diff --git a/Control/Monad/Primitive.hs b/Control/Monad/Primitive.hs
--- a/Control/Monad/Primitive.hs
+++ b/Control/Monad/Primitive.hs
@@ -1,6 +1,7 @@
 {-# LANGUAGE CPP, MagicHash, UnboxedTuples, TypeFamilies #-}
 {-# LANGUAGE FlexibleContexts, FlexibleInstances, UndecidableInstances #-}
 {-# LANGUAGE ScopedTypeVariables #-}
+{-# OPTIONS_GHC -fno-warn-deprecations #-}
 
 -- |
 -- Module      : Control.Monad.Primitive
@@ -77,6 +78,8 @@
 -- Unlike 'PrimMonad', this typeclass requires that the @Monad@ be fully
 -- expressed as a state transformer, therefore disallowing other monad
 -- transformers on top of the base @IO@ or @ST@.
+--
+-- @since 0.6.0.0
 class PrimMonad m => PrimBase m where
   -- | Expose the internal structure of the monad
   internal :: m a -> State# (PrimState m) -> (# State# (PrimState m), a #)
@@ -97,41 +100,52 @@
   internal (IO p) = p
   {-# INLINE internal #-}
 
+-- | @since 0.6.3.0
 instance PrimMonad m => PrimMonad (ContT r m) where
   type PrimState (ContT r m) = PrimState m
   primitive = lift . primitive
   {-# INLINE primitive #-}
+
 instance PrimMonad m => PrimMonad (IdentityT m) where
   type PrimState (IdentityT m) = PrimState m
   primitive = lift . primitive
   {-# INLINE primitive #-}
+
+-- | @since 0.6.2.0
 instance PrimBase m => PrimBase (IdentityT m) where
   internal (IdentityT m) = internal m
   {-# INLINE internal #-}
+
 instance PrimMonad m => PrimMonad (ListT m) where
   type PrimState (ListT m) = PrimState m
   primitive = lift . primitive
   {-# INLINE primitive #-}
+
 instance PrimMonad m => PrimMonad (MaybeT m) where
   type PrimState (MaybeT m) = PrimState m
   primitive = lift . primitive
   {-# INLINE primitive #-}
+
 instance (Error e, PrimMonad m) => PrimMonad (ErrorT e m) where
   type PrimState (ErrorT e m) = PrimState m
   primitive = lift . primitive
   {-# INLINE primitive #-}
+
 instance PrimMonad m => PrimMonad (ReaderT r m) where
   type PrimState (ReaderT r m) = PrimState m
   primitive = lift . primitive
   {-# INLINE primitive #-}
+
 instance PrimMonad m => PrimMonad (StateT s m) where
   type PrimState (StateT s m) = PrimState m
   primitive = lift . primitive
   {-# INLINE primitive #-}
+
 instance (Monoid w, PrimMonad m) => PrimMonad (WriterT w m) where
   type PrimState (WriterT w m) = PrimState m
   primitive = lift . primitive
   {-# INLINE primitive #-}
+
 instance (Monoid w, PrimMonad m) => PrimMonad (RWST r w s m) where
   type PrimState (RWST r w s m) = PrimState m
   primitive = lift . primitive
@@ -145,6 +159,7 @@
 #endif
 
 #if MIN_VERSION_transformers(0,5,3)
+-- | @since 0.6.3.0
 instance ( Monoid w
          , PrimMonad m
 # if !(MIN_VERSION_base(4,8,0))
@@ -164,10 +179,12 @@
   type PrimState (Strict.StateT s m) = PrimState m
   primitive = lift . primitive
   {-# INLINE primitive #-}
+
 instance (Monoid w, PrimMonad m) => PrimMonad (Strict.WriterT w m) where
   type PrimState (Strict.WriterT w m) = PrimState m
   primitive = lift . primitive
   {-# INLINE primitive #-}
+
 instance (Monoid w, PrimMonad m) => PrimMonad (Strict.RWST r w s m) where
   type PrimState (Strict.RWST r w s m) = PrimState m
   primitive = lift . primitive
@@ -205,11 +222,15 @@
 primToST = primToPrim
 
 -- | Convert an 'IO' action to a 'PrimMonad'.
+-- 
+-- @since 0.6.2.0
 ioToPrim :: (PrimMonad m, PrimState m ~ RealWorld) => IO a -> m a
 {-# INLINE ioToPrim #-}
 ioToPrim = primToPrim
 
 -- | Convert an 'ST' action to a 'PrimMonad'.
+--
+-- @since 0.6.2.0
 stToPrim :: PrimMonad m => ST (PrimState m) a -> m a
 {-# INLINE stToPrim #-}
 stToPrim = primToPrim
@@ -233,12 +254,16 @@
 
 -- | Convert an 'ST' action with an arbitraty state token to any 'PrimMonad'.
 -- This operation is highly unsafe!
+-- 
+-- @since 0.6.2.0
 unsafeSTToPrim :: PrimMonad m => ST s a -> m a
 {-# INLINE unsafeSTToPrim #-}
 unsafeSTToPrim = unsafePrimToPrim
 
 -- | Convert an 'IO' action to any 'PrimMonad'. This operation is highly
 -- unsafe!
+--
+-- @since 0.6.2.0
 unsafeIOToPrim :: PrimMonad m => IO a -> m a
 {-# INLINE unsafeIOToPrim #-}
 unsafeIOToPrim = unsafePrimToPrim
@@ -261,6 +286,8 @@
         $ (primitive (\s -> case touch# x s of { s' -> (# s', () #) }) :: IO ())
 
 -- | Create an action to force a value; generalizes 'Control.Exception.evaluate'
+--
+-- @since 0.6.2.0
 evalPrim :: forall a m . PrimMonad m => a -> m a
 #if MIN_VERSION_base(4,4,0)
 evalPrim a = primitive (\s -> seq# a s)
diff --git a/Data/Primitive.hs b/Data/Primitive.hs
--- a/Data/Primitive.hs
+++ b/Data/Primitive.hs
@@ -11,15 +11,75 @@
 -- Reexports all primitive operations
 --
 module Data.Primitive (
-  module Data.Primitive.Types,
-  module Data.Primitive.Array,
-  module Data.Primitive.ByteArray,
-  module Data.Primitive.Addr,
-
-  sizeOf, alignment
+  -- * Re-exports
+  module Data.Primitive.Types
+  ,module Data.Primitive.Array
+  ,module Data.Primitive.ByteArray
+  ,module Data.Primitive.Addr
+  ,module Data.Primitive.SmallArray
+  ,module Data.Primitive.UnliftedArray
+  ,module Data.Primitive.PrimArray
+  ,module Data.Primitive.MutVar
+  -- * Naming Conventions
+  -- $namingConventions
 ) where
 
 import Data.Primitive.Types
 import Data.Primitive.Array
 import Data.Primitive.ByteArray
 import Data.Primitive.Addr
+import Data.Primitive.SmallArray
+import Data.Primitive.UnliftedArray
+import Data.Primitive.PrimArray
+import Data.Primitive.MutVar
+
+{- $namingConventions
+For historical reasons, this library embraces the practice of suffixing
+the name of a function with the type it operates on. For example, three
+of the variants of the array indexing function are:
+
+> indexArray      ::           Array      a -> Int -> a
+> indexSmallArray ::           SmallArray a -> Int -> a
+> indexPrimArray  :: Prim a => PrimArray  a -> Int -> a
+
+In a few places, where the language sounds more natural, the array type
+is instead used as a prefix. For example, @Data.Primitive.SmallArray@
+exports @smallArrayFromList@, which would sound unnatural if it used
+@SmallArray@ as a suffix instead.
+
+This library provides several functions traversing, building, and filtering
+arrays. These functions are suffixed with an additional character to
+indicate their the nature of their effectfulness:
+
+* No suffix: A non-effectful pass over the array.
+* @-A@ suffix: An effectful pass over the array, where the effect is 'Applicative'.
+* @-P@ suffix: An effectful pass over the array, where the effect is 'PrimMonad'.
+
+Additionally, an apostrophe can be used to indicate strictness in the elements.
+The variants with an apostrophe are used in @Data.Primitive.Array@ but not
+in @Data.Primitive.PrimArray@ since the array type it provides is always strict in the element.
+For example, there are three variants of the function that filters elements
+from a primitive array.
+
+> filterPrimArray  :: (Prim a               ) => (a ->   Bool) -> PrimArray a ->    PrimArray a
+> filterPrimArrayA :: (Prim a, Applicative f) => (a -> f Bool) -> PrimArray a -> f (PrimArray a)
+> filterPrimArrayP :: (Prim a, PrimMonad   m) => (a -> m Bool) -> PrimArray a -> m (PrimArray a)
+
+As long as the effectful context is a monad that is sufficiently affine
+the behaviors of the 'Applicative' and 'PrimMonad' variants produce the same results
+and differ only in their strictness. Monads that are sufficiently affine
+include:
+
+* 'IO' and 'ST'
+* Any combination of 'MaybeT', 'ExceptT', 'StateT' and 'Writer' on top
+  of another sufficiently affine monad.
+
+There is one situation where the names deviate from effectful suffix convention
+described above. Throughout the haskell ecosystem, the 'Applicative' variant of
+'map' is known as 'traverse', not @mapA@. Consequently, we adopt the following
+naming convention for mapping:
+
+> mapPrimArray :: (Prim a, Prim b) => (a -> b) -> PrimArray a -> PrimArray b
+> traversePrimArray :: (Applicative f, Prim a, Prim b) => (a -> f b) -> PrimArray a -> f (PrimArray b)
+> traversePrimArrayP :: (PrimMonad m, Prim a, Prim b) => (a -> m b) -> PrimArray a -> m (PrimArray b)
+-}
diff --git a/Data/Primitive/Addr.hs b/Data/Primitive/Addr.hs
--- a/Data/Primitive/Addr.hs
+++ b/Data/Primitive/Addr.hs
@@ -1,4 +1,4 @@
-{-# LANGUAGE MagicHash, UnboxedTuples #-}
+{-# LANGUAGE MagicHash, UnboxedTuples, CPP #-}
 
 -- |
 -- Module      : Data.Primitive.Addr
@@ -22,11 +22,21 @@
   indexOffAddr, readOffAddr, writeOffAddr,
 
   -- * Block operations
-  copyAddr, moveAddr, setAddr
+  copyAddr,
+#if __GLASGOW_HASKELL__ >= 708
+  copyAddrToByteArray,
+#endif
+  moveAddr, setAddr,
+
+  -- * Conversion
+  addrToInt
 ) where
 
 import Control.Monad.Primitive
 import Data.Primitive.Types
+#if __GLASGOW_HASKELL__ >= 708
+import Data.Primitive.ByteArray
+#endif
 
 import GHC.Base ( Int(..) )
 import GHC.Prim
@@ -84,6 +94,23 @@
 copyAddr (Addr dst#) (Addr src#) n
   = unsafePrimToPrim $ copyBytes (Ptr dst#) (Ptr src#) n
 
+#if __GLASGOW_HASKELL__ >= 708
+-- | Copy the given number of bytes from the 'Addr' to the 'MutableByteArray'.
+--   The areas may not overlap. This function is only available when compiling
+--   with GHC 7.8 or newer.
+--   
+--   @since 0.6.4.0
+copyAddrToByteArray :: PrimMonad m
+  => MutableByteArray (PrimState m) -- ^ destination
+  -> Int -- ^ offset into the destination array
+  -> Addr -- ^ source
+  -> Int -- ^ number of bytes to copy
+  -> m ()
+{-# INLINE copyAddrToByteArray #-}
+copyAddrToByteArray (MutableByteArray marr) (I# off) (Addr addr) (I# len) =
+  primitive_ $ copyAddrToByteArray# addr marr off len
+#endif
+
 -- | Copy the given number of bytes from the second 'Addr' to the first. The
 -- areas may overlap.
 moveAddr :: PrimMonad m => Addr         -- ^ destination address
@@ -100,3 +127,7 @@
 {-# INLINE setAddr #-}
 setAddr (Addr addr#) (I# n#) x = primitive_ (setOffAddr# addr# 0# n# x)
 
+-- | Convert an 'Addr' to an 'Int'.
+addrToInt :: Addr -> Int
+{-# INLINE addrToInt #-}
+addrToInt (Addr addr#) = I# (addr2Int# addr#)
diff --git a/Data/Primitive/Array.hs b/Data/Primitive/Array.hs
--- a/Data/Primitive/Array.hs
+++ b/Data/Primitive/Array.hs
@@ -16,13 +16,15 @@
 module Data.Primitive.Array (
   Array(..), MutableArray(..),
 
