is there anything wrong with doing cleanup like this in a flow builder block?

flow {
  for (i in collection) {
    try {
      emit(i)
    } finally {
      doCleanupOn(i)
    }
  }
}

is there a better way to accomplish this?

I'd say that this is the preferred way to do it

cool. and this will have the semantics that the suspension will only continue once whatever downstream of this call to emit is completed?

AFAIK, it's not a good practice to have your

emits

between

try-catch

actually thinking through this i don’t want the finally

There's an

onCompletion

operator

is onCompletion equivalent to that?

onCompletion

does the same as having your flow's collection inside a

try-finally

except it doesn’t have the element in scope

Yup

yeah exactly

Solved then 😛

assuming the semantics are what i said above, which is that all downstream operations on the flow complete before it resumes to doCleanupOn

No problem, happy to do so 😊

I think this thread is endemic of a problem with coroutines: they present as a way to solve the worst flow problems of concurrency, but do they? or like most things, do they just mean you have to find new ways to accommodate same?

I am not sure what you mean by that. In general they claim to solve the issues of non-sequential-like async programming, but being able to

emit(i)

and then immediately call

doCleanupOn(i)

is very sequential-like, so this seems to be an example of coroutines working well towards their goal

But there are still catches, and questions of when and where things will happen.. the means the idea of a deterministic sequence is not really there right?

I feel like that's more like an question of Flow in particular, which is separate from coroutines. And the model is fairly simple, downstream flows pass in a FlowCollector, the upstream calls

emit(i)

, and in

emit

, the downstream uses the element, processes it, etc. So clearly exceptions from downstream are thrown from

emit

.

catch

's implementation uses

collect

&

emit

,

collect

throws the errors from upstream generation and the

emit

calls, and special handling around emit figures out if exceptions thrown from

collect

are from up or downstream. It's not very complicated at all.

One thing is for sure: the solutions, are towers erected in sand. The architecture patterns from 2 years ago looked good, had good MVVM approach, they are mostly dust now, Compose ushered in a whole new approach. You don’t even have time to learn things before they are mothballed. My problem with Rx and its spawns (have not looked enough at flow yet) is that they obscure logic. The argument in the Meyer Bible OOSC is that exceptions are useful for that reason primarily, is pretty compelling, but here we are talking about how you have to figure exactly how to sequence and then gird for upstream emissions. I hope you ARE right because I do think some flow is necessary, but if it overwhelms the code and the logic is every 3rd line, we’re doomed. Might as well go back to C where they had to have if statements on each SDK call, and then block-based handlers for integer return types…

I don't think this is the correct way to do this. The assumption that the element is fully processed after

emit

returns does not apply to some flow operators, namely buffering ones. But it might work in your use case.

is there anything you would recommend short of threading the item through the whole flow?

I dont have an answer.

Hmm, but.. doesnt that presume that the subscriber doesnt switch threads?i.e. if he does, then cleanup will run before subscriber's code, which will then act on a closed instance