Is it good practice, if you're in a suspending function, to use `withContext(Dispatchers.Default) { ...

# serialization

[Erik]

Is it good practice, if you're in a suspending function, to use withContext(Dispatchers.Default) { Json./*... some (de)serialization function*/() }, so that (de)serialization happens on a thread optimised for CPU-heavy work? Or is there built-in coroutine support in kotlinx.serialization, so that I can just call suspending versions of (de)serialiszation functions?

[CLOVIS]

Does your server have two kinds of cores for CPU and IO work?

[Erik]

This is on mobile Android devices, so there are likely multiple cores and certainly threads. Not sure about different kinds

[CLOVIS]

There is no standard dispatcher with just the "faster CPU cores", maybe the Android libraries have that somewhere?

[Erik]

I only know that the IO dispatcher on Android is optimised for IO work. The Main dispatcher is not to be blocked. And then there's the Default dispatcher for CPU work. And the coroutines lib smartly can switch between these dispatchers by (sometimes) not switching JVM threads underneath for efficiency.

So if I'm about to serialise something to post in a request, I think I could:

[Erik]

Edit: improved version below, ignore this message

val serialzedData = withContext(Dispatchers.Default) { Json.encodeToString(data) }
withContext(<http://Dispatchers.IO|Dispatchers.IO>) { post(serializedData) }

[CLOVIS]

Is your data so large that it makes a difference to serialize it somewhere else than the IO dispatcher?

[Erik]

Probably not 😅 though in this project I actually don't know the data being (de)serialized. It's dynamic
But I'm wondering still (hence my original question): is it a good idea to do this by default?

[Erik]

Note what I said earlier (and see the 'Implementation note' here: https://kotlinlang.org/api/kotlinx.coroutines/kotlinx-coroutines-core/kotlinx.coroutines/-dispatchers/-i-o.html):

The IO and Default dispatchers share underlying threads and do a best effort to prevent thread switches

[Erik]

Improved version:

withContext(Dispatchers.Default) {
    val serialzedData = Json.encodeToString(data)
    withContext(<http://Dispatchers.IO|Dispatchers.IO>) { post(serializedData) }
}

To ensure that post happens from within the IO dispatcher in the Default dispatcher, so that that optimisation kicks in

[Erik]

(And I know that the context switch to IO should probably be encapsulated inside of the post function, so that all these context switches are less visible and the code looks more streamlined)