-  newArray, readArray, writeArray, indexArray, indexArrayM,
-  freezeArray, thawArray,
+  newArray, readArray, writeArray, indexArray, indexArrayM, indexArray##,
+  freezeArray, thawArray, runArray,
   unsafeFreezeArray, unsafeThawArray, sameMutableArray,
   copyArray, copyMutableArray,
   cloneArray, cloneMutableArray,
   sizeofArray, sizeofMutableArray,
-  fromListN, fromList
+  fromListN, fromList,
+  mapArray',
+  traverseArrayP
 ) where
 
 import Control.Monad.Primitive
@@ -42,7 +44,7 @@
 import Control.Monad.ST(ST,runST)
 
 import Control.Applicative
-import Control.Monad (MonadPlus(..))
+import Control.Monad (MonadPlus(..), when)
 import Control.Monad.Fix
 #if MIN_VERSION_base(4,4,0)
 import Control.Monad.Zip
@@ -53,31 +55,35 @@
 import Data.Monoid
 #endif
 #if MIN_VERSION_base(4,9,0)
+import qualified GHC.ST as GHCST
 import qualified Data.Foldable as F
 import Data.Semigroup
 #endif
+#if MIN_VERSION_base(4,8,0)
+import Data.Functor.Identity
+#endif
+#if MIN_VERSION_base(4,10,0)
+import GHC.Exts (runRW#)
+#elif MIN_VERSION_base(4,9,0)
+import GHC.Base (runRW#)
+#endif
 
 import Text.ParserCombinators.ReadP
 
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+import Data.Functor.Classes (Eq1(..),Ord1(..),Show1(..),Read1(..))
+#endif
+
 -- | Boxed arrays
 data Array a = Array
-             { array# :: Array# a
-#if (__GLASGOW_HASKELL__ < 702)
-             , sizeofArray :: {-# UNPACK #-} !Int
-#endif
-             }
+  { array# :: Array# a }
   deriving ( Typeable )
 
 -- | Mutable boxed arrays associated with a primitive state token.
 data MutableArray s a = MutableArray
-                      { marray# :: MutableArray# s a
-#if (__GLASGOW_HASKELL__ < 702)
-                      , sizeofMutableArray :: {-# UNPACK #-} !Int
-#endif
-                      }
+  { marray# :: MutableArray# s a }
   deriving ( Typeable )
 
-#if (__GLASGOW_HASKELL__ >= 702)
 sizeofArray :: Array a -> Int
 sizeofArray a = I# (sizeofArray# (array# a))
 {-# INLINE sizeofArray #-}
@@ -85,7 +91,6 @@
 sizeofMutableArray :: MutableArray s a -> Int
 sizeofMutableArray a = I# (sizeofMutableArray# (marray# a))
 {-# INLINE sizeofMutableArray #-}
-#endif
 
 -- | Create a new mutable array of the specified size and initialise all
 -- elements with the given value.
@@ -95,9 +100,6 @@
    (\s# -> case newArray# n# x s# of
              (# s'#, arr# #) ->
                let ma = MutableArray arr#
-#if (__GLASGOW_HASKELL__ < 702)
-                          (I# n#)
-#endif
                in (# s'# , ma #))
 
 -- | Read a value from the array at the given index.
@@ -115,6 +117,13 @@
 {-# INLINE indexArray #-}
 indexArray arr (I# i#) = case indexArray# (array# arr) i# of (# x #) -> x
 
+-- | Read a value from the immutable array at the given index, returning
+-- the result in an unboxed unary tuple. This is currently used to implement
+-- folds.
+indexArray## :: Array a -> Int -> (# a #)
+indexArray## arr (I# i) = indexArray# (array# arr) i
+{-# INLINE indexArray## #-}
+
 -- | Monadically read a value from the immutable array at the given index.
 -- This allows us to be strict in the array while remaining lazy in the read
 -- element which is very useful for collective operations. Suppose we want to
@@ -154,16 +163,9 @@
   -> Int                          -- ^ length
   -> m (Array a)
 {-# INLINE freezeArray #-}
-#if (__GLASGOW_HASKELL__ >= 702)
 freezeArray (MutableArray ma#) (I# off#) (I# len#) =
   primitive $ \s -> case freezeArray# ma# off# len# s of
     (# s', a# #) -> (# s', Array a# #)
-#else
-freezeArray src off len = do
-  dst <- newArray len (die "freezeArray" "impossible")
-  copyMutableArray dst 0 src off len
-  unsafeFreezeArray dst
-#endif
 
 -- | Convert a mutable array to an immutable one without copying. The
 -- array should not be modified after the conversion.
@@ -173,9 +175,6 @@
   = primitive (\s# -> case unsafeFreezeArray# (marray# arr) s# of
                         (# s'#, arr'# #) ->
                           let a = Array arr'#
-#if (__GLASGOW_HASKELL__ < 702)
-                                    (sizeofMutableArray arr)
-#endif
                           in (# s'#, a #))
 
 -- | Create a mutable array from a slice of an immutable array.
@@ -189,16 +188,9 @@
   -> Int     -- ^ length
   -> m (MutableArray (PrimState m) a)
 {-# INLINE thawArray #-}
-#if (__GLASGOW_HASKELL__ >= 702)
 thawArray (Array a#) (I# off#) (I# len#) =
   primitive $ \s -> case thawArray# a# off# len# s of
     (# s', ma# #) -> (# s', MutableArray ma# #)
-#else
-thawArray src off len = do
-  dst <- newArray len (die "thawArray" "impossible")
-  copyArray dst 0 src off len
-  return dst
-#endif
 
 -- | Convert an immutable array to an mutable one without copying. The
 -- immutable array should not be used after the conversion.
@@ -208,9 +200,6 @@
   = primitive (\s# -> case unsafeThawArray# (array# a) s# of
                         (# s'#, arr'# #) ->
                           let ma = MutableArray arr'#
-#if (__GLASGOW_HASKELL__ < 702)
-                                     (sizeofArray a)
-#endif
                           in (# s'#, ma #))
 
 -- | Check whether the two arrays refer to the same memory block.
@@ -275,15 +264,8 @@
            -> Int     -- ^ number of elements to copy
            -> Array a
 {-# INLINE cloneArray #-}
-#if __GLASGOW_HASKELL__ >= 702
 cloneArray (Array arr#) (I# off#) (I# len#)
   = case cloneArray# arr# off# len# of arr'# -> Array arr'#
-#else
-cloneArray arr off len = runST $ do
-    marr2 <- newArray len $ die "cloneArray" "impossible"
-    copyArray marr2 0 arr off len
-    unsafeFreezeArray marr2
-#endif
 
 -- | Return a newly allocated MutableArray. with the specified subrange of
 -- the provided MutableArray. The provided MutableArray should contain the
@@ -294,112 +276,209 @@
         -> Int                          -- ^ number of elements to copy
         -> m (MutableArray (PrimState m) a)
 {-# INLINE cloneMutableArray #-}
-#if __GLASGOW_HASKELL__ >= 702
 cloneMutableArray (MutableArray arr#) (I# off#) (I# len#) = primitive
    (\s# -> case cloneMutableArray# arr# off# len# s# of
              (# s'#, arr'# #) -> (# s'#, MutableArray arr'# #))
-#else
-cloneMutableArray marr off len = do
-        marr2 <- newArray len $ die "cloneMutableArray" "impossible"
-        let go !i !j c
-                | c >= len = return marr2
-                | otherwise = do
-                    b <- readArray marr i
-                    writeArray marr2 j b
-                    go (i+1) (j+1) (c+1)
-        go off 0 0
-#endif
 
 emptyArray :: Array a
 emptyArray =
   runST $ newArray 0 (die "emptyArray" "impossible") >>= unsafeFreezeArray
 {-# NOINLINE emptyArray #-}
 
+#if !MIN_VERSION_base(4,9,0)
 createArray
   :: Int
   -> a
   -> (forall s. MutableArray s a -> ST s ())
   -> Array a
 createArray 0 _ _ = emptyArray
-createArray n x f = runST $ do
-  ma <- newArray n x
-  f ma
-  unsafeFreezeArray ma
+createArray n x f = runArray $ do
+  mary <- newArray n x
+  f mary
+  pure mary
 
+runArray
+  :: (forall s. ST s (MutableArray s a))
+  -> Array a
+runArray m = runST $ m >>= unsafeFreezeArray
+
+#else /* Below, runRW# is available. */
+
+-- This low-level business is designed to work with GHC's worker-wrapper
+-- transformation. A lot of the time, we don't actually need an Array
+-- constructor. By putting it on the outside, and being careful about
+-- how we special-case the empty array, we can make GHC smarter about this.
+-- The only downside is that separately created 0-length arrays won't share
+-- their Array constructors, although they'll share their underlying
+-- Array#s.
+createArray
+  :: Int
+  -> a
+  -> (forall s. MutableArray s a -> ST s ())
+  -> Array a
+createArray 0 _ _ = Array (emptyArray# (# #))
+createArray n x f = runArray $ do
+  mary <- newArray n x
+  f mary
+  pure mary
+
+runArray
+  :: (forall s. ST s (MutableArray s a))
+  -> Array a
+runArray m = Array (runArray# m)
+
+runArray#
+  :: (forall s. ST s (MutableArray s a))
+  -> Array# a
+runArray# m = case runRW# $ \s ->
+  case unST m s of { (# s', MutableArray mary# #) ->
+  unsafeFreezeArray# mary# s'} of (# _, ary# #) -> ary#
+
+unST :: ST s a -> State# s -> (# State# s, a #)
+unST (GHCST.ST f) = f
+
+emptyArray# :: (# #) -> Array# a
+emptyArray# _ = case emptyArray of Array ar -> ar
+{-# NOINLINE emptyArray# #-}
+#endif
+
+
 die :: String -> String -> a
 die fun problem = error $ "Data.Primitive.Array." ++ fun ++ ": " ++ problem
 
+arrayLiftEq :: (a -> b -> Bool) -> Array a -> Array b -> Bool
+arrayLiftEq p a1 a2 = sizeofArray a1 == sizeofArray a2 && loop (sizeofArray a1 - 1)
+  where loop i | i < 0     = True
+               | (# x1 #) <- indexArray## a1 i
+               , (# x2 #) <- indexArray## a2 i
+               , otherwise = p x1 x2 && loop (i-1)
+
 instance Eq a => Eq (Array a) where
-  a1 == a2 = sizeofArray a1 == sizeofArray a2 && loop (sizeofArray a1 - 1)
-   where loop i | i < 0     = True
-                | otherwise = indexArray a1 i == indexArray a2 i && loop (i-1)
+  a1 == a2 = arrayLiftEq (==) a1 a2
 
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+-- | @since 0.6.4.0
+instance Eq1 Array where
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,5,0)
+  liftEq = arrayLiftEq
+#else
+  eq1 = arrayLiftEq (==)
+#endif
+#endif
+
 instance Eq (MutableArray s a) where
   ma1 == ma2 = isTrue# (sameMutableArray# (marray# ma1) (marray# ma2))
 
+arrayLiftCompare :: (a -> b -> Ordering) -> Array a -> Array b -> Ordering
+arrayLiftCompare elemCompare a1 a2 = loop 0
+  where
+  mn = sizeofArray a1 `min` sizeofArray a2
+  loop i
+    | i < mn
+    , (# x1 #) <- indexArray## a1 i
+    , (# x2 #) <- indexArray## a2 i
+    = elemCompare x1 x2 `mappend` loop (i+1)
+    | otherwise = compare (sizeofArray a1) (sizeofArray a2)
+
+-- | Lexicographic ordering. Subject to change between major versions.
 instance Ord a => Ord (Array a) where
-  compare a1 a2 = loop 0
-   where
-   mn = sizeofArray a1 `min` sizeofArray a2
-   loop i
-     | i < mn    = compare (indexArray a1 i) (indexArray a2 i) `mappend` loop (i+1)
-     | otherwise = compare (sizeofArray a1) (sizeofArray a2)
+  compare a1 a2 = arrayLiftCompare compare a1 a2
 
