Question

I have a bunch of functions like:

f1 :: String -> String -> ... -> String -> ()
f1 a b ... z = g [("a", a), ("b", b), ... ("z", z)]
...
fn :: String -> Int -> String -> ... -> String -> ()
fn a b ... z = g [("a", a), ("b", show b), ... ("z", z)]

So user can just call them like f1 "abc" "def". I don’t want him to do this because he can easily swap “abc” and “def” by mistake (and God knows how much time would be wasted while debugging). I want him to pass arguments like fk (A "abc") (B "def")
As far as I can see, there are 2 options:

1. Massive data construction and massive unpack function:

   data Value = A String
              | B String
              | C Int
              | D String
              ...
   
    unpack :: Value -> String
    unpack (A a) = a
    unpack (B b) = b
    unpack (C c) = show c
    unpack (D c) = d
   

   Lots of code.

2. Common typeclass and newtypes:
   EDIT: Okay then, we can use GeneralizedNewtypeDeriving in such simple case.

     {-# LANGUAGE GeneralizedNewtypeDeriving #-}
   
     class Value a where
       unpack :: a -> String
     instance Value String where
       unpack = id
     instance Value Int where
       unpack = show
   
     newtype A = A String deriving Value
     newtype B = B String deriving Value
     newtype C = C Int deriving Value
     newtype D = D String deriving Value
   
     ...
   

   Looks much better but all fk would look like

      fk a b ... z = g [("a", unpack a), ("b", unpack b), ... ("z", unpack z)]
   

   Lots of code and duplication.

What I want is some magic trick which would allow me:

1. fk a b ... z = g [("a", a), ("b", b), ... ("z", z)]
2. g = h . map (second unpack)

Answer 1

I think the problems boils down to this: The list can have elements of the same type only; which means that either you have to ‘coalesce’ it into a single type in your f, or you cannot rely on haskells type checks. E.g. the following code would work for you, but the type check is runtime:

{-# LANGUAGE GADTs #-}

import Control.Arrow (second)

data Item where
    A :: String -> Item
    B :: Int -> Item

unpack (A s) = s
unpack (B i) = show i

myf a@(A {}) b@(B {}) c@(B {}) = 
    let g = [("a", a), ("b", b), ("c", c)]
    in map (second unpack) g
myf _ _ _ = error "Bad types"

main = do
    putStrLn $ show $ myf (A "test") (B 13) (B 14)
    putStrLn $ show $ myf (A "test") (B 13) (A "xxx")

When you want compile-time type check, you can do something like this; however, you still have to retype the parameters to the same type, so in some sense, there is not much difference between unpacking it, only it might by slightly less error-prone, though. A nice trick comes from the json packages – they redefine some operator (e.g. =:) to create the type, so you would have:

{-# LANGUAGE ExistentialQuantification #-}
import Control.Arrow (second)

class Value a where
    unpack :: a -> String
newtype A = A String
newtype B = B Int

instance Value A where
    unpack (A a) = a

instance Value B where
    unpack (B b) = show b

data Item = forall b. Value b => Item b
a =: b = (a, Item b)

myf :: A -> B -> B -> [(String, String)]
myf a b c = 
    let g = ["a" =: a, "b" =: b, "c" =: c]
    in map (second (\(Item x) -> unpack x)) g

main = do
    putStrLn $ show $ myf (A "test") (B 13) (B 14)

It’s not that much different from just defining a =: b = (a, unpack b) though.