I'm looking for a way to have Enum a => UArray a (which makes sense to me as we can trivially map enums to Int and back by toEnum and fromEnum)
So far I tried to steal code of UArray Int from Data.Array.Base and smuggle a few toEnums and fromEnums here and there:
{-# LANGUAGE MagicHash, UnboxedTuples #-}
{-# LANGUAGE MultiParamTypeClasses #-}
{-# LANGUAGE FlexibleInstances #-}
module UArrays where
import Data.Array.Base
import Data.Array.ST
import Data.Array.Unboxed
import GHC.Base -- (Int(I#), Int#(..))
import GHC.Prim -- (indexIntArray#, readIntArray#, writeIntArray#)
import GHC.ST (ST(..), runST)
import Unsafe.Coerce
instance (Enum a, Bounded a) => IArray UArray a where
{-# INLINE bounds #-}
bounds (UArray l u _ _) = (l, u)
{-# INLINE numElements #-}
numElements (UArray _ _ n _) = n
{-# INLINE unsafeArray #-}
unsafeArray lu ies = runST (unsafeArrayUArray lu ies minBound)
{-# INLINE unsafeAt #-}
unsafeAt (UArray _ _ _ arr#) (I# i#) =
I# $ fromEnum (indexIntArray# arr# i#)
{-# INLINE unsafeReplace #-}
unsafeReplace arr ies = runST (unsafeReplaceUArray arr ies)
{-# INLINE unsafeAccum #-}
unsafeAccum f arr ies = runST (unsafeAccumUArray f arr ies)
{-# INLINE unsafeAccumArray #-}
unsafeAccumArray f initialValue lu ies =
runST (unsafeAccumArrayUArray f initialValue lu ies)
-- data STUArray s i e = STUArray !i !i !Int (GHC.Prim.MutableByteArray# s)
instance (Enum a, Bounded a) => MArray (STUArray s) a (ST s) where
{-# INLINE getBounds #-}
getBounds (STUArray l u _ _) = return (l, u)
{-# INLINE getNumElements #-}
getNumElements (STUArray _ _ n _) = return n
{-# INLINE unsafeNewArray_ #-}
unsafeNewArray_ (l, u) = unsafeNewArraySTUArray_ (l, u) wORD_SCALE
{-# INLINE newArray_ #-}
newArray_ arrBounds = newArray arrBounds minBound
{-# INLINE unsafeRead #-}
-- unsafeRead :: GHC.Arr.Ix i => a i e -> Int -> m e
unsafeRead (STUArray _ _ _ marr#) (I# i#) =
ST $ \ s1# ->
case readIntArray# marr# i# s1# of
(# s2#, e# #) -> (# s2#, I# (toEnum e#) #)
{-# INLINE unsafeWrite #-}
-- unsafeWrite :: GHC.Arr.Ix i => a i e -> Int -> e -> m ()
unsafeWrite (STUArray _ _ _ marr#) (I# i#) (I# e#) =
ST $ \ s1# ->
case writeIntArray# marr# (unsafeCoerce i#) (I# $ fromEnum e#) s1# of
s2# -> (# s2#, () #)
But of course it doesn't compile:
[2 of 4] Compiling UArrays ( UArrays.hs, interpreted )
UArrays.hs:27:14:
Couldn't match expected type `Int#' with actual type `Int'
In the return type of a call of `fromEnum'
In the second argument of `($)', namely
`fromEnum (indexIntArray# arr# i#)'
In the expression: I# $ fromEnum (indexIntArray# arr# i#)
UArrays.hs:52:45:
Couldn't match expected type `Int' with actual type `Int#'
In the first argument of `toEnum', namely `e#'
In the first argument of `I#', namely `(toEnum e#)'
In the expression: I# (toEnum e#)
UArrays.hs:57:57:
Couldn't match expected type `Int#' with actual type `Int'
In the return type of a call of `fromEnum'
In the second argument of `($)', namely `fromEnum e#'
In the third argument of `writeIntArray#', namely
`(I# $ fromEnum e#)'
Failed, modules loaded: Utils.
Also there is no magical unboxInt :: Int -> Int# in GHC.*, and pattern-matching on I# doesn't yield Int but an Int# instead, yet somehow UArray Int exists and works on Int#s.
I have also found a post about making an UArray for newtypes, but it doesn't seem to apply because it relies on unsafeCoerce. I tried that but it made a funny listArray (0, 54) $ cycle [Red, Yellow, Green] in which all constructors were Blue.
Am I on the wrong track?
Update:
It works now, here is the source code:
- UArrays.hs: http://hpaste.org/56728
- BenchmarkUArray.hs: http://hpaste.org/56729
- Makefile: http://hpaste.org/56727
You need to realize that
Intis a boxed integer that is constructed from an unboxed integerInt#via the constructorI#. So:Since
fromEnumalready returns a boxed integer, you don't have to re-box it by applyingI#again, so in this code snippet, for instance:... you can simply leave out the
I#constructor. Similarly, when usingtoEnum, you should apply theI#constructor to get a boxed integer out of an unboxed integer.As @leftaroundabout mentioned, this boxing and unboxing in addition to the complexity that
fromEnumandtoEnumcan have (Especially for tuples, etc) might lead to less performance compared to using simple boxedArrays.