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+-- | @since 0.6.4.0
+instance Ord1 Array where
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,5,0)
+  liftCompare = arrayLiftCompare
+#else
+  compare1 = arrayLiftCompare compare
+#endif
+#endif
+
 instance Foldable Array where
-  foldr f z a = go 0
-   where go i | i < sizeofArray a = f (indexArray a i) (go $ i+1)
-              | otherwise         = z
+  -- Note: we perform the array lookups eagerly so we won't
+  -- create thunks to perform lookups even if GHC can't see
+  -- that the folding function is strict.
+  foldr f = \z !ary ->
+    let
+      !sz = sizeofArray ary
+      go i
+        | i == sz = z
+        | (# x #) <- indexArray## ary i
+        = f x (go (i+1))
+    in go 0
   {-# INLINE foldr #-}
-  foldl f z a = go (sizeofArray a - 1)
-   where go i | i < 0     = z
-              | otherwise = f (go $ i-1) (indexArray a i)
+  foldl f = \z !ary ->
+    let
+      go i
+        | i < 0 = z
+        | (# x #) <- indexArray## ary i
+        = f (go (i-1)) x
+    in go (sizeofArray ary - 1)
   {-# INLINE foldl #-}
-  foldr1 f a | sz < 0    = die "foldr1" "empty array"
-             | otherwise = go 0
-   where sz = sizeofArray a - 1
-         z = indexArray a sz
-         go i | i < sz    = f (indexArray a i) (go $ i+1)
-              | otherwise = z
+  foldr1 f = \ !ary ->
+    let
+      !sz = sizeofArray ary - 1
+      go i =
+        case indexArray## ary i of
+          (# x #) | i == sz -> x
+                  | otherwise -> f x (go (i+1))
+    in if sz < 0
+       then die "foldr1" "empty array"
+       else go 0
   {-# INLINE foldr1 #-}
-  foldl1 f a | sz == 0   = die "foldl1" "empty array"
-             | otherwise = go $ sz-1
-   where sz = sizeofArray a
-         z = indexArray a 0
-         go i | i < 1     = f (go $ i-1) (indexArray a i)
-              | otherwise = z
+  foldl1 f = \ !ary ->
+    let
+      !sz = sizeofArray ary - 1
+      go i =
+        case indexArray## ary i of
+          (# x #) | i == 0 -> x
+                  | otherwise -> f (go (i - 1)) x
+    in if sz < 0
+       then die "foldl1" "empty array"
+       else go sz
   {-# INLINE foldl1 #-}
 #if MIN_VERSION_base(4,6,0)
-  foldr' f z a = go (sizeofArray a - 1) z
-   where go i !acc | i < 0     = acc
-                   | otherwise = go (i-1) (f (indexArray a i) acc)
+  foldr' f = \z !ary ->
+    let
+      go i !acc
+        | i == -1 = acc
+        | (# x #) <- indexArray## ary i
+        = go (i-1) (f x acc)
+    in go (sizeofArray ary - 1) z
   {-# INLINE foldr' #-}
-  foldl' f z a = go 0 z
-   where go i !acc | i < sizeofArray a = go (i+1) (f acc $ indexArray a i)
-                   | otherwise         = acc
+  foldl' f = \z !ary ->
+    let
+      !sz = sizeofArray ary
+      go i !acc
+        | i == sz = acc
+        | (# x #) <- indexArray## ary i
+        = go (i+1) (f acc x)
+    in go 0 z
   {-# INLINE foldl' #-}
 #endif
 #if MIN_VERSION_base(4,8,0)
-  toList a = Exts.build $ \c z -> let
-      sz = sizeofArray a
-      go i | i < sz    = c (indexArray a i) (go $ i+1)
-           | otherwise = z
-    in go 0
-  {-# INLINE toList #-}
   null a = sizeofArray a == 0
   {-# INLINE null #-}
   length = sizeofArray
   {-# INLINE length #-}
-  maximum a | sz == 0   = die "maximum" "empty array"
-            | otherwise = go 1 (indexArray a 0)
-   where sz = sizeofArray a
-         go i !e | i < sz    = go (i+1) (max e $ indexArray a i)
-                 | otherwise = e
+  maximum ary | sz == 0   = die "maximum" "empty array"
+              | (# frst #) <- indexArray## ary 0
+              = go 1 frst
+   where
+     sz = sizeofArray ary
+     go i !e
+       | i == sz = e
+       | (# x #) <- indexArray## ary i
+       = go (i+1) (max e x)
   {-# INLINE maximum #-}
-  minimum a | sz == 0   = die "minimum" "empty array"
-            | otherwise = go 1 (indexArray a 0)
-   where sz = sizeofArray a
-         go i !e | i < sz    = go (i+1) (min e $ indexArray a i)
-                 | otherwise = e
+  minimum ary | sz == 0   = die "minimum" "empty array"
+              | (# frst #) <- indexArray## ary 0
+              = go 1 frst
+   where sz = sizeofArray ary
+         go i !e
+           | i == sz = e
+           | (# x #) <- indexArray## ary i
+           = go (i+1) (min e x)
   {-# INLINE minimum #-}
   sum = foldl' (+) 0
   {-# INLINE sum #-}
@@ -407,56 +486,155 @@
   {-# INLINE product #-}
 #endif
 
+newtype STA a = STA {_runSTA :: forall s. MutableArray# s a -> ST s (Array a)}
+
+runSTA :: Int -> STA a -> Array a
+runSTA !sz = \ (STA m) -> runST $ newArray_ sz >>= \ ar -> m (marray# ar)
+{-# INLINE runSTA #-}
+
+newArray_ :: Int -> ST s (MutableArray s a)
+newArray_ !n = newArray n badTraverseValue
+
+badTraverseValue :: a
+badTraverseValue = die "traverse" "bad indexing"
+{-# NOINLINE badTraverseValue #-}
+
 instance Traversable Array where
-  traverse f a =
-    fromListN (sizeofArray a)
-      <$> traverse (f . indexArray a) [0 .. sizeofArray a - 1]
+  traverse f = traverseArray f
+  {-# INLINE traverse #-}
 
+traverseArray
+  :: Applicative f
+  => (a -> f b)
+  -> Array a
+  -> f (Array b)
+traverseArray f = \ !ary ->
+  let
+    !len = sizeofArray ary
+    go !i
+      | i == len = pure $ STA $ \mary -> unsafeFreezeArray (MutableArray mary)
+      | (# x #) <- indexArray## ary i
+      = liftA2 (\b (STA m) -> STA $ \mary ->
+                  writeArray (MutableArray mary) i b >> m mary)
+               (f x) (go (i + 1))
+  in if len == 0
+     then pure emptyArray
+     else runSTA len <$> go 0
+{-# INLINE [1] traverseArray #-}
+
+{-# RULES
+"traverse/ST" forall (f :: a -> ST s b). traverseArray f =
+   traverseArrayP f
+"traverse/IO" forall (f :: a -> IO b). traverseArray f =
+   traverseArrayP f
+ #-}
+#if MIN_VERSION_base(4,8,0)
+{-# RULES
+"traverse/Id" forall (f :: a -> Identity b). traverseArray f =
+   (coerce :: (Array a -> Array (Identity b))
+           -> Array a -> Identity (Array b)) (fmap f)
+ #-}
+#endif
+
+-- | This is the fastest, most straightforward way to traverse
+-- an array, but it only works correctly with a sufficiently
+-- "affine" 'PrimMonad' instance. In particular, it must only produce
+-- *one* result array. 'Control.Monad.Trans.List.ListT'-transformed
+-- monads, for example, will not work right at all.
+traverseArrayP
+  :: PrimMonad m
+  => (a -> m b)
+  -> Array a
+  -> m (Array b)
+traverseArrayP f = \ !ary ->
+  let
+    !sz = sizeofArray ary
+    go !i !mary
+      | i == sz
+      = unsafeFreezeArray mary
+      | otherwise
+      = do
+          a <- indexArrayM ary i
+          b <- f a
+          writeArray mary i b
+          go (i + 1) mary
+  in do
+    mary <- newArray sz badTraverseValue
+    go 0 mary
+{-# INLINE traverseArrayP #-}
+
+-- | Strict map over the elements of the array.
+mapArray' :: (a -> b) -> Array a -> Array b
+mapArray' f a =
+  createArray (sizeofArray a) (die "mapArray'" "impossible") $ \mb ->
+    let go i | i == sizeofArray a
+             = return ()
+             | otherwise
+             = do x <- indexArrayM a i
+                  -- We use indexArrayM here so that we will perform the
+                  -- indexing eagerly even if f is lazy.
+                  let !y = f x
+                  writeArray mb i y >> go (i+1)
+     in go 0
+{-# INLINE mapArray' #-}
+
+arrayFromListN :: Int -> [a] -> Array a
+arrayFromListN n l =
+  createArray n (die "fromListN" "uninitialized element") $ \sma ->
+    let go !ix [] = if ix == n
+          then return ()
+          else die "fromListN" "list length less than specified size"
+        go !ix (x : xs) = if ix < n
+          then do
+            writeArray sma ix x
+            go (ix+1) xs
+          else die "fromListN" "list length greater than specified size"
+    in go 0 l
+
+arrayFromList :: [a] -> Array a
+arrayFromList l = arrayFromListN (length l) l
+
 #if MIN_VERSION_base(4,7,0)
 instance Exts.IsList (Array a) where
   type Item (Array a) = a
-  fromListN n l =
-    createArray n (die "fromListN" "mismatched size and list") $ \mi ->
-      let go i (x:xs) = writeArray mi i x >> go (i+1) xs
-          go _ [    ] = return ()
-       in go 0 l
-  fromList l = Exts.fromListN (length l) l
+  fromListN = arrayFromListN
+  fromList = arrayFromList
   toList = toList
 #else
 fromListN :: Int -> [a] -> Array a
-fromListN n l =
-  createArray n (die "fromListN" "mismatched size and list") $ \mi ->
-    let go i (x:xs) = writeArray mi i x >> go (i+1) xs
-        go _ [    ] = return ()
-     in go 0 l
+fromListN = arrayFromListN
 
 fromList :: [a] -> Array a
-fromList l = fromListN (length l) l
+fromList = arrayFromList
 #endif
 
 instance Functor Array where
   fmap f a =
     createArray (sizeofArray a) (die "fmap" "impossible") $ \mb ->
-      let go i | i < sizeofArray a = return ()
-               | otherwise         = writeArray mb i (f $ indexArray a i)
-                                  >> go (i+1)
+      let go i | i == sizeofArray a
+               = return ()
+               | otherwise
+               = do x <- indexArrayM a i
+                    writeArray mb i (f x) >> go (i+1)
        in go 0
 #if MIN_VERSION_base(4,8,0)
-  e <$ a = runST $ newArray (sizeofArray a) e >>= unsafeFreezeArray
+  e <$ a = createArray (sizeofArray a) e (\ !_ -> pure ())
 #endif
 
 instance Applicative Array where
-  pure x = runST $ newArray 1 x >>= unsafeFreezeArray
-  ab <*> a = runST $ do
-    mb <- newArray (szab*sza) $ die "<*>" "impossible"
-    let go1 i
-          | i < szab  = go2 (i*sza) (indexArray ab i) 0 >> go1 (i+1)
-          | otherwise = return ()
-        go2 off f j
-          | j < sza   = writeArray mb (off + j) (f $ indexArray a j)
-          | otherwise = return ()
-    go1 0
-    unsafeFreezeArray mb
+  pure x = runArray $ newArray 1 x
+  ab <*> a = createArray (szab*sza) (die "<*>" "impossible") $ \mb ->
+    let go1 i = when (i < szab) $
+            do
+              f <- indexArrayM ab i
+              go2 (i*sza) f 0
+              go1 (i+1)
+        go2 off f j = when (j < sza) $
+            do
+              x <- indexArrayM a j
+              writeArray mb (off + j) (f x)
+              go2 off f (j + 1)
+    in go1 0
    where szab = sizeofArray ab ; sza = sizeofArray a
   a *> b = createArray (sza*szb) (die "*>" "impossible") $ \mb ->
     let go i | i < sza   = copyArray mb (i * szb) b 0 szb
@@ -466,7 +644,9 @@
   a <* b = createArray (sza*szb) (die "<*" "impossible") $ \ma ->
     let fill off i e | i < szb   = writeArray ma (off+i) e >> fill off (i+1) e
                      | otherwise = return ()
-        go i | i < sza   = fill (i*szb) 0 (indexArray a i) >> go (i+1)
+        go i | i < sza
+             = do x <- indexArrayM a i
+                  fill (i*szb) 0 x >> go (i+1)
              | otherwise = return ()
      in go 0
    where sza = sizeofArray a ; szb = sizeofArray b
@@ -481,20 +661,36 @@
   many a | sizeofArray a == 0 = pure []
          | otherwise = die "many" "infinite arrays are not well defined"
 
+data ArrayStack a
+  = PushArray !(Array a) !(ArrayStack a)
+  | EmptyStack
+-- See the note in SmallArray about how we might improve this.
+
 instance Monad Array where
   return = pure
   (>>) = (*>)
-  a >>= f = push 0 [] (sizeofArray a - 1)
+
+  ary >>= f = collect 0 EmptyStack (la-1)
    where
-   push !sz bs i
-     | i < 0 = build sz bs
-     | otherwise = let b = f $ indexArray a i
-                    in push (sz + sizeofArray b) (b:bs) (i+1)
+   la = sizeofArray ary
+   collect sz stk i
+     | i < 0 = createArray sz (die ">>=" "impossible") $ fill 0 stk
+     | (# x #) <- indexArray## ary i
+     , let sb = f x
+           lsb = sizeofArray sb
+       -- If we don't perform this check, we could end up allocating
+       -- a stack full of empty arrays if someone is filtering most
+       -- things out. So we refrain from pushing empty arrays.
+     = if lsb == 0
+       then collect sz stk (i - 1)
+       else collect (sz + lsb) (PushArray sb stk) (i-1)
 
-   build sz stk = createArray sz (die ">>=" "impossible") $ \mb ->
-     let go off (b:bs) = copyArray mb off b 0 (sizeofArray b) >> go (off + sizeofArray b) bs
-         go _   [    ] = return ()
-      in go 0 stk
+   fill _   EmptyStack         _   = return ()
+   fill off (PushArray sb sbs) smb
+     | let lsb = sizeofArray sb
+     = copyArray smb off sb 0 (lsb)
+         *> fill (off + lsb) sbs smb
+
   fail _ = empty
 
 instance MonadPlus Array where
@@ -503,10 +699,13 @@
 
 zipW :: String -> (a -> b -> c) -> Array a -> Array b -> Array c
 zipW s f aa ab = createArray mn (die s "impossible") $ \mc ->
-  let go i
-        | i < mn    = writeArray mc i (f (indexArray aa i) (indexArray ab i))
-                   >> go (i+1)
-        | otherwise = return ()
+  let go i | i < mn
+           = do
+               x <- indexArrayM aa i
+               y <- indexArrayM ab i
+               writeArray mc i (f x y)
+               go (i+1)
+           | otherwise = return ()
    in go 0
  where mn = sizeofArray aa `min` sizeofArray ab
 {-# INLINE zipW #-}
@@ -520,7 +719,7 @@
     ma <- newArray sz (die "munzip" "impossible")
     mb <- newArray sz (die "munzip" "impossible")
     let go i | i < sz = do
-          let (a, b) = indexArray aab i
+          (a, b) <- indexArrayM aab i
           writeArray ma i a
           writeArray mb i b
           go (i+1)
@@ -530,9 +729,17 @@
 #endif
 
 instance MonadFix Array where
-  mfix f = let l = mfix (toList . f) in fromListN (length l) l
+  mfix f = createArray (sizeofArray (f err))
+                       (die "mfix" "impossible") $ flip fix 0 $
+    \r !i !mary -> when (i < sz) $ do
+                      writeArray mary i (fix (\xi -> f xi `indexArray` i))
+                      r (i + 1) mary
+    where
+      sz = sizeofArray (f err)
+      err = error "mfix for Data.Primitive.Array applied to strict function."
 
 #if MIN_VERSION_base(4,9,0)
+-- | @since 0.6.3.0
 instance Semigroup (Array a) where
   (<>) = (<|>)
   sconcat = mconcat . F.toList
@@ -550,19 +757,50 @@
      in go 0 l
    where sz = sum . fmap sizeofArray $ l
 
+arrayLiftShowsPrec :: (Int -> a -> ShowS) -> ([a] -> ShowS) -> Int -> Array a -> ShowS
+arrayLiftShowsPrec elemShowsPrec elemListShowsPrec p a = showParen (p > 10) $
+  showString "fromListN " . shows (sizeofArray a) . showString " "
+    . listLiftShowsPrec elemShowsPrec elemListShowsPrec 11 (toList a)
+
+-- this need to be included for older ghcs
+listLiftShowsPrec :: (Int -> a -> ShowS) -> ([a] -> ShowS) -> Int -> [a] -> ShowS
+listLiftShowsPrec _ sl _ = sl
+
 instance Show a => Show (Array a) where
-  showsPrec p a = showParen (p > 10) $
-    showString "fromListN " . shows (sizeofArray a) . showString " "
-      . shows (toList a)
+  showsPrec p a = arrayLiftShowsPrec showsPrec showList p a
 
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+-- | @since 0.6.4.0
+instance Show1 Array where
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,5,0)
+  liftShowsPrec = arrayLiftShowsPrec
+#else
+  showsPrec1 = arrayLiftShowsPrec showsPrec showList
+#endif
+#endif
+
+arrayLiftReadsPrec :: (Int -> ReadS a) -> ReadS [a] -> Int -> ReadS (Array a)
+arrayLiftReadsPrec _ listReadsPrec p = readParen (p > 10) . readP_to_S $ do
+  () <$ string "fromListN"
+  skipSpaces
+  n <- readS_to_P reads
+  skipSpaces
+  l <- readS_to_P listReadsPrec
+  return $ arrayFromListN n l
+
 instance Read a => Read (Array a) where
-  readsPrec p = readParen (p > 10) . readP_to_S $ do
-    () <$ string "fromListN"
-    skipSpaces
-    n <- readS_to_P reads
-    skipSpaces
-    l <- readS_to_P reads
-    return $ fromListN n l
+  readsPrec = arrayLiftReadsPrec readsPrec readList
+
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+-- | @since 0.6.4.0
+instance Read1 Array where
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,5,0)
+  liftReadsPrec = arrayLiftReadsPrec
+#else
+  readsPrec1 = arrayLiftReadsPrec readsPrec readList
+#endif
+#endif
+
 
 arrayDataType :: DataType
 arrayDataType = mkDataType "Data.Primitive.Array.Array" [fromListConstr]
diff --git a/Data/Primitive/ByteArray.hs b/Data/Primitive/ByteArray.hs
--- a/Data/Primitive/ByteArray.hs
+++ b/Data/Primitive/ByteArray.hs
@@ -1,4 +1,4 @@
-{-# LANGUAGE CPP, MagicHash, UnboxedTuples, UnliftedFFITypes, DeriveDataTypeable #-}
+{-# LANGUAGE BangPatterns, CPP, MagicHash, UnboxedTuples, UnliftedFFITypes, DeriveDataTypeable #-}
 {-# LANGUAGE ScopedTypeVariables #-}
 {-# LANGUAGE TypeFamilies #-}
 
@@ -19,10 +19,14 @@
 
   -- * Allocation
   newByteArray, newPinnedByteArray, newAlignedPinnedByteArray,
+  resizeMutableByteArray,
 
   -- * Element access
   readByteArray, writeByteArray, indexByteArray,
 
+  -- * Constructing
+  byteArrayFromList, byteArrayFromListN,
+
   -- * Folding
   foldrByteArray,
 
@@ -30,17 +34,25 @@
   unsafeFreezeByteArray, unsafeThawByteArray,
 
   -- * Block operations
-  copyByteArray, copyMutableByteArray, moveByteArray,
+  copyByteArray, copyMutableByteArray,
+#if __GLASGOW_HASKELL__ >= 708
+  copyByteArrayToAddr, copyMutableByteArrayToAddr,
+#endif
+  moveByteArray,
   setByteArray, fillByteArray,
 
   -- * Information
-  sizeofByteArray, sizeofMutableByteArray, sameMutableByteArray,
+  sizeofByteArray,
+  sizeofMutableByteArray, getSizeofMutableByteArray, sameMutableByteArray,
+#if __GLASGOW_HASKELL__ >= 802
+  isByteArrayPinned, isMutableByteArrayPinned,
+#endif
   byteArrayContents, mutableByteArrayContents
+
 ) where
 
 import Control.Monad.Primitive
 import Control.Monad.ST
-import Control.Monad ( zipWithM_ )
 import Data.Primitive.Types
 
 import Foreign.C.Types
@@ -58,8 +70,26 @@
 import Data.Data ( Data(..) )
 import Data.Primitive.Internal.Compat ( isTrue#, mkNoRepType )
 import Numeric
-import System.IO.Unsafe
 
+#if MIN_VERSION_base(4,9,0)
+import qualified Data.Semigroup as SG
+import qualified Data.Foldable as F
+#endif
+
+#if !(MIN_VERSION_base(4,8,0))
+import Data.Monoid (Monoid(..))
+#endif
+
+#if __GLASGOW_HASKELL__ >= 802
+import GHC.Exts as Exts (isByteArrayPinned#,isMutableByteArrayPinned#)
+#endif
+
+#if __GLASGOW_HASKELL__ >= 804
+import GHC.Exts (compareByteArrays#)
+#else
+import System.IO.Unsafe (unsafeDupablePerformIO)
+#endif
+
 -- | Byte arrays
 data ByteArray = ByteArray ByteArray# deriving ( Typeable )
 
@@ -83,9 +113,12 @@
                         (# s'#, arr# #) -> (# s'#, MutableByteArray arr# #))
 
 -- | Create a /pinned/ byte array of the specified size in bytes and with the
--- give alignment. The garbage collector is guaranteed not to move it.
+-- given alignment. The garbage collector is guaranteed not to move it.
 newAlignedPinnedByteArray
-  :: PrimMonad m => Int -> Int -> m (MutableByteArray (PrimState m))
+  :: PrimMonad m
+  => Int  -- ^ size
+  -> Int  -- ^ alignment
+  -> m (MutableByteArray (PrimState m))
 {-# INLINE newAlignedPinnedByteArray #-}
 newAlignedPinnedByteArray (I# n#) (I# k#)
   = primitive (\s# -> case newAlignedPinnedByteArray# n# k# s# of
@@ -112,6 +145,47 @@
 sameMutableByteArray (MutableByteArray arr#) (MutableByteArray brr#)
   = isTrue# (sameMutableByteArray# arr# brr#)
 
+-- | Resize a mutable byte array. The new size is given in bytes.
+--
+-- This will either resize the array in-place or, if not possible, allocate the
+-- contents into a new, unpinned array and copy the original array's contents.
+--
+-- To avoid undefined behaviour, the original 'MutableByteArray' shall not be
+-- accessed anymore after a 'resizeMutableByteArray' has been performed.
+-- Moreover, no reference to the old one should be kept in order to allow
+-- garbage collection of the original 'MutableByteArray' in case a new
+-- 'MutableByteArray' had to be allocated.
+--
+-- @since 0.6.4.0
+resizeMutableByteArray
+  :: PrimMonad m => MutableByteArray (PrimState m) -> Int
+                 -> m (MutableByteArray (PrimState m))
+{-# INLINE resizeMutableByteArray #-}
+#if __GLASGOW_HASKELL__ >= 710
+resizeMutableByteArray (MutableByteArray arr#) (I# n#)
+  = primitive (\s# -> case resizeMutableByteArray# arr# n# s# of
+                        (# s'#, arr'# #) -> (# s'#, MutableByteArray arr'# #))
+#else
+resizeMutableByteArray arr n
+  = do arr' <- newByteArray n
+       copyMutableByteArray arr' 0 arr 0 (min (sizeofMutableByteArray arr) n)
+       return arr'
+#endif
+
+-- | Get the size of a byte array in bytes. Unlike 'sizeofMutableByteArray',
+-- this function ensures sequencing in the presence of resizing.
+getSizeofMutableByteArray
+  :: PrimMonad m => MutableByteArray (PrimState m) -> m Int
+{-# INLINE getSizeofMutableByteArray #-}
+#if __GLASGOW_HASKELL__ >= 801
+getSizeofMutableByteArray (MutableByteArray arr#)
+  = primitive (\s# -> case getSizeofMutableByteArray# arr# s# of
+                        (# s'#, n# #) -> (# s'#, I# n# #))
+#else
+getSizeofMutableByteArray arr
+  = return (sizeofMutableByteArray arr)
+#endif
+
 -- | Convert a mutable byte array to an immutable one without copying. The
 -- array should not be modified after the conversion.
 unsafeFreezeByteArray
@@ -134,11 +208,35 @@
 {-# INLINE sizeofByteArray #-}
 sizeofByteArray (ByteArray arr#) = I# (sizeofByteArray# arr#)
 
--- | Size of the mutable byte array in bytes.
+-- | Size of the mutable byte array in bytes. This function\'s behavior 
+-- is undefined if 'resizeMutableByteArray' is ever called on the mutable
+-- byte array given as the argument. Consequently, use of this function
+-- is discouraged. Prefer 'getSizeofMutableByteArray', which ensures correct
+-- sequencing in the presence of resizing.
 sizeofMutableByteArray :: MutableByteArray s -> Int
 {-# INLINE sizeofMutableByteArray #-}
 sizeofMutableByteArray (MutableByteArray arr#) = I# (sizeofMutableByteArray# arr#)
 
