Hi Corentin,

If I undestand your idea, then it is possible to do with help of continuations wrapped in a monad computation, at least with help of two continuations: (1) the main branch of computation; (2) the immediate cancellation if the result is already known.

The approach is implemented in the F# async workflow (so called the asynchronous computation) and .. in my simulation library Aivika (to model the discontinuous process). Both are monads. Only like the F# developers, I use a plain cancellation without returning the meaningful result, but nothing prevents from returning some data. Also like F# my library supports the third continuation for handling exceptions, namely the IO exceptions in terms of Haskell.

So, the keywords are: canceling the F# asynchronous computation.

Best regards,
David Sorokin


20 июня 2014 г., в 16:59, Corentin Dupont <corentin.dupont@gmail.com> написал(а):

Nobody on that?
What I really want is: a computation that can be interrupted if enough result is already there.
I would be surprise if that concept doesn't exist already somewhere :)


On Thu, Jun 19, 2014 at 5:17 PM, Corentin Dupont <corentin.dupont@gmail.com> wrote:
Hi guys,
I am making a DSL for event composition (inspired from digestive-functor & reactive-banana) and I find myself wanting a primitive like that:

-- given a list of events, evaluate the function on those that have already fired.
-- returns an event yelding the result when ready.
ListEvent  :: [Event a] -> ([a] -> Maybe b) -> Event b 

Does this type tell you something? Is there a better abstraction that can encompass it?
My DSL is like that:

-- | Composable events
data Event a where
   SumEvent   :: Event a -> Event a -> Event a        -- The first event to fire will be returned
   AppEvent   :: Event (a -> b) -> Event a -> Event b -- Both events should fire, and then the result is returned
   PureEvent  :: a -> Event a                         -- Create a fake event. The result is useable with no delay.
   EmptyEvent :: Event a                              -- An event that is never fired.
   BaseEvent  :: (Typeable a) => Field a -> Event a   -- Embed a base event

It's easy to define instances for Applicative and Alternative.
But then I have a use case that I cannot solve:
In the case of votations (which use this DSL), sometime you can end prematuraly the voting, even if not everybody voted yet: for example if the majority is already reached.
In this case no need to wait for everybody, we can end the vote.

This does not seem to be possible with Applicative or Monad interfaces: things have to be sequenced in a monad, they cannot be evaluated in parralel.
For example:

do
   v1 <- getVote1
   v2 <- getVote2
   v3 <- getVote3
   evalMajority(v1, v2, v3)

In this example I have to wait for all 3 votes to be completed, but in fact only two would suffice to achieve majority.
That's why I added "ListEvent" to the DSL above which allow to achieve that but I'm wondering if there is a better/bigger abstraction to put it in.

Thanks!!
Corentin


_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe