Sometimes I find a need for `withScope` instead of...

# coroutines

[Dico]

Sometimes I find a need for withScope instead of withContext. Now you have to write withContext(myScope.coroutineContext) but, given the intended purpose 😉 of CoroutineScope, I feel there should be a scope variant of it.

[Dias]

is it not what coroutineScope {} does?

[Dico]

You cant pass an explicit scope to that

[Dico]

I feel one potential issue which is that you cant really have 2 parents, when ideally, withScope can give you 2 parents. I assume withContext makes the paraneter's Job take precedence. Just entertaining some thoughts.

[elizarov]

What would be your use-case for two parents?

[Dico]

That's a good question. I think the reason is that you have to drop one relationship when you want to change to another CoroutineScope. Maybe that's how it should be, and maybe that's why it's good that withScope doesn't exist.

[elizarov]

But what is the use-case? Where it comes up?

[Dico]

I have an application where I'm rendering using OpenGL. The application is slowed down significantly by a particular aspect of it because it can make around 2 million queries to the memory of a particular process per frame. Hitting 20ms of frame time is not ideal. All of that work could be done on another thread, as it's just creating vertices in a buffer.
That job should ideally be dependent on both scopes being alive; the main scope and the one I want to add.

[elizarov]

What’s your main scope and which one you want to add?

[Dico]

My main scope is the one with the opengl context, it wraps an event loop. The new one would be on the default dispatcher.

[Dico]

I think a good solution to that would be to cancel the new scope when the main scope dies

[Dico]

Anyway, I'm just giving the idea of multiple parents a thought, I'm not necessarily advocating for it. I think it would bring many problems with it that would ultimately make it a bad idea.

[Dico]

What I end up doing each loop in the second thread, at the end is this:

MainScope.launch {
    vertexBuffer.sync() // upload to GPU
}.join() // ensure it's uploaded before next iteration

[Dico]

This is OK. But withContext like mechanism would be more elegant. That's all.