+#if __GLASGOW_HASKELL__ >= 802
+-- | Check whether or not the byte array is pinned. Pinned byte arrays cannot
+--   be moved by the garbage collector. It is safe to use 'byteArrayContents'
+--   on such byte arrays. This function is only available when compiling with
+--   GHC 8.2 or newer.
+--
+--   @since 0.6.4.0
+isByteArrayPinned :: ByteArray -> Bool
+{-# INLINE isByteArrayPinned #-}
+isByteArrayPinned (ByteArray arr#) = isTrue# (Exts.isByteArrayPinned# arr#)
+
+-- | Check whether or not the mutable byte array is pinned. This function is
+--   only available when compiling with GHC 8.2 or newer.
+--
+--   @since 0.6.4.0
+isMutableByteArrayPinned :: MutableByteArray s -> Bool
+{-# INLINE isMutableByteArrayPinned #-}
+isMutableByteArrayPinned (MutableByteArray marr#) = isTrue# (Exts.isMutableByteArrayPinned# marr#)
+#endif
+
 -- | Read a primitive value from the byte array. The offset is given in
 -- elements of type @a@ rather than in bytes.
 indexByteArray :: Prim a => ByteArray -> Int -> a
@@ -168,18 +266,27 @@
     go i
       | sizeofByteArray arr > i * sz = f (indexByteArray arr i) (go (i+1))
       | otherwise                    = z
-    sz = sizeofByteArray arr
+    sz = sizeOf (undefined :: a)
 
-fromListN :: Prim a => Int -> [a] -> ByteArray
-fromListN n xs = runST $ do
-    marr <- newByteArray (n * sizeOf (head xs))
-    zipWithM_ (writeByteArray marr) [0..n] xs
+byteArrayFromList :: Prim a => [a] -> ByteArray
+byteArrayFromList xs = byteArrayFromListN (length xs) xs
+
+byteArrayFromListN :: Prim a => Int -> [a] -> ByteArray
+byteArrayFromListN n ys = runST $ do
+    marr <- newByteArray (n * sizeOf (head ys))
+    let go !ix [] = if ix == n
+          then return ()
+          else die "byteArrayFromListN" "list length less than specified size"
+        go !ix (x : xs) = if ix < n
+          then do
+            writeByteArray marr ix x
+            go (ix + 1) xs
+          else die "byteArrayFromListN" "list length greater than specified size"
+    go 0 ys
     unsafeFreezeByteArray marr
 
-#if __GLASGOW_HASKELL__ >= 702
 unI# :: Int -> Int#
 unI# (I# n#) = n#
-#endif
 
 -- | Copy a slice of an immutable byte array to a mutable byte array.
 copyByteArray
@@ -192,13 +299,7 @@
                  -> m ()
 {-# INLINE copyByteArray #-}
 copyByteArray (MutableByteArray dst#) doff (ByteArray src#) soff sz
-#if __GLASGOW_HASKELL__ >= 702
   = primitive_ (copyByteArray# src# (unI# soff) dst# (unI# doff) (unI# sz))
-#else
-  = unsafePrimToPrim
-  $ memcpy_ba dst# (fromIntegral doff) src# (fromIntegral soff)
-                 (fromIntegral sz)
-#endif
 
 -- | Copy a slice of a mutable byte array into another array. The two slices
 -- may not overlap.
@@ -214,12 +315,40 @@
 {-# INLINE copyMutableByteArray #-}
 copyMutableByteArray (MutableByteArray dst#) doff
                      (MutableByteArray src#) soff sz
-#if __GLASGOW_HASKELL__ >= 702
   = primitive_ (copyMutableByteArray# src# (unI# soff) dst# (unI# doff) (unI# sz))
-#else
-  = unsafePrimToPrim
-  $ memcpy_mba dst# (fromIntegral doff) src# (fromIntegral soff)
-                    (fromIntegral sz)
+
+#if __GLASGOW_HASKELL__ >= 708
+-- | Copy a slice of a byte array to an unmanaged address. These must not
+--   overlap. This function is only available when compiling with GHC 7.8
+--   or newer.
+--
+--   @since 0.6.4.0
+copyByteArrayToAddr
+  :: PrimMonad m
+  => Addr -- ^ destination
+  -> ByteArray -- ^ source array
+  -> Int -- ^ offset into source array
+  -> Int -- ^ number of bytes to copy
+  -> m ()
+{-# INLINE copyByteArrayToAddr #-}
+copyByteArrayToAddr (Addr dst#) (ByteArray src#) soff sz
+  = primitive_ (copyByteArrayToAddr# src# (unI# soff) dst# (unI# sz))
+
+-- | Copy a slice of a mutable byte array to an unmanaged address. These must
+--   not overlap. This function is only available when compiling with GHC 7.8
+--   or newer.
+--
+--   @since 0.6.4.0
+copyMutableByteArrayToAddr
+  :: PrimMonad m
+  => Addr -- ^ destination
+  -> MutableByteArray (PrimState m) -- ^ source array
+  -> Int -- ^ offset into source array
+  -> Int -- ^ number of bytes to copy
+  -> m ()
+{-# INLINE copyMutableByteArrayToAddr #-}
+copyMutableByteArrayToAddr (Addr dst#) (MutableByteArray src#) soff sz
+  = primitive_ (copyMutableByteArrayToAddr# src# (unI# soff) dst# (unI# sz))
 #endif
 
 -- | Copy a slice of a mutable byte array into another, potentially
@@ -263,18 +392,6 @@
 {-# INLINE fillByteArray #-}
 fillByteArray = setByteArray
 
-#if __GLASGOW_HASKELL__ < 702
-foreign import ccall unsafe "primitive-memops.h hsprimitive_memcpy"
-  memcpy_mba :: MutableByteArray# s -> CInt
-             -> MutableByteArray# s -> CInt
-             -> CSize -> IO ()
-
-foreign import ccall unsafe "primitive-memops.h hsprimitive_memcpy"
-  memcpy_ba :: MutableByteArray# s -> CInt
-            -> ByteArray# -> CInt
-            -> CSize -> IO ()
-#endif
-
 foreign import ccall unsafe "primitive-memops.h hsprimitive_memmove"
   memmove_mba :: MutableByteArray# s -> CInt
               -> MutableByteArray# s -> CInt
@@ -290,6 +407,7 @@
   gunfold _ _ = error "gunfold"
   dataTypeOf _ = mkNoRepType "Data.Primitive.ByteArray.MutableByteArray"
 
+-- | @since 0.6.3.0
 instance Show ByteArray where
   showsPrec _ ba =
       showString "[" . go 0
@@ -301,9 +419,25 @@
           comma | i == 0    = id
                 | otherwise = showString ", "
 
+
+compareByteArrays :: ByteArray -> ByteArray -> Int -> Ordering
+{-# INLINE compareByteArrays #-}
+#if __GLASGOW_HASKELL__ >= 804
+compareByteArrays (ByteArray ba1#) (ByteArray ba2#) (I# n#) =
+  compare (I# (compareByteArrays# ba1# 0# ba2# 0# n#)) 0
+#else
+-- Emulate GHC 8.4's 'GHC.Prim.compareByteArrays#'
+compareByteArrays (ByteArray ba1#) (ByteArray ba2#) (I# n#)
+    = compare (fromCInt (unsafeDupablePerformIO (memcmp_ba ba1# ba2# n))) 0
+  where
+    n = fromIntegral (I# n#) :: CSize
+    fromCInt = fromIntegral :: CInt -> Int
+
 foreign import ccall unsafe "primitive-memops.h hsprimitive_memcmp"
   memcmp_ba :: ByteArray# -> ByteArray# -> CSize -> IO CInt
+#endif
 
+
 sameByteArray :: ByteArray# -> ByteArray# -> Bool
 sameByteArray ba1 ba2 =
     case reallyUnsafePtrEquality# (unsafeCoerce# ba1 :: ()) (unsafeCoerce# ba2 :: ()) of
@@ -314,33 +448,102 @@
       0# -> False
 #endif
 
+-- | @since 0.6.3.0
 instance Eq ByteArray where
   ba1@(ByteArray ba1#) == ba2@(ByteArray ba2#)
-    | sameByteArray ba1# ba2#                    = True
-    | sizeofByteArray ba1 /= sizeofByteArray ba2 = False
-    | otherwise =
-        case unsafeDupablePerformIO $ memcmp_ba ba1# ba2# (fromIntegral $ sizeofByteArray ba1) of
-          0 -> True
-          _ -> False
+    | sameByteArray ba1# ba2# = True
+    | n1 /= n2 = False
+    | otherwise = compareByteArrays ba1 ba2 n1 == EQ
+    where
+      n1 = sizeofByteArray ba1
+      n2 = sizeofByteArray ba2
 
+-- | Non-lexicographic ordering. This compares the lengths of
+-- the byte arrays first and uses a lexicographic ordering if
+-- the lengths are equal. Subject to change between major versions.
+-- 
+-- @since 0.6.3.0
 instance Ord ByteArray where
   ba1@(ByteArray ba1#) `compare` ba2@(ByteArray ba2#)
     | sameByteArray ba1# ba2# = EQ
-    | n1 /= n2                = n1 `compare` n2
-    | otherwise =
-        case unsafeDupablePerformIO $ memcmp_ba ba1# ba2# (fromIntegral n1) of
-          x | x >  0 -> GT
-            | x == 0 -> EQ
-            | otherwise -> LT
+    | n1 /= n2 = n1 `compare` n2
+    | otherwise = compareByteArrays ba1 ba2 n1
     where
       n1 = sizeofByteArray ba1
       n2 = sizeofByteArray ba2
+-- Note: On GHC 8.4, the primop compareByteArrays# performs a check for pointer
+-- equality as a shortcut, so the check here is actually redundant. However, it
+-- is included here because it is likely better to check for pointer equality
+-- before checking for length equality. Getting the length requires deferencing
+-- the pointers, which could cause accesses to memory that is not in the cache.
+-- By contrast, a pointer equality check is always extremely cheap.
 
+appendByteArray :: ByteArray -> ByteArray -> ByteArray
+appendByteArray a b = runST $ do
+  marr <- newByteArray (sizeofByteArray a + sizeofByteArray b)
+  copyByteArray marr 0 a 0 (sizeofByteArray a)
+  copyByteArray marr (sizeofByteArray a) b 0 (sizeofByteArray b)
+  unsafeFreezeByteArray marr
+
+concatByteArray :: [ByteArray] -> ByteArray
+concatByteArray arrs = runST $ do
+  let len = calcLength arrs 0
+  marr <- newByteArray len
+  pasteByteArrays marr 0 arrs
+  unsafeFreezeByteArray marr
+
+pasteByteArrays :: MutableByteArray s -> Int -> [ByteArray] -> ST s ()
+pasteByteArrays !_ !_ [] = return ()
+pasteByteArrays !marr !ix (x : xs) = do
+  copyByteArray marr ix x 0 (sizeofByteArray x)
+  pasteByteArrays marr (ix + sizeofByteArray x) xs
+
+calcLength :: [ByteArray] -> Int -> Int
+calcLength [] !n = n
+calcLength (x : xs) !n = calcLength xs (sizeofByteArray x + n)
+
+emptyByteArray :: ByteArray
+emptyByteArray = runST (newByteArray 0 >>= unsafeFreezeByteArray)
+
+replicateByteArray :: Int -> ByteArray -> ByteArray
+replicateByteArray n arr = runST $ do
+  marr <- newByteArray (n * sizeofByteArray arr)
+  let go i = if i < n
+        then do
+          copyByteArray marr (i * sizeofByteArray arr) arr 0 (sizeofByteArray arr)
+          go (i + 1)
+        else return ()
+  go 0
+  unsafeFreezeByteArray marr
+
+#if MIN_VERSION_base(4,9,0)
+instance SG.Semigroup ByteArray where
+  (<>) = appendByteArray
+  sconcat = mconcat . F.toList
+  stimes i arr
+    | itgr < 1 = emptyByteArray
+    | itgr <= (fromIntegral (maxBound :: Int)) = replicateByteArray (fromIntegral itgr) arr
+    | otherwise = error "Data.Primitive.ByteArray#stimes: cannot allocate the requested amount of memory"
+    where itgr = toInteger i :: Integer
+#endif
+
+instance Monoid ByteArray where
+  mempty = emptyByteArray
+#if !(MIN_VERSION_base(4,11,0))
+  mappend = appendByteArray
+#endif
+  mconcat = concatByteArray
+
 #if __GLASGOW_HASKELL__ >= 708
+-- | @since 0.6.3.0
 instance Exts.IsList ByteArray where
   type Item ByteArray = Word8
 
