is the limitation on delay related to the granularity of the

Question

is the limitation on delay related to the granularity of the underlying worker pool dispatcher implementation

louiscad · Answer

Most OSes don t have true nanosecond precision so milliseconds was used It s also more natural to human mind You can still register a callback to execute after nanoseconds if the platform you re developing for has that capability despite it probably not having precision guaranteeds using `suspendCancellableCoroutine` like with any single execution cancellable unregisterable callback

brett.wooldridge · Answer

The relationship between `suspendCancellableCoroutine` and the ability to suspend for nanoseconds or microseconds is not clear to me

louiscad · Answer

You can write your own `delayNanos` function with it and an API from the platform that allows delaying with nanoseconds parameter

louiscad · Answer

single backticks will work for you try editing your message

brett.wooldridge · Answer

Ok I understand I was just curious why Kotlin did not choose to offer such a method Obviously all JVM based platforms have `parkNanos `` available obviously with actual accuracy depending on the underlying OS

brett.wooldridge · Answer

One question I have is that obviously `parkNanos ` is a blocking call but `delay ` is a suspending call so how is the delay itself achieved without invoking an underlying blocking platform call

octylFractal · Answer

remember that Kotlin is not a JVM only language it also supports JavaScript which does not offer nanosecond accuracy in most browsers iirc

brett.wooldridge · Answer

So that is the reason for its omission or is that supposition Just like `parkNanos `` makes no guarantees on granularity it seems a similar `delay `` could be offered with similar best effort contract

brett.wooldridge · Answer

after all all delays sleeps have variance except in realtime OSes so a `delay 20ms ` may not actually resume until tens hundreds or even thousands of milliseconds later

elizarov · Answer

No hidden thoughts Millis are simply the most common unit on JVM With the upcoming Duration support in stdlib you ll be able to request delay with any precision to exploit the full potential of your target platform

octylFractal · Answer

re `parkNanos ` is a blocking call but `delay ` is a suspending call it depends on the underlying implementation but typically the thread pool will do the blocking the difference being the thread pool can do other work that is ready while waiting for your task to finish delaying since it knows of all the tasks it needs to be executing essentially when you call `delay ` Kotlin schedules a task after the delay that calls back into your method to continue execution see the generated bytecode it basically is a switch statement to which part of the method it should be executing and suspending will just tell it to call the next part

brett.wooldridge · Answer

yeah i ve watched the videos on the underlying implementation of the resume dispatching

brett.wooldridge · Answer

It looks like the jvm implem entation of `CoroutineScheduler` already uses `parkNanos ` and does all internal math based on nanoseconds would be nice to have it exposed When is `Duration` expected to be added to stdlib 1 4