问题
I think that I copied the following dist
function from Applicative Programming with Effects:
The paper prefaces this function with:
Have you noticed that sequence and transpose now look rather alike? The details that distinguish the two programs are inferred by the compiler from their types. Both are instances of the applicative distributor for lists
dist :: Applicative f => [f a] -> f [a]
dist [] = [[]]
dist (v : vs) = [(:) v (dist vs)]
However, I get the following compile-time error:
ghci> :l ApplicativePaper.hs
[1 of 1] Compiling Main ( ApplicativePaper.hs, interpreted )
ApplicativePaper.hs:12:20:
Could not deduce (a ~ f a)
from the context (Applicative f)
bound by the type signature for
dist :: Applicative f => [f a] -> f [a]
at ApplicativePaper.hs:10:9-39
`a' is a rigid type variable bound by
the type signature for dist :: Applicative f => [f a] -> f [a]
at ApplicativePaper.hs:10:9
Relevant bindings include
vs :: [f a] (bound at ApplicativePaper.hs:12:9)
v :: f a (bound at ApplicativePaper.hs:12:7)
dist :: [f a] -> f [a] (bound at ApplicativePaper.hs:11:1)
In the first argument of `(:)', namely `v'
In the expression: (:) v (dist vs)
Failed, modules loaded: none.
Please let me know what I'm doing wrong. Also, please provide an intuition for this function.
回答1:
In the paper, the double bracket 〚 f u1 … un 〛
is defined on page 4 to be the same as pure f <*> u1 <*> … <*> un
.
The definition of dist
in the text is
dist :: Applicative f ⇒ [f a] → f [a ]
dist [] = 〚 [] 〛
dist (v : vs) = 〚 (:) v (dist vs) 〛
Translating the other symbols into Haskell and making the substitution for the definition of 〚 … 〛
results in
dist :: Applicative f => [f a] -> f [a ]
dist [] = pure []
dist (v : vs) = pure (:) <*> v <*> (dist vs)
回答2:
Wrong brackets. Conor McBride and Ross Paterson use so called idiom brackets. Here is your definition in imaginary syntax (but it probably compiles with the Strathclyde Haskell Enhancement, I'm not sure):
dist :: Applicative f => [f a] -> f [a]
dist [] = (| [] |)
dist (v : vs) = (| (:) v (dist vs) |)
(| f x y z |)
elaborates to pure f <*> x <*> y <*> z
. So dist
defined right is
dist :: Applicative f => [f a] -> f [a]
dist [] = pure []
dist (v : vs) = (:) <$> v <*> dist vs -- f <$> x == pure f <*> x
There is a very similar function, defined in the Control.Monad
module:
-- | Evaluate each action in the sequence from left to right,
-- and collect the results.
sequence :: Monad m => [m a] -> m [a]
{-# INLINE sequence #-}
sequence ms = foldr k (return []) ms
where
k m m' = do { x <- m; xs <- m'; return (x:xs) }
Only it requires a monad instead of an applicative.
There is also a generalization of dist
, provided by the Traversable
typeclass:
sequenceA :: Applicative f => t (f a) -> f (t a)
So you have some t
(which is both a Functor
and a Foldable
) instead of []
here.
This description of sequence
-- | Evaluate each action in the sequence from left to right,
-- and collect the results.
is pretty exhaustive. For example
main = dist [print 3, print 4, print 5]
prints
3
4
5
Since every monad is an applicative, dist
works properly here with the IO monad. However there is no need in collecting the results here. That's why there is the sequence_
function, that returns ()
as the result.
It's easy to define dist_
:
dist_ :: Applicative f => [f a] -> f ()
dist_ [] = pure ()
dist_ (v : vs) = v *> dist_ vs
So you simply perform all actions in a list.
For example
main = do
print $ dist_ [Just 3, Just 4, Just 5]
print $ dist_ [Nothing, Just 3]
prints
Just ()
Nothing
because
Just smth1 *> Just smth2 == Just smth2
Nothing *> Just smth2 == Nothing
The only thing, that dist
adds to this machinery, is collecting results.
So
main = do
print $ dist [Just 3, Just 4, Just 5]
print $ dist [Just 3, Nothing]
prints
Just [3,4,5]
Nothing
Also, have a look at the Typeclassopedia.
来源:https://stackoverflow.com/questions/27350924/applicative-distributor-for-list-dist-function