn't take advantage of Kotlin coroutine's structured concurrency. `GlobalScope`'s lifecycle is the life of your app. So the coroutine's async work isn't automatically cancelled at the end of a more intentionally defined `CoroutineScope`'s lifecycle.

Your code does work, but the coroutine lives longer than it needs to before determining it's no longer needed. You could do something like this instead, and explicitly cancel the running coroutine when it's invalidated:

class AsyncSchedulerCaller(
    delayTime: Long,
    suspendFunction: suspend () -> Unit
) {
    private val job = GlobalScope.launch {
        delay(delayTime)
        isRunning = true
        suspendFunction()
        isRunning = false
    }

    var isRunning = false
        private set

    fun invalidate() {
        job.cancel()
        println("Invalidated.")
    }
}

If invalidate() is being called from iOS, this may be the most straightforward solution. Alternatively, if you're able to define a `CoroutineScope` that is `cancel()`ed when you'd be invalidating, you could pass the scope and have the coroutine cancelled for you:

class AsyncSchedulerCaller(
    delayTime: Long,
    scope: CoroutineScope,
    suspendFunction: suspend () -> Unit
) {
    private val job = scope.launch {
        try {
            delay(delayTime)
        } catch (e: CancellationException) {
            println("Invalidated.")
            throw e
        }
        isRunning = true
        suspendFunction()
        isRunning = false
    }

    var isRunning = false
        private set
}

The advantage of using CoroutineScope is that you can run multiple coroutines within the scope and they will all be cancelled when the scope is cancelled at the end of its lifecycle.

[Big Chungus]

Isn't freezing obsolete with new mm?

[Pat Teruel]

@Jeff Lockhart Thanks for this comprehensive answer! I did realize I could actually store the async value into a variable and realize it's a Job type, so yeah I pretty much added the same code you did here.

Another question though. Since I don't want the coroutine to live longer than it should, can I actually just "invalidate", or cancel the job INSIDE the coroutine scope after the suspend function is executed completely? or is it a bad practice to do that?

[Jeff Lockhart]

Once a coroutine builder's suspend function completes, the coroutine is finished as well. So there's no need to explicitly cancel it after suspendFunction() completes, if I understand your question correctly.

There's nothing wrong with cancelling a coroutine job from within its suspend function. But if you do need to cancel a coroutine early from within its own suspend function, it's generally easier to just return early. Cancelling a coroutine's job only works to cancel a coroutine while it's in a cancellable state, where the code is actively checking the cancelled state of the coroutine and returning early or throwing CancellationException in response. Well behaving long-running suspend functions, such as delay(), will do this for you. You can also simply throw CancellationException at any point in a running coroutine to cancel it. See docs on coroutine cancellation.

Martynas brings up a good question. You mention needing to freeze your suspend function. Are you not using the new Kotlin/Native memory model?

[Pat Teruel]

By freezing, I meant that the Suspend function has to wait for the nested suspend function to finish before the suspend function itself is finished.

e.g.

suspend fun a() {
    b()
}
suspend fun b() {
    delay(1000)
}

When I do it this way, a() will not be finished until b() is finished.
What I wanted to do, as I already mentioned on this thread, is to not have a() wait for b() to finish, and instead schedule b() to be executed IF there’s no “new” b() to be executed.
so:

suspend fun a() {
   existingCaller?.invalidate()
   existingCaller = AsyncCaller {
        b()
   } 
}

Maybe the terms I used is wrong. But anyway, I’m using the new memory model because it’s awesome. That’s the one in the gradle properties, right?

#Native
kotlin.native.binary.objcExportSuspendFunctionLaunchThreadRestriction=none

[Jeff Lockhart]

The new memory model gradle property is:

kotlin.native.binary.memoryModel=experimental

But it's enabled by default since Kotlin 1.7.20.

The property

kotlin.native.binary.objcExportSuspendFunctionLaunchThreadRestriction=none

is an additional option that can be enabled with the new memory model though.

The term "freezing" was just a bit confusing because the old memory model required freezing objects as read-only in order to access an object from multiple threads.

I understand your intention is just to be able to replace a previously scheduled suspend routine with a new routine. 👍🏼