So I've been experimenting with fixed points lately and have finally struggled through regular fixed points enough to discover some uses; now I'm moving onto comonadic fixed points and I'm afraid I've gotten stuck;
Here's a few examples of what I've tried and what has/hasn't worked:
{-# language DeriveFunctor #-}
{-# language FlexibleInstances #-}
module WFix where
import Control.Comonad
import Control.Comonad.Cofree
import Control.Monad.Fix
So I started with loeb's theorem as a list; each element of the list is a function which takes the end result to compute its answer; this lets me do 'spreadsheet' calculations where values can depend on other values.
spreadSheetFix :: [Int]
spreadSheetFix = fix $ \result -> [length result, (result !! 0) * 10, (result !! 1) + 1, sum (take 3 result)]
Okay, so I have basic fix working, time to move on to the comonad types! Here's a few simple comonads to use for examples:
data Stream a = S a (Stream a)
deriving (Eq, Show, Functor)
next :: Stream a -> Stream a
next (S _ s) = s
instance Comonad Stream where
extract (S a _) = a
duplicate s@(S _ r) = S s (duplicate r)
instance ComonadApply Stream where
(S f fs) <@> (S a as) = S (f a) (fs <@> as)
data Tape a = Tape [a] a [a]
deriving (Show, Eq, Functor)
moveLeft, moveRight :: Tape a -> Tape a
moveLeft w@(Tape [] _ _) = w
moveLeft (Tape (l:ls) a rs) = Tape ls l (a:rs)
moveRight w@(Tape _ _ []) = w
moveRight (Tape ls a (r:rs)) = Tape (a:ls) r rs
instance Comonad Tape where
extract (Tape _ a _) = a
duplicate w@(Tape l _ r) = Tape lefts w rights
where
lefts = zipWith const (tail $ iterate moveLeft w) l
rights = zipWith const (tail $ iterate moveRight w) r
instance ComonadApply Tape where
Tape l f r <@> Tape l' a r' = Tape (zipWith ($) l l') (f a) (zipWith ($) r r')
Okay so the following combinators come from Control.Comonad;
wfix :: Comonad w => w (w a -> a) -> a
wfix w = extract w (extend wfix w)
cfix :: Comonad w => (w a -> a) -> w a
cfix f = fix (extend f)
kfix :: ComonadApply w => w (w a -> a) -> w a
kfix w = fix $ \u -> w <@> duplicate u
I started with trying out wfix:
streamWFix :: Int
streamWFix = wfix st
where
incNext = succ . extract . next
st = (S incNext (S incNext (S (const 0) st)))
> streamWFix
-- 2
This one seems to work by calling the first w a -> a
on w until reaching
a resolution const 0
in this case; that makes sense. We can also do this
with a Tape:
selfReferentialWFix :: Int
selfReferentialWFix = wfix $ Tape [const 10] ((+5) . extract . moveLeft) []
-- selfReferentialWFix == 15
K, I think I get that one, but the next ones I'm kind of stuck, I don't seem to have an intuition for what cfix is supposed to do. Even the simplest possible thing I could think of spins forever when I evaluate it; even trying to extract the first element of the stream using getOne fails.
getOne :: Stream a -> a
getOne (S a _) = a
simpleCFix :: Stream Int
simpleCFix = cfix go
where
go _ = 0
Similarly with kfix; even simple tries don't seem to terminate. My understanding of kfix was that the function in each 'slot' gets passed a copy of the evaluated comonad focused on that spot; is that the case?
I tried using 'getOne' on this:
streamKFix :: Stream Int
streamKFix = kfix st
where
go _ = 0
st = S go st
Here's a finite attempt using Tape which also fails to run:
tapeKFix :: Tape Int
tapeKFix = kfix $ Tape [] (const 0) []
So; down to my question, could someone please offer some runnable (non-trivial) examples of using cfix and kfix, and explain how they work? I plan to use kfix to eventually do a "Conway's game of life" style experiment, am I correct in thinking that kfix would be useful in working with neighbourhoods around a given cell?
Feel free to ask any clarifying questions and help me expand my knowledge and intuition of fix!
Thanks!