Haskell: Heterogeneous list for data with phantom variable - haskell

I'm learning about existential quantification, phantom types, and GADTs at the moment. How do I go about creating a heterogeneous list of a data type with a phantom variable? For example:
{-# LANGUAGE GADTs #-}
{-# LANGUAGE ExistentialQuantification #-}
data Toy a where
TBool :: Bool -> Toy Bool
TInt :: Int -> Toy Int
instance Show (Toy a) where
show (TBool b) = "TBool " ++ show b
show (TInt i) = "TInt " ++ show i
bools :: [Toy Bool]
bools = [TBool False, TBool True]
ints :: [Toy Int]
ints = map TInt [0..9]
Having functions like below are OK:
isBool :: Toy a -> Bool
isBool (TBool _) = True
isBool (TInt _) = False
addOne :: Toy Int -> Toy Int
addOne (TInt a) = TInt $ a + 1
However, I would like to be able to declare a heterogeneous list like so:
zeros :: [Toy a]
zeros = [TBool False, TInt 0]
I tried using an empty type class to restrict the type on a by:
class Unify a
instance Unify Bool
instance Unify Int
zeros :: Unify a => [Toy a]
zeros = [TBool False, TInt 0]
But the above would fail to compile. I was able to use existential quantification to do get the following:
data T = forall a. (Forget a, Show a) => T a
instance Show T where
show (T a) = show a
class (Show a) => Forget a
instance Forget (Toy a)
instance Forget T
zeros :: [T]
zeros = [T (TBool False), T (TInt 0)]
But this way, I cannot apply a function that was based on the specific type of a in Toy a to T e.g. addOne above.
In conclusion, what are some ways I can create a heterogeneous list without forgetting/losing the phantom variable?

Start with the Toy type:
data Toy a where
TBool :: Bool -> Toy Bool
TInt :: Int -> Toy Int
Now you can wrap it up in an existential without over-generalizing with the class system:
data WrappedToy where
Wrap :: Toy a -> WrappedToy
Since the wrapper only holds Toys, we can unwrap them and get Toys back:
incIfInt :: WrappedToy -> WrappedToy
incIfInt (Wrap (TInt n)) = Wrap (TInt (n+1))
incIfInt w = w
And now you can distinguish things within the list:
incIntToys :: [WrappedToy] -> [WrappedToy]
incIntToys = map incIfInt
Edit
As Cirdec points out, the different pieces can be teased apart a bit:
onInt :: (Toy Int -> WrappedToy) -> WrappedToy -> WrappedToy
onInt f (Wrap t#(TInt _)) = f t
onInt _ w = w
mapInt :: (Int -> Int) -> Toy Int -> Toy Int
mapInt f (TInt x) = TInt (f x)
incIntToys :: [WrappedToy] -> [WrappedToy]
incIntToys = map $ onInt (Wrap . mapInt (+1))
I should also note that nothing here so far really justifies the Toy GADT. bheklilr's simpler approach of using a plain algebraic datatype should work just fine.

There was a very similar question a few days ago.
In your case it would be
{-# LANGUAGE GADTs, PolyKinds, Rank2Types #-}
data Exists :: (k -> *) -> * where
This :: p x -> Exists p
type Toys = [Exists Toy]
zeros :: Toys
zeros = [This (TBool False), This (TInt 0)]
It's easy to eliminate an existential:
recEx :: (forall x. p x -> c) -> Exists p -> c
recEx f (This x) = f x
Then if you have a recursor for the Toy datatype
recToy :: (Toy Bool -> c) -> (Toy Int -> c) -> Toy a -> c
recToy f g x#(TBool _) = f x
recToy f g x#(TInt _) = g x
you can map a wrapped toy:
mapToyEx :: (Toy Bool -> p x) -> (Toy Int -> p y) -> Exists Toy -> Exists p
mapToyEx f g = recEx (recToy (This . f) (This . g))
For example
non_zeros :: Toys
non_zeros = map (mapToyEx (const (TBool True)) addOne) zeros
This approach is similar to one in #dfeuer's answer, but it's less ad hoc.

