In a non lifecycle class, how should we use coroutines? I have read that your class should extend Co...

# android

[kd]

In a non lifecycle class, how should we use coroutines? I have read that your class should extend CoroutineScope if it is a lifecycle class (Activity, fragment, viewmodel). But in just a simple business logic class, how would you use CoroutineScope?

[Allan Wang]

If there's truly no lifecycle, then maybe something like global scope

[Allan Wang]

however, in reality, anything you do that is tied to the UI will have a lifecycle. If you are running things in a service, the service itself also has a lifecycle

[kd]

Good point that everything is eventually tied to a lifecycle, so we should pass the coroutine scope as a dependency to the non lifecycle classes?

[Allan Wang]

There are a few ways of doing things.

You can create functions that extend CoroutineScope so the caller supplies their own scope.

You can also write suspension functions that either checks for cancellation or is wrapped by a cancellable coroutine.

Coroutines need to cooperate to be cancelled, but for the most part you don't really have to worry about explicitly cancelling it within the code

[Allan Wang]

For me personally, I use a lot of suspension functions, and I declare the context from which it runs (IO vs UI). Switching contexts will already halt if the scope is cancelled. Otherwise, you can use yield(), which allows other coroutines to run and will cancel if the job is cancelled

[ghedeon]

yeah.. seems like the official suggestion is to proxy your CoroutineScope all the way down, from UI to Data... The code starts looking really strange tho.. If you're lucky to have only coroutineScope{} use cases than its cleaner and suspend is enough.

[kd]

Great, thanks for the comments guys!

[gildor]

You need to pass scope only if you run some background tasks, otherwise suspend and coroutineScope as @ghedeon mentioned.
Also, not sure why API will look strange, this is usual thing without coroutines. If you don’t want leak background job you have to manage lifecycle somehow

[Davide Giuseppe Farella]

Basically:
* I wanna wait for this function to complete and, eventually, get a result suspend fun doJob() { }
* same as above, but I need that function to do some works asynchronously suspend fun doJob() = coroutineScope { }
* I wanna launch something and don't wait for it to complete fun CoroutineScope.doJob() = launch { }

[gildor]

yes, or just:

fun doJob(scope: CoroutineScope) = scope.launch {  }

But launch probably is a bad example, I don’t see use cases for launch. Only valid case what I see it’s actor{} or produce{}

[Davide Giuseppe Farella]

I didn't need that yet, yesterday I dealt with it for the first time and launch seems to fit my use case, I'll take a look at those apis 🍻

[gildor]

nono

[gildor]

I don’t think that you shoul use actor or produce isntead of launch

[gildor]

I just talking about your example:

fun CoroutineScope.doJob() = launch {  }

[Davide Giuseppe Farella]

Yep, me too 😋

[gildor]

instead of launch, just use common suspend function to do job, and if client code want to launch it as task, just wrap it to launch inside your scope

[Davide Giuseppe Farella]

operator fun CoroutineScope.invoke(
            channels: SendChannel<IChannel>,
            groups: SendChannel<ChannelGroup>,
            errors: SendChannel<ChannelError>
    ) = launch { /* do smtng and deliver to Channels */ }

[Davide Giuseppe Farella]

That's my use case, I used launch since I don't want to wait for this function tu complete, once is called

[gildor]

But why?

[gildor]

I see what you mean. Channels ccase

[gildor]

Yes, this is valid case. also you can use produce{} for this

[gildor]

But this API looks a bit strange for me

[Davide Giuseppe Farella]

Which one? You mean produce or my function?

[gildor]

invoke for coroutine scope

[Davide Giuseppe Farella]

Oh ok. That's just a parser class, that makes sense to have only a function that is invoke, since the class purpose it's already clear 😁. What is weird for me is the way to call the api parser.run { this@scope( /* channels */ ) }. Still working on it 😁

[Davide Giuseppe Farella]

It would be great to have launch { parser( /* channels */ ) } anyway I'm goin off topic 😋 I already posted that question on #general and ghideon told me that already thousands of people made the same question 😁

[gildor]

I believe this highly depends on use case

[gildor]

It would be great to have launch { parser( /* channels */ ) }

But you can have such API

[Davide Giuseppe Farella]

I can't figure out how 🤔
https://kotlinlang.slack.com/archives/C0922A726/p1548800405769100?thread_ts=1548800405.769100&cid=C0922A726

[kd]

@gildor and Davide, thanks for the discussions. They help a lot. Coming to unit testing coroutines.
How would you test
fun CoroutineScope.doJob() = launch { }
Runblocking?

[Davide Giuseppe Farella]

I successfully tested with runBlocking, AFAIK it's the best ( only? ) way

[kd]

Yup, was just curious if there is any other way.
One thing I have been recommended is to separate the launch and the actual code inside launch to a suspend method and just test the suspend method using runBlocking

[gildor]

runBlocking will work, but because launch doesn't return any result you have to check side effect, test suspend function is way easier, it's the same as testing side effect and pure functions

[gildor]

And again, in your place I would just consider do you really need launch (so some background job that does side effect only) or you can replace it with suspend function