   toList = foldrByteArray (:) []
-  fromList xs = fromListN (length xs) xs
-  fromListN = fromListN
+  fromList xs = byteArrayFromListN (length xs) xs
+  fromListN = byteArrayFromListN
 #endif
+
+die :: String -> String -> a
+die fun problem = error $ "Data.Primitive.ByteArray." ++ fun ++ ": " ++ problem
+
diff --git a/Data/Primitive/MVar.hs b/Data/Primitive/MVar.hs
new file mode 100644
--- /dev/null
+++ b/Data/Primitive/MVar.hs
@@ -0,0 +1,155 @@
+{-# LANGUAGE BangPatterns #-}
+{-# LANGUAGE CPP #-}
+{-# LANGUAGE MagicHash #-}
+{-# LANGUAGE UnboxedTuples #-}
+
+-- |
+-- Module      : Data.Primitive.MVar
+-- License     : BSD2
+-- Portability : non-portable
+--
+-- Primitive operations on @MVar@. This module provides a similar interface
+-- to "Control.Concurrent.MVar". However, the functions are generalized to
+-- work in any 'PrimMonad' instead of only working in 'IO'. Note that all
+-- of the functions here are completely deterministic. Users of 'MVar' are
+-- responsible for designing abstractions that guarantee determinism in
+-- the presence of multi-threading.
+--
+-- @since 0.6.4.0
+module Data.Primitive.MVar
+  ( MVar(..)
+  , newMVar
+  , isEmptyMVar
+  , newEmptyMVar
+  , putMVar
+  , readMVar
+  , takeMVar
+  , tryPutMVar
+  , tryReadMVar
+  , tryTakeMVar
+  ) where
+
+import Control.Monad.Primitive
+import Data.Primitive.Internal.Compat (isTrue#)
+import GHC.Exts (MVar#,newMVar#,takeMVar#,sameMVar#,putMVar#,tryTakeMVar#,
+  isEmptyMVar#,tryPutMVar#,(/=#))
+
+#if __GLASGOW_HASKELL__ >= 708
+import GHC.Exts (readMVar#,tryReadMVar#)
+#endif
+
+data MVar s a = MVar (MVar# s a)
+
+instance Eq (MVar s a) where
+  MVar mvar1# == MVar mvar2# = isTrue# (sameMVar# mvar1# mvar2#)
+
+-- | Create a new 'MVar' that is initially empty.
+newEmptyMVar :: PrimMonad m => m (MVar (PrimState m) a)
+newEmptyMVar = primitive $ \ s# ->
+  case newMVar# s# of
+    (# s2#, svar# #) -> (# s2#, MVar svar# #)
+
+
+-- | Create a new 'MVar' that holds the supplied argument.
+newMVar :: PrimMonad m => a -> m (MVar (PrimState m) a)
+newMVar value =
+  newEmptyMVar >>= \ mvar ->
+  putMVar mvar value >>
+  return mvar
+
+-- | Return the contents of the 'MVar'.  If the 'MVar' is currently
+-- empty, 'takeMVar' will wait until it is full.  After a 'takeMVar',
+-- the 'MVar' is left empty.
+takeMVar :: PrimMonad m => MVar (PrimState m) a -> m a
+takeMVar (MVar mvar#) = primitive $ \ s# -> takeMVar# mvar# s#
+
+-- | Atomically read the contents of an 'MVar'.  If the 'MVar' is
+-- currently empty, 'readMVar' will wait until it is full.
+-- 'readMVar' is guaranteed to receive the next 'putMVar'.
+--
+-- /Multiple Wakeup:/ 'readMVar' is multiple-wakeup, so when multiple readers
+-- are blocked on an 'MVar', all of them are woken up at the same time.
+--
+-- /Compatibility note:/ On GHCs prior to 7.8, 'readMVar' is a combination
+-- of 'takeMVar' and 'putMVar'. Consequently, its behavior differs in the
+-- following ways:
+--
+-- * It is single-wakeup instead of multiple-wakeup.
+-- * It might not receive the value from the next call to 'putMVar' if
+--   there is already a pending thread blocked on 'takeMVar'.
+-- * If another thread puts a value in the 'MVar' in between the
+--   calls to 'takeMVar' and 'putMVar', that value may be overridden.
+readMVar :: PrimMonad m => MVar (PrimState m) a -> m a
+#if __GLASGOW_HASKELL__ >= 708
+readMVar (MVar mvar#) = primitive $ \ s# -> readMVar# mvar# s#
+#else
+readMVar mv = do
+  a <- takeMVar mv
+  putMVar mv a
+  return a
+#endif
+
+-- |Put a value into an 'MVar'.  If the 'MVar' is currently full,
+-- 'putMVar' will wait until it becomes empty.
+putMVar :: PrimMonad m => MVar (PrimState m) a -> a -> m ()
+putMVar (MVar mvar#) x = primitive_ (putMVar# mvar# x)
+
+-- |A non-blocking version of 'takeMVar'.  The 'tryTakeMVar' function
+-- returns immediately, with 'Nothing' if the 'MVar' was empty, or
+-- @'Just' a@ if the 'MVar' was full with contents @a@.  After 'tryTakeMVar',
+-- the 'MVar' is left empty.
+tryTakeMVar :: PrimMonad m => MVar (PrimState m) a -> m (Maybe a)
+tryTakeMVar (MVar m) = primitive $ \ s ->
+  case tryTakeMVar# m s of
+    (# s', 0#, _ #) -> (# s', Nothing #) -- MVar is empty
+    (# s', _,  a #) -> (# s', Just a  #) -- MVar is full
+
+
+-- |A non-blocking version of 'putMVar'.  The 'tryPutMVar' function
+-- attempts to put the value @a@ into the 'MVar', returning 'True' if
+-- it was successful, or 'False' otherwise.
+tryPutMVar :: PrimMonad m => MVar (PrimState m) a -> a -> m Bool
+tryPutMVar (MVar mvar#) x = primitive $ \ s# ->
+    case tryPutMVar# mvar# x s# of
+        (# s, 0# #) -> (# s, False #)
+        (# s, _  #) -> (# s, True #)
+
+-- | A non-blocking version of 'readMVar'.  The 'tryReadMVar' function
+-- returns immediately, with 'Nothing' if the 'MVar' was empty, or
+-- @'Just' a@ if the 'MVar' was full with contents @a@.
+--
+-- /Compatibility note:/ On GHCs prior to 7.8, 'tryReadMVar' is a combination
+-- of 'tryTakeMVar' and 'putMVar'. Consequently, its behavior differs in the
+-- following ways:
+--
+-- * It is single-wakeup instead of multiple-wakeup.
+-- * In the presence of other threads calling 'putMVar', 'tryReadMVar'
+--   may block.
+-- * If another thread puts a value in the 'MVar' in between the
+--   calls to 'tryTakeMVar' and 'putMVar', that value may be overridden.
+tryReadMVar :: PrimMonad m => MVar (PrimState m) a -> m (Maybe a)
+#if __GLASGOW_HASKELL__ >= 708
+tryReadMVar (MVar m) = primitive $ \ s ->
+    case tryReadMVar# m s of
+        (# s', 0#, _ #) -> (# s', Nothing #)      -- MVar is empty
+        (# s', _,  a #) -> (# s', Just a  #)      -- MVar is full
+#else
+tryReadMVar mv = do
+  ma <- tryTakeMVar mv
+  case ma of
+    Just a -> do
+      putMVar mv a
+      return (Just a)
+    Nothing -> return Nothing
+#endif
+
+-- | Check whether a given 'MVar' is empty.
+--
+-- Notice that the boolean value returned  is just a snapshot of
+-- the state of the MVar. By the time you get to react on its result,
+-- the MVar may have been filled (or emptied) - so be extremely
+-- careful when using this operation.   Use 'tryTakeMVar' instead if possible.
+isEmptyMVar :: PrimMonad m => MVar (PrimState m) a -> m Bool
+isEmptyMVar (MVar mv#) = primitive $ \ s# ->
+  case isEmptyMVar# mv# s# of
+    (# s2#, flg #) -> (# s2#, isTrue# (flg /=# 0#) #)
diff --git a/Data/Primitive/PrimArray.hs b/Data/Primitive/PrimArray.hs
new file mode 100644
--- /dev/null
+++ b/Data/Primitive/PrimArray.hs
@@ -0,0 +1,969 @@
+{-# LANGUAGE BangPatterns #-}
+{-# LANGUAGE CPP #-}
+{-# LANGUAGE MagicHash #-}
+{-# LANGUAGE RankNTypes #-}
+{-# LANGUAGE ScopedTypeVariables #-}
+{-# LANGUAGE TypeFamilies #-}
+{-# LANGUAGE UnboxedTuples #-}
+
+{-# OPTIONS_GHC -Wall #-}
+
+-- |
+-- Module      : Data.Primitive.PrimArray
+-- Copyright   : (c) Roman Leshchinskiy 2009-2012
+-- License     : BSD-style
+--
+-- Maintainer  : Roman Leshchinskiy <rl@cse.unsw.edu.au>
+-- Portability : non-portable
+--
+-- Arrays of unboxed primitive types. The function provided by this module
+-- match the behavior of those provided by @Data.Primitive.ByteArray@, and
+-- the underlying types and primops that back them are the same.
+-- However, the type constructors 'PrimArray' and 'MutablePrimArray' take one additional
+-- argument than their respective counterparts 'ByteArray' and 'MutableByteArray'.
+-- This argument is used to designate the type of element in the array.
+-- Consequently, all function this modules accepts length and incides in
+-- terms of elements, not bytes.
+--
+-- @since 0.6.4.0
+module Data.Primitive.PrimArray
+  ( -- * Types
+    PrimArray(..)
+  , MutablePrimArray(..)
+    -- * Allocation
+  , newPrimArray
+  , resizeMutablePrimArray
+#if __GLASGOW_HASKELL__ >= 710
+  , shrinkMutablePrimArray
+#endif
+    -- * Element Access
+  , readPrimArray
+  , writePrimArray
+  , indexPrimArray
+    -- * Freezing and Thawing
+  , unsafeFreezePrimArray
+  , unsafeThawPrimArray
+    -- * Block Operations
+  , copyPrimArray
+  , copyMutablePrimArray
+#if __GLASGOW_HASKELL__ >= 708
+  , copyPrimArrayToPtr
+  , copyMutablePrimArrayToPtr
+#endif
+  , setPrimArray
+    -- * Information
+  , sameMutablePrimArray
+  , getSizeofMutablePrimArray
+  , sizeofMutablePrimArray
+  , sizeofPrimArray
+    -- * List Conversion
+  , primArrayToList
+  , primArrayFromList
+  , primArrayFromListN
+    -- * Folding
+  , foldrPrimArray
+  , foldrPrimArray'
+  , foldlPrimArray
+  , foldlPrimArray'
+  , foldlPrimArrayM'
+    -- * Effectful Folding
+  , traversePrimArray_
+  , itraversePrimArray_
+    -- * Map/Create
+  , mapPrimArray
+  , imapPrimArray
+  , generatePrimArray
+  , replicatePrimArray
+  , filterPrimArray
+  , mapMaybePrimArray
+    -- * Effectful Map/Create
+    -- $effectfulMapCreate
+    -- ** Lazy Applicative
+  , traversePrimArray
+  , itraversePrimArray
+  , generatePrimArrayA
+  , replicatePrimArrayA
+  , filterPrimArrayA
+  , mapMaybePrimArrayA
+    -- ** Strict Primitive Monadic
+  , traversePrimArrayP
+  , itraversePrimArrayP
+  , generatePrimArrayP
+  , replicatePrimArrayP
+  , filterPrimArrayP
+  , mapMaybePrimArrayP
+  ) where
+
+import GHC.Prim
+import GHC.Base ( Int(..) )
+import GHC.Exts (build)
+import GHC.Ptr
+import Data.Primitive.Internal.Compat (isTrue#)
+import Data.Primitive.Types
+import Data.Primitive.ByteArray (ByteArray(..))
+import Data.Monoid (Monoid(..),(<>))
+import Control.Applicative
+import Control.Monad.Primitive
+import Control.Monad.ST
+import qualified Data.List as L
+import qualified Data.Primitive.ByteArray as PB
+import qualified Data.Primitive.Types as PT
+
+#if MIN_VERSION_base(4,7,0)
+import GHC.Exts (IsList(..))
+#endif
+
+#if MIN_VERSION_base(4,9,0)
+import Data.Semigroup (Semigroup)
+import qualified Data.Semigroup as SG
+#endif
+
+-- | Arrays of unboxed elements. This accepts types like 'Double', 'Char',
+-- 'Int', and 'Word', as well as their fixed-length variants ('Word8',
+-- 'Word16', etc.). Since the elements are unboxed, a 'PrimArray' is strict