The ordinary heterogeneous list indexed by a list of the types of its elements is
{-# LANGUAGE DataKinds #-}
{-# LANGUAGE TypeOperators #-}
{-# LANGUAGE GADTs #-}
data HList l where
HNil :: HList '[]
HCons :: a -> HList l -> HList (a ': l)
We can modify this to hold values inside some f :: * -> *.
data HList1 f l where
HNil1 :: HList1 f '[]
HCons1 :: f a -> HList1 f l -> HList1 f (a ': l)
Which you can use to write zeros without forgetting the type variables.
zeros :: HList1 Toy [Bool, Int]
zeros = HCons1 (TBool False) $ HCons1 (TInt 0) $ HNil1

Have you played with Data.Typeable? A Typeable constraint allows you to make guesses at the type hidden by the existential, and cast to that type when you guess right.
Not your example, but some example code I have lying around:
{-# LANGUAGE GADTs, ScopedTypeVariables, TypeOperators #-}
import Data.Typeable
data Showable where
-- Note that this is an existential defined in GADT form
Showable :: (Typeable a, Show a) => a -> Showable
instance Show Showable where
show (Showable value) = "Showable " ++ show value
-- Example of casting Showable to Integer
castToInteger :: Showable -> Maybe Integer
castToInteger (Showable (value :: a)) =
case eqT :: Maybe (a :~: Integer) of
Just Refl -> Just value
Nothing -> Nothing
example1 = [Showable "foo", Showable 5]
example2 = map castToInteger example1

Related

Is there a way to apply Maybe constructor to each field of record with generics?

I have two data types and the second one is the copy of first, but with Maybe on each field.
data A = {a :: Int, b :: String}
data B = {c :: Maybe Int, d :: Maybe String}
Is there a way to make a functions
f :: A -> B
g :: B -> A -> A
without any knowledge about fields itself? (if value of first argument is nothing g will take default value from second argument)
This can be done with generics-sop, a library that extends the default Generics machinery of GHC.
"generics-sop" can take a regular record and deduce a generic representation for it. This representation has a type parameter that wraps every field, and the library allows Applicative sequence-like operations across the record fields.
{-# language TypeOperators #-}
{-# language DeriveGeneric #-}
{-# language TypeFamilies #-}
{-# language DataKinds #-}
import qualified GHC.Generics as GHC
import Generics.SOP
data A = A {a :: Int, b :: String} deriving (Show,GHC.Generic)
instance Generic A -- this Generic is from generics-sop
defaulty :: (Generic a, Code a ~ '[ xs ]) => NP Maybe xs -> a -> a
defaulty maybes r = case (from r) of
SOP (Z np) -> let result = hliftA2 (\m i -> maybe i I m) maybes np
in to (SOP (Z result))
main :: IO ()
main = do
print $ defaulty (Nothing :* Just "bar" :* Nil) (A 99 "foo")
Nothing :* Just "bar" :* Nil is a generic representation that matches the list of fields in the original record definition. Notice that each field in the representation is wrapped in Maybe.
See here for another example of generics-sop.
How about:
{-# LANGUAGE RankNTypes #-}
data R f = R { a :: f Int, b :: f String, c :: f Char }
newtype I a = I { unI :: a }
fromMaybeI :: I a -> Maybe a -> I a
fromMaybeI a Nothing = a
fromMaybeI _ (Just a) = I a
fromMaybeR :: R I -> R Maybe -> R I
fromMaybeR ri rm =
R (go a) (go b) (go c)
where
go :: (forall f. R f -> f a) -> I a
go x = fromMaybeI (x ri) (x rm)
R Maybe is the record with Maybe values, R I is the record with concrete values.
Using RankNTypes reduces the amount of boilerplate code in fromMaybeR.
One downside is that you have use I and unI to construct and
access the field values.

Haskell: finite types of user-defined values?

