[Haskell-cafe] Parallel interruptible computations

[David Sorokin]

Corentin,

Only in my case the computations are executed ultimately from the event queue and they are bound up with the modeling time. If the time as a factor is not needed in your case, then this is not a problem to exclude the queue from consideration. The main idea is how we can use continuations, one of which is invoked when canceling (interrupting) the computation.

David

08 сент. 2014 г., в 5:53, David Sorokin <david.sorokin at gmail.com> написал(а):

> Corentin,
> 
> As far as I understand, it is similar in implementation to what I have in my simulation library Aivika [1]. Please correct me if I am wrong:
> 
> -- | Execute the specified computations in parallel within
> -- the current computation and return their results. The cancellation
> -- of any of the nested computations affects the current computation.
> -- The exception raised in any of the nested computations is propagated
> -- to the current computation as well.
> processParallel :: [Process a] -> Process [a]
> 
> It looks like that my Process type is an equivalent of your type Event in some sense. Only here we would have to cancel (or, interrupt in your terms) all other Process computations right after we receive a final result.
> 
> So, if it is true then it is possible to write a new function based on the stated above so that it would be similar to the ShortcutEvents function:
> 
> shortcutProcesses: [Process a] -> ([Maybe a] -> Maybe b) -> Process b
> 
> Probably, I should add such a function to my library too.
> 
> Thanks,
> David 
> 
> [1] http://hackage.haskell.org/package/aivika
> 
> 
> 08 сент. 2014 г., в 0:23, Corentin Dupont <corentin.dupont at gmail.com> написал(а):
> 
>> Hi guys,
>> thanks for the nice answers!
>> I'll give you a little bit more context: I'm designing an event engine. I have instances for Applicative, Alternative, Monad, MonadPlus.
>> It's 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.
>>    BindEvent      :: Event a -> (a -> Event b) -> Event b     -- The first event should fire, then a second event is created using the result.
>>    BaseEvent      :: BaseEvent a -> Event a                   -- Embed a base event.
>>    ShortcutEvents :: [Event a] -> ([Maybe a] -> Maybe b) -> Event b -- The function is called each time an event fires, as soon as the result can be computed from the available data, it is returned, dismissing the events that haven't fired yet.
>> 
>> 
>> instance Functor Event where
>>    fmap f a = pure f <*> a
>> 
>> instance Applicative Event where
>>    pure = PureEvent
>>    (<*>) = AppEvent
>> 
>> instance Alternative Event where
>>    (<|>) = SumEvent
>>    empty = EmptyEvent
>> 
>> instance Monad Event where
>>    (>>=) = BindEvent
>>    return = PureEvent
>> 
>> instance MonadPlus Event where
>>    mplus = SumEvent
>>    mzero = EmptyEvent
>> 
>> 
>> The Applicative instance is good if you have two events and you want both of them to fire ("and"). The Alternative instance is good if you have two events and you need only one to fire ("or").
>> But what if you have several events, but you need only a part of them to fire in order to construct a final result? Say you have 10 events, but the 5 first to fire will give you enough data to construct a result.
>> You cannot do that with Applicative/Alternative because with Applicative, you need *all* events results, with Alternative you need *only one*.
>> 
>> That's why I added this primitive "ShortcutEvents" in my DSL, but I'm not convinced by it. So my questions are:
>> 1. is ShortcutEvents expressible in term of Applicative/Alternative/Monad/MonadPlus?
>> 2. if not is their a well known typeclass that covers this case?
>> 3. if not is their a better way to write it? I especially don't like the list of Event, I'd prefer a more generic writing. What if I want a structure containing the events, instead of a list? What if I want event of various types (say a pair (Event a, Event b) for example)?
>> 
>> 
>> Note that I'm not working with streams of events (like in traditional FRP frameworks): just with single events (the "BaseEvents") that I want to combine with each other. Those "BaseEvents" will fire only once. The final result of the combination of events will trigger a callback.
>> 
>> Cheers,
>> Corentin
>> 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.haskell.org/pipermail/haskell-cafe/attachments/20140908/0d9c5836/attachment.html>