+-- in its elements. This differs from the behavior of 'Array', which is lazy
+-- in its elements.
+data PrimArray a = PrimArray ByteArray#
+
+-- | Mutable primitive arrays associated with a primitive state token.
+-- These can be written to and read from in a monadic context that supports
+-- sequencing such as 'IO' or 'ST'. Typically, a mutable primitive array will
+-- be built and then convert to an immutable primitive array using
+-- 'unsafeFreezePrimArray'. However, it is also acceptable to simply discard
+-- a mutable primitive array since it lives in managed memory and will be
+-- garbage collected when no longer referenced.
+data MutablePrimArray s a = MutablePrimArray (MutableByteArray# s)
+
+sameByteArray :: ByteArray# -> ByteArray# -> Bool
+sameByteArray ba1 ba2 =
+    case reallyUnsafePtrEquality# (unsafeCoerce# ba1 :: ()) (unsafeCoerce# ba2 :: ()) of
+#if __GLASGOW_HASKELL__ >= 708
+      r -> isTrue# r
+#else
+      1# -> True
+      _ -> False
+#endif
+
+-- | @since 0.6.4.0
+instance (Eq a, Prim a) => Eq (PrimArray a) where
+  a1@(PrimArray ba1#) == a2@(PrimArray ba2#)
+    | sameByteArray ba1# ba2# = True
+    | sz1 /= sz2 = False
+    | otherwise = loop (quot sz1 (sizeOf (undefined :: a)) - 1)
+    where
+    -- Here, we take the size in bytes, not in elements. We do this
+    -- since it allows us to defer performing the division to
+    -- calculate the size in elements.
+    sz1 = PB.sizeofByteArray (ByteArray ba1#)
+    sz2 = PB.sizeofByteArray (ByteArray ba2#)
+    loop !i
+      | i < 0 = True
+      | otherwise = indexPrimArray a1 i == indexPrimArray a2 i && loop (i-1)
+
+-- | Lexicographic ordering. Subject to change between major versions.
+-- 
+--   @since 0.6.4.0
+instance (Ord a, Prim a) => Ord (PrimArray a) where
+  compare a1@(PrimArray ba1#) a2@(PrimArray ba2#)
+    | sameByteArray ba1# ba2# = EQ
+    | otherwise = loop 0
+    where
+    sz1 = PB.sizeofByteArray (ByteArray ba1#)
+    sz2 = PB.sizeofByteArray (ByteArray ba2#)
+    sz = quot (min sz1 sz2) (sizeOf (undefined :: a))
+    loop !i
+      | i < sz = compare (indexPrimArray a1 i) (indexPrimArray a2 i) <> loop (i+1)
+      | otherwise = compare sz1 sz2
+
+#if MIN_VERSION_base(4,7,0)
+-- | @since 0.6.4.0
+instance Prim a => IsList (PrimArray a) where
+  type Item (PrimArray a) = a
+  fromList = primArrayFromList
+  fromListN = primArrayFromListN
+  toList = primArrayToList
+#endif
+
+-- | @since 0.6.4.0
+instance (Show a, Prim a) => Show (PrimArray a) where
+  showsPrec p a = showParen (p > 10) $
+    showString "fromListN " . shows (sizeofPrimArray a) . showString " "
+      . shows (primArrayToList a)
+
+die :: String -> String -> a
+die fun problem = error $ "Data.Primitive.PrimArray." ++ fun ++ ": " ++ problem
+
+primArrayFromList :: Prim a => [a] -> PrimArray a
+primArrayFromList vs = primArrayFromListN (L.length vs) vs
+
+primArrayFromListN :: forall a. Prim a => Int -> [a] -> PrimArray a
+primArrayFromListN len vs = runST run where
+  run :: forall s. ST s (PrimArray a)
+  run = do
+    arr <- newPrimArray len
+    let go :: [a] -> Int -> ST s ()
+        go [] !ix = if ix == len
+          then return ()
+          else die "fromListN" "list length less than specified size"
+        go (a : as) !ix = if ix < len
+          then do
+            writePrimArray arr ix a
+            go as (ix + 1)
+          else die "fromListN" "list length greater than specified size"
+    go vs 0
+    unsafeFreezePrimArray arr
+
+-- | Convert the primitive array to a list.
+{-# INLINE primArrayToList #-}
+primArrayToList :: forall a. Prim a => PrimArray a -> [a]
+primArrayToList xs = build (\c n -> foldrPrimArray c n xs)
+
+primArrayToByteArray :: PrimArray a -> PB.ByteArray
+primArrayToByteArray (PrimArray x) = PB.ByteArray x
+
+byteArrayToPrimArray :: ByteArray -> PrimArray a
+byteArrayToPrimArray (PB.ByteArray x) = PrimArray x
+
+#if MIN_VERSION_base(4,9,0)
+-- | @since 0.6.4.0
+instance Semigroup (PrimArray a) where
+  x <> y = byteArrayToPrimArray (primArrayToByteArray x SG.<> primArrayToByteArray y)
+  sconcat = byteArrayToPrimArray . SG.sconcat . fmap primArrayToByteArray
+  stimes i arr = byteArrayToPrimArray (SG.stimes i (primArrayToByteArray arr))
+#endif
+
+-- | @since 0.6.4.0
+instance Monoid (PrimArray a) where
+  mempty = emptyPrimArray
+#if !(MIN_VERSION_base(4,11,0))
+  mappend x y = byteArrayToPrimArray (mappend (primArrayToByteArray x) (primArrayToByteArray y))
+#endif
+  mconcat = byteArrayToPrimArray . mconcat . map primArrayToByteArray
+
+-- | The empty primitive array.
+emptyPrimArray :: PrimArray a
+{-# NOINLINE emptyPrimArray #-}
+emptyPrimArray = runST $ primitive $ \s0# -> case newByteArray# 0# s0# of
+  (# s1#, arr# #) -> case unsafeFreezeByteArray# arr# s1# of
+    (# s2#, arr'# #) -> (# s2#, PrimArray arr'# #)
+
+-- | Create a new mutable primitive array of the given length. The
+-- underlying memory is left uninitialized.
+newPrimArray :: forall m a. (PrimMonad m, Prim a) => Int -> m (MutablePrimArray (PrimState m) a)
+{-# INLINE newPrimArray #-}
+newPrimArray (I# n#)
+  = primitive (\s# -> 
+      case newByteArray# (n# *# sizeOf# (undefined :: a)) s# of
+        (# s'#, arr# #) -> (# s'#, MutablePrimArray arr# #)
+    )
+
+-- | Resize a mutable primitive array. The new size is given in elements.
+--
+-- This will either resize the array in-place or, if not possible, allocate the
+-- contents into a new, unpinned array and copy the original array\'s contents.
+--
+-- To avoid undefined behaviour, the original 'MutablePrimArray' shall not be
+-- accessed anymore after a 'resizeMutablePrimArray' has been performed.
+-- Moreover, no reference to the old one should be kept in order to allow
+-- garbage collection of the original 'MutablePrimArray' in case a new
+-- 'MutablePrimArray' had to be allocated.
+resizeMutablePrimArray :: forall m a. (PrimMonad m, Prim a)
+  => MutablePrimArray (PrimState m) a
+  -> Int -- ^ new size
+  -> m (MutablePrimArray (PrimState m) a)
+{-# INLINE resizeMutablePrimArray #-}
+#if __GLASGOW_HASKELL__ >= 710
+resizeMutablePrimArray (MutablePrimArray arr#) (I# n#)
+  = primitive (\s# -> case resizeMutableByteArray# arr# (n# *# sizeOf# (undefined :: a)) s# of
+                        (# s'#, arr'# #) -> (# s'#, MutablePrimArray arr'# #))
+#else
+resizeMutablePrimArray arr n
+  = do arr' <- newPrimArray n
+       copyMutablePrimArray arr' 0 arr 0 (min (sizeofMutablePrimArray arr) n)
+       return arr'
+#endif
+
+-- Although it is possible to shim resizeMutableByteArray for old GHCs, this
+-- is not the case with shrinkMutablePrimArray.
+#if __GLASGOW_HASKELL__ >= 710
+-- | Shrink a mutable primitive array. The new size is given in elements.
+-- It must be smaller than the old size. The array will be resized in place.
+-- This function is only available when compiling with GHC 7.10 or newer.
+shrinkMutablePrimArray :: forall m a. (PrimMonad m, Prim a)
+  => MutablePrimArray (PrimState m) a
+  -> Int -- ^ new size
+  -> m ()
+{-# INLINE shrinkMutablePrimArray #-}
+shrinkMutablePrimArray (MutablePrimArray arr#) (I# n#)
+  = primitive_ (shrinkMutableByteArray# arr# (n# *# sizeOf# (undefined :: a)))
+#endif
+
+readPrimArray :: (Prim a, PrimMonad m) => MutablePrimArray (PrimState m) a -> Int -> m a
+{-# INLINE readPrimArray #-}
+readPrimArray (MutablePrimArray arr#) (I# i#)
+  = primitive (readByteArray# arr# i#)
+
+-- | Write an element to the given index.
+writePrimArray ::
+     (Prim a, PrimMonad m)
+  => MutablePrimArray (PrimState m) a -- ^ array
+  -> Int -- ^ index
+  -> a -- ^ element
+  -> m ()
+{-# INLINE writePrimArray #-}
+writePrimArray (MutablePrimArray arr#) (I# i#) x
+  = primitive_ (writeByteArray# arr# i# x)
+
+-- | Copy part of a mutable array into another mutable array.
+--   In the case that the destination and
+--   source arrays are the same, the regions may overlap.
+copyMutablePrimArray :: forall m a.
+     (PrimMonad m, Prim a)
+  => MutablePrimArray (PrimState m) a -- ^ destination array
+  -> Int -- ^ offset into destination array
+  -> MutablePrimArray (PrimState m) a -- ^ source array
+  -> Int -- ^ offset into source array
+  -> Int -- ^ number of elements to copy
+  -> m ()
+{-# INLINE copyMutablePrimArray #-}
+copyMutablePrimArray (MutablePrimArray dst#) (I# doff#) (MutablePrimArray src#) (I# soff#) (I# n#)
+  = primitive_ (copyMutableByteArray#
+      src# 
+      (soff# *# (sizeOf# (undefined :: a)))
+      dst#
+      (doff# *# (sizeOf# (undefined :: a)))
+      (n# *# (sizeOf# (undefined :: a)))
+    )
+
+-- | Copy part of an array into another mutable array.
+copyPrimArray :: forall m a.
+     (PrimMonad m, Prim a)
+  => MutablePrimArray (PrimState m) a -- ^ destination array
+  -> Int -- ^ offset into destination array
+  -> PrimArray a -- ^ source array
+  -> Int -- ^ offset into source array
+  -> Int -- ^ number of elements to copy
+  -> m ()
+{-# INLINE copyPrimArray #-}
+copyPrimArray (MutablePrimArray dst#) (I# doff#) (PrimArray src#) (I# soff#) (I# n#)
+  = primitive_ (copyByteArray#
+      src# 
+      (soff# *# (sizeOf# (undefined :: a)))
+      dst#
+      (doff# *# (sizeOf# (undefined :: a)))
+      (n# *# (sizeOf# (undefined :: a)))
+    )
+
+#if __GLASGOW_HASKELL__ >= 708
+-- | Copy a slice of an immutable primitive array to an address.
+--   The offset and length are given in elements of type @a@.
+--   This function assumes that the 'Prim' instance of @a@
+--   agrees with the 'Storable' instance. This function is only
+--   available when building with GHC 7.8 or newer.
+copyPrimArrayToPtr :: forall m a. (PrimMonad m, Prim a)
+  => Ptr a -- ^ destination pointer
+  -> PrimArray a -- ^ source array
+  -> Int -- ^ offset into source array
+  -> Int -- ^ number of prims to copy
+  -> m ()
+{-# INLINE copyPrimArrayToPtr #-}
+copyPrimArrayToPtr (Ptr addr#) (PrimArray ba#) (I# soff#) (I# n#) =
+    primitive (\ s# ->
+        let s'# = copyByteArrayToAddr# ba# (soff# *# siz#) addr# (n# *# siz#) s#