Part of the power of Haskell is delegating range checks to the type system (see for example Numeric.Natural). Is this possible to do for types whose set of values is defined once at runtime? I'd effectively like an Enum whose values are unknown at compile-time.
Edit: In terms of example usage:
-- Defines the list of allowed values
init :: [a] -> ?
-- Constructs a new instance
construct :: a -> ? -> Maybe Foo
-- Then just usable like an enum
bar :: Int -> Foo -> Bar
Ideally I'd be able to use things like Bounded on it too.
Your sample code is, unfortunately, too sparse to indicate what you really mean. I'm guessing you may be after dependent types, as n.m. suggested. If that is the case, you're likely better off looking at something like Agda instead of Haskell. If you want a somewhat safer version of what Daniel Wagner suggested, you can get it with the reflection package.
{-# LANGUAGE ScopedTypeVariables #-}
{-# LANGUAGE MultiParamTypeClasses #-}
{-# LANGUAGE FlexibleContexts #-}
{-# LANGUAGE UndecidableInstances #-}
{-# LANGUAGE RankNTypes #-}
{-# LANGUAGE GeneralizedNewtypeDeriving #-}
{-# LANGUAGE StandaloneDeriving #-}
module DynEnum (.... not including newtype constructors) where
import Data.Reflection
import Data.Proxy
import Data.Set (Set, splitMember, size, lookupIndex, fromList, elemAt, member, findMin, findMax)
import Data.Foldable
import Data.Bool
import Data.Type.Coercion
-- Just Enum
newtype Limited a s = Limited { unLimited :: a }
type role Limited representational nominal
-- We can safely conflate types of values that come
-- from the same set.
coerceLimited :: (Reifies s (Set a), Reifies t (Set a), Ord a)
=> Maybe (Coercion (Limited a s) (Limited a t))
coerceLimited
| reflect (Proxy :: Proxy s) == reflect (Proxy :: Proxy t)
= Just Coercion
| otherwise = Nothing
instance (Ord a, Reifies s (Set a)) => Enum (Limited a s) where
toEnum i
| 0 <= i && i < size values = Limited $ elemAt i values
| otherwise = error "Limited toEnum: out of range"
where values = reflect (Proxy :: Proxy s)
fromEnum x = case lookupIndex (unLimited x) (reflect x) of
Nothing -> error "Limited fromEnum: out of range"
Just i -> i
enumFrom (Limited a) = case splitMember a (reflect (Proxy :: Proxy s)) of
(_, False, s) -> fmap Limited $ toList s
(_, True, s) -> Limited a : fmap Limited (toList s)
enumFromTo (Limited a) (Limited b) = case splitMember a (reflect (Proxy :: Proxy s)) of
(_, inclFirst, s) -> case splitMember b s of
(t, inclLast, _) -> bool id (Limited a:) inclFirst
. (map Limited (toList t) ++)
$ bool [] [Limited b] inclLast
initialize :: Ord a
=> [a]
-> (forall s . Enum (Limited a s) => Proxy s -> r)
-> r
initialize vals f = reify (fromList vals) f
construct :: forall s a . (Ord a, Reifies s (Set a)) => a -> Maybe (Limited a s)
construct x
| x `member` reflect (Proxy :: Proxy s) = Just (Limited x)
| otherwise = Nothing
newtype Bound a b = Bound a deriving (Enum)
type role Bound representational nominal
instance Reifies b (a, a) => Bounded (Bound a b) where
minBound = Bound . fst $ reflect (Proxy :: Proxy b)
maxBound = Bound . snd $ reflect (Proxy :: Proxy b)
initializeBounded :: (a, a)
-> (forall b . Bounded (Bound a b) => Proxy b -> r)
-> r
initializeBounded bounds f = reify bounds f
newtype LimitedB a s b = LimitedB (Bound (Limited a s) b)
deriving instance (Ord a, Reifies s (Set a)) => Enum (LimitedB a s b)
deriving instance Reifies b (Limited a s, Limited a s) => Bounded (LimitedB a s b)
initializeLimitedB :: Ord a
=> [a]
-> (forall s b . (Enum (LimitedB a s b), Bounded (LimitedB a s b)) => Proxy s -> Proxy b -> r)
-> r
initializeLimitedB [] _f = error "Cannot initialize LimitedB with an empty list"
initializeLimitedB vals f = reify set $ \ps ->
reify (Limited (findMin set), Limited (findMax set)) $ \pb ->
f ps pb
where
set = fromList vals
Perhaps a Set is suitable for your needs. We have:
initialize :: Ord a => [a] -> Set a
initialize = fromList
construct :: Ord a => a -> Set a -> Maybe a
construct x xs = guard (x `member` xs) >> return x
dynamicMinBound :: Set a -> Maybe a
dynamicMinBound xs = fst <$> minView xs
dynamicMaxBound :: Set a -> Maybe a
dynamicMaxBound xs = fst <$> maxView xs
enumerate :: Set a -> [a]
enumerate = toList
dynamicToEnum :: Int -> Set a -> Maybe a
dynamicToEnum n xs = guard (inRange n (0, size xs-1)) >> return (elemAt n xs)
dynamicFromEnum :: Ord a => a -> Set a -> Maybe Int
dynamicFromEnum = lookupIndex
I believe this covers the operations you asked for, though I could easily have misunderstood something -- your specification is not 100% clear to me.

