Using Maybe and Writer to filter a list and track filter hits

Question

I'm filtering a list by using chained functions that return Maybe element. This part works fine.

{-# LANGUAGE TypeSynonymInstances, FlexibleInstances, OverlappingInstances #-}
import Control.Monad
import Control.Monad.Trans.Maybe
import Control.Monad.Writer
import Data.Map (Map, alter, empty, unionWith)

------------------------------------------------

main = do
  let numberList = [1..6]
  let result = filter ((\z -> case z of Just _ -> True; Nothing -> False) . numFilter) numberList
  (putStrLn . show) result

{-
 [2,3,4]
-}

--- Maybe
bigOne :: Int -> Maybe Int
bigOne n | n > 1     = Just n
         | otherwise = Nothing

lessFive :: Int -> Maybe Int
lessFive n | n < 5     = Just n
           | otherwise = Nothing

numFilter :: Int -> Maybe Int
numFilter num = bigOne num
            >>= lessFive

But then I also want to count the times when different functions have caught an element. I'm now using a Writer with a Map to collect the hits. I tried wrapping this inside a MaybeT but this causes the whole filter to fail in case of an unwanted element and returns and empty list.

-------------------------------
type FunctionName = String
type Count = Int
type CountMap = Map FunctionName Count

instance Monoid CountMap where
  mempty = empty :: CountMap
  -- default mappend on maps overwrites values with same key,
  -- this increments them
  mappend x y = unionWith (+) x y

{-
  Helper monad to track the filter hits.
-}
type CountWriter = Writer CountMap

incrementCount :: String -> CountMap
incrementCount key = alter addOne key empty

addOne :: Maybe Int -> Maybe Int
addOne Nothing = Just 1
addOne (Just n) = Just (n + 1)

bigOneMW :: Int -> MaybeT CountWriter Int
bigOneMW n | n > 1     = MaybeT $ return (Just n)
           | otherwise = do
                          tell (incrementCount "bigOne")
                          MaybeT $ return Nothing

lessFiveMW :: Int -> MaybeT CountWriter Int
lessFiveMW n | n < 5     = MaybeT $ return (Just n)
             | otherwise = do
                           tell (incrementCount "lessFive")
                           MaybeT $ return Nothing

chainMWBool :: Int -> MaybeT CountWriter Bool
chainMWBool n = do
             a <- bigOneMW n
             b <- lessFiveMW a
             return True

chainerMW :: [Int] -> MaybeT CountWriter [Int]
chainerMW ns = do
               result <- filterM chainMWBool ns
               return result
{-
> runWriter (runMaybeT (chainerMW [1..3]))
(Nothing,fromList [("bigOne",1)])
> runWriter (runMaybeT (chainerMW [2..5]))
(Nothing,fromList [("lessFive",1)])
> runWriter (runMaybeT (chainerMW [2..4]))
(Just [2,3,4],fromList [])
-}

I just can't figure out how get it to do what I want. I guess the type signature I'm looking for is [Int] -> CountWriter [Int], but how to get a result like this when input is [1..6]:

([2,3,4], fromList[("bigOne", 1), ("lessFive", 2)])

Answer 1

You were closer than you realized when you said:

but how to get a result like this when input is [1..6]:

([2,3,4], fromList[("bigOne", 1), ("lessFive", 2)])

In other words, you want something that takes a list as an input and returns a list and a map as output:

newtype Filter a = Filter { runFilter :: [a] -> (CountMap, [a]) }

Why not just encode all of your filters directly using the representation you actually wanted:

import Data.List (partition)
import qualified Data.Map as M
import Data.Monoid

newtype CountMap = CountMap (M.Map String Int)

instance Show CountMap where
    show (CountMap m) = show m

instance Monoid CountMap where
    mempty = CountMap M.empty
    mappend (CountMap x) (CountMap y) = CountMap (M.unionWith (+) x y)

filterOn :: String -> (a -> Bool) -> Filter a
filterOn str pred = Filter $ \as ->
    let (pass, fail) = partition pred as
    in  (CountMap (M.singleton str (length fail)), pass)

bigOne :: Filter Int
bigOne = filterOn "bigOne" (> 1)

lessFive :: Filter Int
lessFive = filterOn "lessFive" (< 5)

We're missing one lass piece of the puzzle: how to combine filters. Well, it turns out that our Filter type is a Monoid:

instance Monoid (Filter a) where
    mempty = Filter (\as -> (mempty, as))
    mappend (Filter f) (Filter g) = Filter $ \as0 ->
        let (map1, as1) = f as0
            (map2, as2) = g as1
        in  (map1 <> map2, as2)

Experienced readers will recognize that this is just the State monad in disguise.

This makes it easy to compose filters using (<>) (i.e. mappend), and we run them just by unwrapping our Filter type:

ghci> runFilter (bigOne <> lessFive) [1..6]
(fromList [("bigOne",1),("lessFive",2)],[2,3,4])

This shows how often the best path is the most direct one!

Answer 2

Alright so the issue here is the use of short circuiting is nuking the CountMap your building up. A quick example

test :: MaybeT (Writer [String]) ()
test = do
       tell ["Blah"] >> mzero
       tell ["Blah"] >> mzero
       tell ["Blah"] >> mzero
       tell ["Blah"] >> mzero


Prelude> runWriter (runMaybeT test)
   (Nothing, ["Blah"])

See the problem?

Fixing it is pretty simple, just don't rely on short circuiting :)

Example*:

bigOneMW n | n > 1     = return True
           | otherwise = tell "bigOne" >> return False
lessFiveMW n | n < 5     = return True
             | otherwise = tell "lessFive" >> return False
chainMWBool n = liftM2 (&&) (bigOneMW n) (lessFiveMW n)
chainerMW ns = filterM chainMWBool ns

Now of course, the MaybeT layer is a bit pointless so we can just ditch that.

Happily that doesn't affect any of the above code.

*You'll notice that the tells are just using a plain string, to do this, I'm using a language extension OverloadedStrings and defined an instance of a typeclass IsString from Data.String. The code to make that work looks like this if you're curious:

instance IsString CountMap where
  -- This is the same as your incrementOne code
  -- Just a bit more reliant on higher order function and
  -- pointfree.
  fromString = flip (alter inc) empty
    where inc = maybe (Just 1) $ Just . (+1)

Whether or not you like that particular trick is up to you :)

Code after all is said and done: http://hpaste.org/88624