I'm going over continuations and I've come across two different approaches to structuring continuation types:
newtype C r a = C {runC :: (a -> r) -> r}
exampleFunction :: String -> C Bool String
exampleFunction s = C $ \t -> if length s > 10 then t s else False
continuationFunction :: String -> Bool
continuationFunction s = True
main = do
let suspendedFunc = exampleFunction "testing"
let completedFunc = runC suspendedFunc $ continuationFunction
versus the approach taken in Poor Mans Concurrency:
type C r a = (a -> r) -> r
exampleFunction :: String -> C Bool String
exampleFunction s = \t -> if length s > 10 then t s else False
...
I understand that the latter approach doesn't use an explicit data constructor.
Will this impact when I try to use this over a general type with a monad? Such as:
data Hole = Hole1 Int | Hole2 String
type C r m a = (a -> m r) -> m r
exampleFunction :: String -> C Bool Maybe Hole
exampleFunction s = \t -> do
x <- t (Hole1 11)
y <- t (Hole2 "test")
...
continuationFunction :: Hole -> Bool
continuationFunction (Hole1 x) = False
continuationFunction (Hole2 y) = True
The differences are the usual differences between type
and newtype
.
A type
synonym is just a new name for an existing type. type
synonyms can't be partially applied, because the compiler expands the definition during type checking. For example, this is no good, even with TypeSynonymInstances
:
type TypeCont r a = (a -> r) -> r
instance Monad (TypeCont r) where -- "The type synonym ‘TypeCont’ should have 2 arguments, but has been given 1"
return x = ($ x)
k >>= f = \q -> k (\x -> (f x) q)
newtype
s, while operationally equivalent to the types they wrap, are separate entities in the type system. This means that newtype
s can be partially applied.
newtype NewtypeCont r a = Cont { runCont :: (a -> r) -> r }
instance Monad (NewtypeCont r) where
return x = Cont ($ x)
Cont k >>= f = Cont $ \q -> k (\x -> runCont (f x) q)