Can I have a polymorphic function argument that may not need to be supplied for some types?

I have a datatype F with a special-case for Int:
{-# LANGUAGE GADTs, RankNTypes #-}
data F a where
FGen :: a -> F a
FInt :: F Int
Without exposing the details of this datatype to callers - the real datatype is more complicated containing internal implementation details - I want to provide an API for using it:
transform :: (a -> b) -> b -> F a -> b
transform f i (FGen v) = f v
transform f i FInt = i
If I'm going to call transform on a F Int, clearly both of the first two arguments are important:
transformInt :: F Int -> Int
transformInt = transform (+1) 5
But if I'm going to call it on a F Char, the second argument is unnecessary as the value can't be a FInt:
transformChar :: F Char -> Char
transformChar = transform id (error "unreachable code")
Is there a way I can express this in the type of transform?
I tried
transform :: (a -> b) -> (a ~ Int => b) -> F a -> b
transform f i (FGen v) = f v
transform f i FInt = i
but then transformChar doesn't compile with
Couldn't match type ‘Char’ with ‘Int’
Inaccessible code in
a type expected by the context: (Char ~ Int) => Char
In the second argument of ‘transform’, namely
‘(error "unreachable code")’
In the expression: transform id (error "unreachable code")
In an equation for ‘transformChar’:
transformChar = transform id (error "unreachable code")
and anyway I'd still want absurd value I could use instead of the error to properly express that the compiler should be able to prove the code will never be used.
We can use the propositional equality type in Data.Type.Equality and we can also express inaccessibility of code from GHC 7.8, using empty case expressions:
{-# LANGUAGE GADTs, RankNTypes, EmptyCase, TypeOperators #-}
import Data.Type.Equality
data F a where
FGen :: a -> F a
FInt :: F Int
transform :: (a -> b) -> ((a :~: Int) -> b) -> F a -> b
transform f i (FGen v) = f v
transform f i FInt = i Refl
transformChar :: F Char -> Char
transformChar = transform id (\p -> case p of {})
-- or (\case {}) with LambdaCase
transformInt :: F Int -> Int
transformInt = transform (+1) (const 5)
I like the answer with a GADT for the type equality proof better. This answer explains how to do the same thing with TypeFamilies. With closed type families we can write functions from types to the unit () and zero Void of the type system to represent prepositional truth and false.
{-# LANGUAGE TypeFamilies #-}
import Data.Void
type family IsInt a where
IsInt Int = ()
IsInt a = Void
The second argument to transform is () -> b when IsInt a and Void -> b (the type of absurd) when a isn't an integer.
transform :: (a -> b) -> (IsInt a -> b) -> F a -> b
transform f i (FGen v) = f v
transform f i FInt = i ()
transformChar can be written in terms of absurd and transformInt must pass in b as a constant function.
transformChar :: F Char -> Char
transformChar = transform id absurd
transformInt :: F Int -> Int
transformInt = transform (+1) (const 5)
More Reusable
At András Kovács suggestion, we can make this more reusable with a type family for type equality (==) that returns lifted Bools.
{-# LANGUAGE TypeFamilies #-}
{-# LANGUAGE TypeOperators #-}
{-# LANGUAGE DataKinds #-}
type family (==) a b :: Bool where
(==) a a = True
(==) a b = False
We could provide another type family to convert True to () and False to Void. For this specific problem it reads better to go all the way from True or False and some type b to () -> b or Void -> b.
type family When p b where
When True b = () -> b
When False b = Void -> b
The type of transform then reads.
transform :: (a -> b) -> When (a == Int) b -> F a -> b

