[External] : Re: A new build and a new structured concurrency API

[Ron Pressler]

> On 17 Nov 2021, at 21:48, Alex Otenko <oleksandr.otenko at gmail.com> wrote:
> 
> Thanks for the great update.
> 
> I am trying to comprehend the implications of fork with onComplete.
> 
> The docs say onComplete may not be invoked at all. What is being saved here, compared to cancel() that guarantees onComplete will be invoked.
> 
> The other thing that seems odd is that onComplete is entirely side-effecting: it has no communication with the task that got completed (the task can't delegate responsibility for anything), and no communication with the consumer of Future (onComplete can't delegate responsibility for anything). Like, I borrow a connection - who's going to return it and how? I'd expect someone to have one last glance at the context of the Callable to clean up.
> 

Completion handlers allows you to factor out policies for the common and simple cases where you need to collect results or shutdown the executor session based on the task’s success or failure. A call to shutdown indicates that the computation is done — either successfully or unsuccessfully — and so there’s no point in processing further results. In more complicated — and, we believe, much rarer — cases, like the connection example in the javadoc, the completion handler is, indeed, insufficient, and you’d want to do cleanup processing inside the task and possibly call shutdown directly (shutdown is the concurrent analogue to a break statement in sequential loop).

— Ron