[Haskell-beginners] Get rid of Maybes in complex types

[Sylvain Henry]

Hi,

You can use something similar to "Trees that grows" in GHC:

{-# LANGUAGE TypeFamilies #-}
{-# LANGUAGE StandaloneDeriving #-}
{-# LANGUAGE FlexibleContexts #-}
{-# LANGUAGE UndecidableInstances #-}

module Main where

import Data.Maybe

data Checked   = Checked   deriving (Show)
data Unchecked = Unchecked deriving (Show)

type family F a b :: * where
    F Unchecked b = Maybe b
    F Checked   b = b

-- data types are decorated with a phantom type indicating if they have 
been checked
-- in which case "Maybe X" are replaced with "X" (see F above)
data A c = A
    { a1 :: F c (B c)
    }

data B c = B
    { b1 :: F c (C c)
    }

data C c = C
    { c1 :: F c Int
    }

deriving instance Show (F c (B c)) => Show (A c)
deriving instance Show (F c (C c)) => Show (B c)
deriving instance Show (F c Int)   => Show (C c)

class Checkable a where
    check :: a Unchecked -> a Checked

instance Checkable A where
    check (A mb) = A (check (fromJust mb))

instance Checkable B where
    check (B mc) = B (check (fromJust mc))

instance Checkable C where
    check (C mi) = C (fromJust mi)

main :: IO ()
main = do
    let
       a :: A Unchecked
       a = A (Just (B (Just (C (Just 10)))))

       a' :: A Checked
       a' = check a
    print a
    print a'


$> ./Test
A {a1 = Just (B {b1 = Just (C {c1 = Just 10})})}
A {a1 = B {b1 = C {c1 = 10}}}


Cheers,
Sylvain


On 06/07/2017 10:12, Baa wrote:
> Hello Dear List!
>
> Consider, I retrieve from external source some data. Internally it's
> represented as some complex type with `Maybe` fields, even more, some
> of fields are record types and have `Maybe` fields too. They are
> Maybe's because some information in this data can be missing (user
> error or it not very valuable and can be skipped):
>
>    data A = A {
>      a1 :: Maybe B
>      ... }
>    data B = B {
>      b1 :: Maybe C
>      ... }
>
> I retrieve it from network, files, i.e. external world, then I validate
> it, report errors of some missing fields, fix another one (which can be
> fixed, for example, replace Nothing with `Just default_value` or even I
> can fix `Just wrong` to `Just right`, etc, etc). After all of this, I
> know that I have "clean" data, so all my complex types now have `Just
> right_value` fields. But I need to process them as optional, with
> possible Nothing case! To avoid it I must create copies of `A`, `B`,
> etc, where `a1`, `b1` will be `B`, `C`, not `Maybe B`, `Maybe C`. Sure,
> it's not a case.
>
> After processing and filtering, I create, for example, some resulting
> objects:
>
>    data Result {
>      a :: A -- not Maybe!
>      ... }
>
> And even more: `a::A` in `Result` (I know it, after filtering) will not
> contain Nothings, only `Just right_values`s.
>
> But each function which consumes `A` must do something with possible
> Nothing values even after filtering and fixing of `A`s.
>
> I have, for example, function:
>
>    createResults :: [A] -> [Result]
>    createResults alst =
>      ...
>      case of (a1 theA) ->
>        Just right_value -> ...
>        Nothing ->
>          logError
>          undefined -- can not happen
>
> Fun here is: that it happens (I found bug in my filtering
> code with this `undefined`). But now I thought about it: what is the
> idiomatic way to solve such situation? When you need to have:
>
>    - COMPLEX type WITH Maybes
>    - the same type WITHOUT Maybes
>
> Alternative is to keep this Maybes to the very end of processing, what I
> don't like. Or to have types copies, which is more terrible, sure.
>
> PS. I threw IOs away to show only the crux of the problem.
>
> ---
> Cheers,
>    Paul
> _______________________________________________
> Beginners mailing list
> Beginners at haskell.org
> http://mail.haskell.org/cgi-bin/mailman/listinfo/beginners