Transform a function with a typeclass constraint into a function taking an explicit typeclass dictionary

It's well known that one way of implementing Haskell typeclasses is via 'typeclass dictionaries'. (This is of course the implementation in ghc, though I make the obligatory remark that Other Implementations are Possible.) To fix ideas, I'll briefly describe how this works. A class declaration like
class (MyClass t) where
test1 :: t -> t -> t
test2 :: t -> String
test3 :: t
can be mechanically transformed into the definition of a datatype like:
data MyClass_ t = MyClass_ {
test1_ :: t -> t -> t,
test2_ :: t -> String,
test3_ :: t,
}
Then we can mechanically transform each instance declaration into an object of that type; for instance:
instance (MyClass Int) where
test1 = (+)
test2 = show
test3 = 3
turns into
instance_MyClass_Int :: MyClass_ Int
instance_MyClass_Int = MyClass_ (+) show 3
and similarly a function which has a typeclass constraint can be turned into a function that takes an extra argument; for instance:
my_function :: (MyClass t) => t -> String
my_function val = test2 . test1 test3
turns into
my_function_ :: MyClass_ t -> t -> String
my_function_ dict val = (test2_ dict) . (test1_ dict) (test3_ dict)
The point is that as long as the compiler knows how to fill in these hidden arguments (which is not totally trivial) then you can translate code that uses classes and instances into code that uses only more basic features of the language.
With that background, here's my question. I have a module M which defines a bunch of classes and functions with class constraints. M is 'opaque'; I can see what it exports (the equivalent of the .hi file) and I can import from it but I can't see its source code. I want to construct a new module N which basically exports the same things but with the transformation above applied. So for instance if M exported
class (Foo t) where
example1 :: t -> t -> t
example2 :: t -- note names and type signatures visible here
-- because they form part of the interface...
instance (Foo String) -- details of implementation invisible
instance (Foo Bool) -- details of implementation invisible
my_fn :: (Foo t) => t -> t -- exported polymorphic fn with class constraint
-- details of implementation invisible
N would start like
module N where
import M
data Foo_ t = Foo_ {example1_ :: t-> t -> t, example2_ :: t}
instance_Foo_String :: Foo_ String
instance_Foo_String = Foo_ example1 example2
instance_Foo_Bool :: Foo_ Bool
instance_Foo_Bool = Foo_ example1 example2
my_fn_ :: Foo_ t -> t -> t
my_fn_ = ???
And my question is what on earth I can put in place of the ???. In other words, what can I write to extract the 'explicit typeclass' version of the function my_fn from the original? It seems rather tricky, and it's infuriating because we all know that 'under the hood' the module M is basically already exporting something like the my_fn_ which I want to create. (Or at least, it is on GHC.)
For the record, I thought I would explain the 'hacky' solution to this which I already know of. I'll basically illustrate it using a series of examples. So let's imagine we're trying to reify the classes, instances and functions in the following (which consists mostly of pretty standard typeclasses, generally simplified somewhat for the exposition):
{-# LANGUAGE FlexibleContexts #-}
{-# LANGUAGE FlexibleInstances #-}
module Src where
import Data.List (intercalate)
class SimpleShow a where
sshow :: a -> String
class SimpleMonoid a where
mempty :: a
mappend :: a -> a -> a
class SimpleFunctor f where
sfmap :: (a -> b) -> f a -> f b
instance SimpleShow Int where
sshow = show
instance SimpleMonoid [a] where
mempty = []
mappend = (++)
instance SimpleMonoid ([a], [b]) where
mempty = ([], [])
mappend (a1, b1) (a2, b2) = (a1 ++ a2, b1 ++ b2)
instance SimpleFunctor [] where
sfmap = map
There's meant to be some generality in these examples: we have
'a' in positive position in the class member
'a' in negative position in the class member
an instance requiring flexible instances
a higher-kinded type
We leave multi-parameter type families as an exercise! Note that I do believe that what I'm presenting is a completely general, syntactic procedure; I just think it's easier to illustrate with examples than by describing the transformation formally. Anyway, let's suppose we've got the following functions to process:
show_2lists :: (SimpleShow a) => [a] -> [a] -> String
show_2lists as1 as2 = "[" ++ intercalate ", " (map sshow as1) ++ "]/["
++ intercalate ", " (map sshow as2) ++ "]"
mconcat :: (SimpleMonoid a) => [a] -> a
mconcat = foldr mappend mempty
example :: (SimpleMonoid (x, y)) => [(x, y)] -> (x, y)
example = foldr mappend mempty
lift_all :: (SimpleFunctor f) => [a -> b] -> [f a -> f b]
lift_all = map sfmap
Then the actual reification looks like:
{-# LANGUAGE PatternGuards #-}
{-# LANGUAGE FlexibleContexts #-}
{-# LANGUAGE FunctionalDependencies #-}
{-# LANGUAGE KindSignatures #-}
{-# LANGUAGE MultiParamTypeClasses #-}
{-# LANGUAGE RankNTypes #-}
{-# LANGUAGE ScopedTypeVariables #-}
{-# LANGUAGE EmptyDataDecls #-}
{-# LANGUAGE UndecidableInstances #-}
{-# LANGUAGE FlexibleInstances #-}
module Main where
import Unsafe.Coerce
import Src
data Proxy k = Proxy
class Reifies s a | s -> a where
reflect :: proxy s -> a
newtype Magic a r = Magic (forall (s :: *). Reifies s a => Proxy s -> r)
reify :: forall a r. a -> (forall (s :: *). Reifies s a => Proxy s -> r) -> r
reify a k = unsafeCoerce (Magic k :: Magic a r) (const a) Proxy
{-# INLINE reify #-}
data SimpleShow_ a = SimpleShow_ {sshow_ :: a -> String}
data SimpleMonoid_ a = SimpleMonoid_ {mempty_ :: a,
mappend_ :: a -> a -> a}
data SimpleFunctor_ f = SimpleFunctor_ {
sfmap_ :: forall a b. (a -> b) -> (f a -> f b)
}
instance_SimpleShow_Int :: SimpleShow_ Int
instance_SimpleShow_Int = SimpleShow_ sshow
instance_SimpleMonoid_lista :: SimpleMonoid_ [a]
instance_SimpleMonoid_lista = SimpleMonoid_ mempty mappend
instance_SimpleMonoid_listpair :: SimpleMonoid_ ([a], [b])
instance_SimpleMonoid_listpair = SimpleMonoid_ mempty mappend
instance_SimpleFunctor_list :: SimpleFunctor_ []
instance_SimpleFunctor_list = SimpleFunctor_ sfmap
---------------------------------------------------------------------
--code to reify show_2lists :: (SimpleShow a) => [a] -> [a] -> String
-- for each type variable that occurs in the constraints, we must
-- create a newtype. Here there is only one tpye variable ('a') so we
-- create one newtype.
newtype Wrap_a a s = Wrap_a { extract_a :: a }
-- for each constraint, we must create an instance of the
-- corresponding typeclass where the instance variables have been
-- replaced by the newtypes we just made, as follows.
instance Reifies s (SimpleShow_ a) => SimpleShow (Wrap_a a s) where
--sshow :: (Wrap_ a s) -> String
sshow = unsafeCoerce sshow__
where sshow__ :: a -> String
sshow__ = sshow_ $ reflect (undefined :: [] s)
-- now we can reify the main function
show_2lists_ :: forall a. SimpleShow_ a -> [a] -> [a] -> String
show_2lists_ dict = let
magic :: forall s. ([Wrap_a a s] -> [Wrap_a a s] -> String)
-> Proxy s -> ([a] -> [a] -> String)
magic v _ arg1 arg2 = let
w_arg1 :: [Wrap_a a s]
w_arg1 = unsafeCoerce (arg1 :: [a])
w_arg2 :: [Wrap_a a s]
w_arg2 = unsafeCoerce (arg2 :: [a])
w_ans :: String
w_ans = v w_arg1 w_arg2
ans :: String
ans = unsafeCoerce w_ans
in ans
in (reify dict $ magic show_2lists)
---------------------------------------------------------------------
--code to reify mconcat :: (SimpleMonoid a) => [a] -> a
-- Here the newtypes begin with Wrap1 to avoid name collisions with
-- the ones above
newtype Wrap1_a a s = Wrap1_a { extract1_a :: a }
instance Reifies s (SimpleMonoid_ a) => SimpleMonoid (Wrap1_a a s) where
--mappend :: (Wrap1_a a s) -> (Wrap1_a a s) -> (Wrap1_a a s)
mappend = unsafeCoerce mappend__
where mappend__ :: a -> a -> a
mappend__ = (mappend_ $ reflect (undefined :: [] s))
--mempty :: (Wrap1_a a s)
mempty = unsafeCoerce mempty__
where mempty__ :: a
mempty__ = (mempty_ $ reflect (undefined :: [] s))
mconcat_ :: forall a. SimpleMonoid_ a -> [a] -> a
mconcat_ dict = let
magic :: forall s. ([Wrap1_a a s] -> (Wrap1_a a s)) -> Proxy s -> ([a] -> a)
magic v _ arg1 = let
w_arg1 :: [Wrap1_a a s]
w_arg1 = unsafeCoerce (arg1 :: [a])
w_ans :: Wrap1_a a s
w_ans = v w_arg1
ans :: a
ans = unsafeCoerce w_ans
in ans
in (reify dict $ magic mconcat)
---------------------------------------------------------------------
--code to reify example :: (SimpleMonoid (x, y)) => [(x, y)] -> (x, y)
newtype Wrap2_x x s = Wrap2_x { extract2_x :: x }
newtype Wrap2_y y s = Wrap2_y { extract2_y :: y }
instance Reifies s (SimpleMonoid_ (x, y))
=> SimpleMonoid (Wrap2_x x s, Wrap2_y y s) where
--mappend :: (Wrap2_x x s, Wrap2_y y s) -> (Wrap2_x x s, Wrap2_y y s)
-- -> (Wrap2_x x s, Wrap2_y y s)
mappend = unsafeCoerce mappend__
where mappend__ :: (x, y) -> (x, y) -> (x, y)
mappend__ = (mappend_ $ reflect (undefined :: [] s))
--mempty :: (Wrap2_x x s, Wrap2_y y s)
mempty = unsafeCoerce mempty__
where mempty__ :: (x, y)
mempty__ = (mempty_ $ reflect (undefined :: [] s))
example_ :: forall x y. SimpleMonoid_ (x, y) -> [(x, y)] -> (x, y)
example_ dict = let
magic :: forall s. ([(Wrap2_x x s, Wrap2_y y s)] -> (Wrap2_x x s, Wrap2_y y s))
-> Proxy s -> ([(x, y)] -> (x, y))
magic v _ arg1 = let
w_arg1 :: [(Wrap2_x x s, Wrap2_y y s)]
w_arg1 = unsafeCoerce (arg1 :: [(x, y)])
w_ans :: (Wrap2_x x s, Wrap2_y y s)
w_ans = v w_arg1
ans :: a
ans = unsafeCoerce w_ans
in ans
in (reify dict $ magic mconcat)
---------------------------------------------------------------------
--code to reify lift_all :: (SimpleFunctor f) => [a -> b] -> [f a -> f b]
newtype Wrap_f f s d = Wrap_f { extract_fd :: f d}
instance Reifies s (SimpleFunctor_ f) => SimpleFunctor (Wrap_f f s) where
--sfmap :: (a -> b) -> (Wrap_f f s a -> Wrap_f f s b)
sfmap = unsafeCoerce sfmap__
where sfmap__ :: (a -> b) -> (f a -> f b)
sfmap__ = sfmap_ $ reflect (undefined :: [] s)
lift_all_ :: forall a b f. SimpleFunctor_ f -> [a -> b] -> [f a -> f b]
lift_all_ dict = let
magic :: forall s. ([a -> b] -> [Wrap_f f s a -> Wrap_f f s b])
-> Proxy s -> ([a -> b] -> [f a -> f b])
magic v _ arg1 = let
w_arg1 :: [a -> b]
w_arg1 = unsafeCoerce (arg1 :: [a -> b])
w_ans :: [Wrap_f f s a -> Wrap_f f s b]
w_ans = v w_arg1
ans :: [f a -> f b]
ans = unsafeCoerce w_ans
in ans
in (reify dict $ magic lift_all)
main :: IO ()
main = do
print (show_2lists_ instance_SimpleShow_Int [3, 4] [6, 9])
print (mconcat_ instance_SimpleMonoid_lista [[1, 2], [3], [4, 5]])
print (example_ instance_SimpleMonoid_listpair
[([1, 2], ["a", "b"]), ([4], ["q"])])
let fns' :: [[Int] -> [Int]]
fns' = lift_all_ instance_SimpleFunctor_list [\ x -> x+1, \x -> x - 1]
print (map ($ [5, 7]) fns')
{- output:
"[3, 4]/[6, 9]"
[1,2,3,4,5]
([1,2,4],["a","b","q"])
[[6,8],[4,6]]
-}
Note that we use a lot of unsafeCoerce, but always relating two types that differ only in the presence of a newtype. Since the run time representations are identical, this is ok.
What you seem to be asking for is known as "local instances". This would mean that you could write something like:
my_fn_ :: forall t. Foo_ t -> t -> t
my_fn_ fooDict = let instance fooDict :: Foo t
in my_fn
Local instances are a natural extension of type classes. They were even standard in the formalism of Wadler and Blott's paper "How to make ad hoc polymorphism less ad hoc". However, they are problematic because they break a property known as principal types. Additionally, they may also break assumptions that there is only ever a single instance of a certain constraint for a specific type (like e.g. Data.Map's assumption about Ord instances). The first problem could be solved by requiring additional type annotations in a local instance and the latter is related to the controversial "orphan instances", which cause a similar problem.
Another relevant paper is Kiselyov and Shan's "Functional pearl: implicit configurations", which contains a variety of type system tricks to simulate local type instances although it doesn't really apply to your situation (pre-existing type class), IIRC.
This isn't a solution in general, but only for some special cases.
There is a hacky way to do this for class methods of a class C t that have the type parameter t appearing in a negative position in their type. e.g., example1 :: Foo t => t -> t -> t is ok, but not example2 :: Foo t => t.
The trick is to create a wrapper data type Wrapper t which comprises the explicit dictionary methods on t paired with a t value, and which has a Foo instance that exploits the appropriate wrapped dictionary methods, e.g.
data Wrapper x = Wrap {example1__ :: (x -> x -> x), val :: x}
instance Foo (Wrapper x) where
example1 (Wrap example1__ x) (Wrap _ y) = Wrap example1__ (example1__ x y)
my_fn_ :: Foo_ t -> t -> t
my_fn_ (Foo_ example1_ example2_) x = val $ my_fn (Wrap example1_ x)
Something tells me this is probably not the solution you are looking for though- it is not general purpose. In the example here, we cannot do anything with example2 because it has no negative occurrence of t with which to "sneak" functions inside. For your example, this means that my_fn in module M can use only example1.

Polymorphic constraint

I have some contrived type:
{-# LANGUAGE DeriveFunctor #-}
data T a = T a deriving (Functor)
... and that type is the instance of some contrived class:
class C t where
toInt :: t -> Int
instance C (T a) where
toInt _ = 0
How can I express in a function constraint that T a is an instance of some class for all a?
For example, consider the following function:
f t = toInt $ fmap Left t
Intuitively, I would expect the above function to work since toInt works on T a for all a, but I cannot express that in the type. This does not work:
f :: (Functor t, C (t a)) => t a -> Int
... because when we apply fmap the type has become Either a b. I can't fix this using:
f :: (Functor t, C (t (Either a b))) => t a -> Int
... because b does not represent a universally quantified variable. Nor can I say:
f :: (Functor t, C (t x)) => t a -> Int
... or use forall x to suggest that the constraint is valid for all x.
So my question is if there is a way to say that a constraint is polymorphic over some of its type variables.
Using the constraints package:
{-# LANGUAGE FlexibleContexts, ConstraintKinds, DeriveFunctor, TypeOperators #-}
import Data.Constraint
import Data.Constraint.Forall
data T a = T a deriving (Functor)
class C t where
toInt :: t -> Int
instance C (T a) where
toInt _ = 0
f :: ForallF C T => T a -> Int
f t = (toInt $ fmap Left t) \\ (instF :: ForallF C T :- C (T (Either a b)))

Resources