+        in (# s'#, () #))
+  where siz# = sizeOf# (undefined :: a)
+
+-- | Copy a slice of an immutable primitive array to an address.
+--   The offset and length are given in elements of type @a@.
+--   This function assumes that the 'Prim' instance of @a@
+--   agrees with the 'Storable' instance. This function is only
+--   available when building with GHC 7.8 or newer.
+copyMutablePrimArrayToPtr :: forall m a. (PrimMonad m, Prim a)
+  => Ptr a -- ^ destination pointer
+  -> MutablePrimArray (PrimState m) a -- ^ source array
+  -> Int -- ^ offset into source array
+  -> Int -- ^ number of prims to copy
+  -> m ()
+{-# INLINE copyMutablePrimArrayToPtr #-}
+copyMutablePrimArrayToPtr (Ptr addr#) (MutablePrimArray mba#) (I# soff#) (I# n#) =
+    primitive (\ s# ->
+        let s'# = copyMutableByteArrayToAddr# mba# (soff# *# siz#) addr# (n# *# siz#) s#
+        in (# s'#, () #))
+  where siz# = sizeOf# (undefined :: a)
+#endif
+
+-- | Fill a slice of a mutable primitive array with a value.
+setPrimArray
+  :: (Prim a, PrimMonad m)
+  => MutablePrimArray (PrimState m) a -- ^ array to fill
+  -> Int -- ^ offset into array
+  -> Int -- ^ number of values to fill
+  -> a -- ^ value to fill with
+  -> m ()
+{-# INLINE setPrimArray #-}
+setPrimArray (MutablePrimArray dst#) (I# doff#) (I# sz#) x
+  = primitive_ (PT.setByteArray# dst# doff# sz# x)
+
+-- | Get the size of a mutable primitive array in elements. Unlike 'sizeofMutablePrimArray',
+-- this function ensures sequencing in the presence of resizing.
+getSizeofMutablePrimArray :: forall m a. (PrimMonad m, Prim a)
+  => MutablePrimArray (PrimState m) a -- ^ array
+  -> m Int
+{-# INLINE getSizeofMutablePrimArray #-}
+#if __GLASGOW_HASKELL__ >= 801
+getSizeofMutablePrimArray (MutablePrimArray arr#)
+  = primitive (\s# -> 
+      case getSizeofMutableByteArray# arr# s# of
+        (# s'#, sz# #) -> (# s'#, I# (quotInt# sz# (sizeOf# (undefined :: a))) #)
+    )
+#else
+-- On older GHCs, it is not possible to resize a byte array, so
+-- this provides behavior consistent with the implementation for
+-- newer GHCs.
+getSizeofMutablePrimArray arr
+  = return (sizeofMutablePrimArray arr)
+#endif
+
+-- | Size of the mutable primitive array in elements. This function shall not
+--   be used on primitive arrays that are an argument to or a result of
+--   'resizeMutablePrimArray' or 'shrinkMutablePrimArray'.
+sizeofMutablePrimArray :: forall s a. Prim a => MutablePrimArray s a -> Int
+{-# INLINE sizeofMutablePrimArray #-}
+sizeofMutablePrimArray (MutablePrimArray arr#) =
+  I# (quotInt# (sizeofMutableByteArray# arr#) (sizeOf# (undefined :: a)))
+
+-- | Check if the two arrays refer to the same memory block.
+sameMutablePrimArray :: MutablePrimArray s a -> MutablePrimArray s a -> Bool
+{-# INLINE sameMutablePrimArray #-}
+sameMutablePrimArray (MutablePrimArray arr#) (MutablePrimArray brr#)
+  = isTrue# (sameMutableByteArray# arr# brr#)
+
+-- | Convert a mutable byte array to an immutable one without copying. The
+-- array should not be modified after the conversion.
+unsafeFreezePrimArray
+  :: PrimMonad m => MutablePrimArray (PrimState m) a -> m (PrimArray a)
+{-# INLINE unsafeFreezePrimArray #-}
+unsafeFreezePrimArray (MutablePrimArray arr#)
+  = primitive (\s# -> case unsafeFreezeByteArray# arr# s# of
+                        (# s'#, arr'# #) -> (# s'#, PrimArray arr'# #))
+
+-- | Convert an immutable array to a mutable one without copying. The
+-- original array should not be used after the conversion.
+unsafeThawPrimArray
+  :: PrimMonad m => PrimArray a -> m (MutablePrimArray (PrimState m) a)
+{-# INLINE unsafeThawPrimArray #-}
+unsafeThawPrimArray (PrimArray arr#)
+  = primitive (\s# -> (# s#, MutablePrimArray (unsafeCoerce# arr#) #))
+
+-- | Read a primitive value from the primitive array.
+indexPrimArray :: forall a. Prim a => PrimArray a -> Int -> a
+{-# INLINE indexPrimArray #-}
+indexPrimArray (PrimArray arr#) (I# i#) = indexByteArray# arr# i#
+
+-- | Get the size, in elements, of the primitive array.
+sizeofPrimArray :: forall a. Prim a => PrimArray a -> Int
+{-# INLINE sizeofPrimArray #-}
+sizeofPrimArray (PrimArray arr#) = I# (quotInt# (sizeofByteArray# arr#) (sizeOf# (undefined :: a)))
+
+-- | Lazy right-associated fold over the elements of a 'PrimArray'.
+{-# INLINE foldrPrimArray #-}
+foldrPrimArray :: forall a b. Prim a => (a -> b -> b) -> b -> PrimArray a -> b
+foldrPrimArray f z arr = go 0
+  where
+    !sz = sizeofPrimArray arr
+    go !i
+      | sz > i = f (indexPrimArray arr i) (go (i+1))
+      | otherwise = z
+
+-- | Strict right-associated fold over the elements of a 'PrimArray'.
+{-# INLINE foldrPrimArray' #-}
+foldrPrimArray' :: forall a b. Prim a => (a -> b -> b) -> b -> PrimArray a -> b
+foldrPrimArray' f z0 arr = go (sizeofPrimArray arr - 1) z0
+  where
+    go !i !acc
+      | i < 0 = acc
+      | otherwise = go (i - 1) (f (indexPrimArray arr i) acc)
+
+-- | Lazy left-associated fold over the elements of a 'PrimArray'.
+{-# INLINE foldlPrimArray #-}
+foldlPrimArray :: forall a b. Prim a => (b -> a -> b) -> b -> PrimArray a -> b
+foldlPrimArray f z arr = go (sizeofPrimArray arr - 1)
+  where
+    go !i
+      | i < 0 = z
+      | otherwise = f (go (i - 1)) (indexPrimArray arr i)
+
+-- | Strict left-associated fold over the elements of a 'PrimArray'.
+{-# INLINE foldlPrimArray' #-}
+foldlPrimArray' :: forall a b. Prim a => (b -> a -> b) -> b -> PrimArray a -> b
+foldlPrimArray' f z0 arr = go 0 z0
+  where
+    !sz = sizeofPrimArray arr
+    go !i !acc
+      | i < sz = go (i + 1) (f acc (indexPrimArray arr i))
+      | otherwise = acc
+
+-- | Strict left-associated fold over the elements of a 'PrimArray'.
+{-# INLINE foldlPrimArrayM' #-}
+foldlPrimArrayM' :: (Prim a, Monad m) => (b -> a -> m b) -> b -> PrimArray a -> m b
+foldlPrimArrayM' f z0 arr = go 0 z0
+  where
+    !sz = sizeofPrimArray arr
+    go !i !acc1
+      | i < sz = do
+          acc2 <- f acc1 (indexPrimArray arr i)
+          go (i + 1) acc2
+      | otherwise = return acc1
+
+-- | Traverse a primitive array. The traversal forces the resulting values and
+-- writes them to the new primitive array as it performs the monadic effects.
+-- Consequently:
+--
+-- >>> traversePrimArrayP (\x -> print x $> bool x undefined (x == 2)) (fromList [1, 2, 3 :: Int])
+-- 1
+-- 2
+-- *** Exception: Prelude.undefined
+--
+-- In many situations, 'traversePrimArrayP' can replace 'traversePrimArray',
+-- changing the strictness characteristics of the traversal but typically improving
+-- the performance. Consider the following short-circuiting traversal:
+--
+-- > incrPositiveA :: PrimArray Int -> Maybe (PrimArray Int)
+-- > incrPositiveA xs = traversePrimArray (\x -> bool Nothing (Just (x + 1)) (x > 0)) xs
+--
+-- This can be rewritten using 'traversePrimArrayP'. To do this, we must
+-- change the traversal context to @MaybeT (ST s)@, which has a 'PrimMonad'
+-- instance:
+--
+-- > incrPositiveB :: PrimArray Int -> Maybe (PrimArray Int)
+-- > incrPositiveB xs = runST $ runMaybeT $ traversePrimArrayP
+-- >   (\x -> bool (MaybeT (return Nothing)) (MaybeT (return (Just (x + 1)))) (x > 0))
+-- >   xs
+-- 
+-- Benchmarks demonstrate that the second implementation runs 150 times
+-- faster than the first. It also results in fewer allocations.
+{-# INLINE traversePrimArrayP #-}
+traversePrimArrayP :: (PrimMonad m, Prim a, Prim b)
+  => (a -> m b)
+  -> PrimArray a
+  -> m (PrimArray b)
+traversePrimArrayP f arr = do
+  let !sz = sizeofPrimArray arr
+  marr <- newPrimArray sz
+  let go !ix = if ix < sz
+        then do
+          b <- f (indexPrimArray arr ix)
+          writePrimArray marr ix b
+          go (ix + 1)
+        else return ()
+  go 0
+  unsafeFreezePrimArray marr
+
+-- | Filter the primitive array, keeping the elements for which the monadic
+-- predicate evaluates true.
+{-# INLINE filterPrimArrayP #-}
+filterPrimArrayP :: (PrimMonad m, Prim a)
+  => (a -> m Bool)
+  -> PrimArray a
+  -> m (PrimArray a)
+filterPrimArrayP f arr = do
+  let !sz = sizeofPrimArray arr
+  marr <- newPrimArray sz
+  let go !ixSrc !ixDst = if ixSrc < sz
+        then do
+          let a = indexPrimArray arr ixSrc
+          b <- f a
+          if b
+            then do
+              writePrimArray marr ixDst a
+              go (ixSrc + 1) (ixDst + 1)
+            else go (ixSrc + 1) ixDst
+        else return ixDst
+  lenDst <- go 0 0
+  marr' <- resizeMutablePrimArray marr lenDst
+  unsafeFreezePrimArray marr'
+
+-- | Map over the primitive array, keeping the elements for which the monadic
+-- predicate provides a 'Just'.
+{-# INLINE mapMaybePrimArrayP #-}
+mapMaybePrimArrayP :: (PrimMonad m, Prim a, Prim b)
+  => (a -> m (Maybe b))
+  -> PrimArray a
+  -> m (PrimArray b)
+mapMaybePrimArrayP f arr = do
+  let !sz = sizeofPrimArray arr
+  marr <- newPrimArray sz
+  let go !ixSrc !ixDst = if ixSrc < sz
+        then do
+          let a = indexPrimArray arr ixSrc
+          mb <- f a
+          case mb of
+            Just b -> do
+              writePrimArray marr ixDst b
+              go (ixSrc + 1) (ixDst + 1)
+            Nothing -> go (ixSrc + 1) ixDst
+        else return ixDst
+  lenDst <- go 0 0
+  marr' <- resizeMutablePrimArray marr lenDst
+  unsafeFreezePrimArray marr'
+
+-- | Generate a primitive array by evaluating the monadic generator function
+-- at each index.
+{-# INLINE generatePrimArrayP #-}
+generatePrimArrayP :: (PrimMonad m, Prim a)
+  => Int -- ^ length
+  -> (Int -> m a) -- ^ generator
+  -> m (PrimArray a)
+generatePrimArrayP sz f = do
+  marr <- newPrimArray sz
+  let go !ix = if ix < sz
+        then do
+          b <- f ix
+          writePrimArray marr ix b
+          go (ix + 1)
+        else return ()
+  go 0
+  unsafeFreezePrimArray marr
+
+-- | Execute the monadic action the given number of times and store the
+-- results in a primitive array.
+{-# INLINE replicatePrimArrayP #-}
+replicatePrimArrayP :: (PrimMonad m, Prim a)
+  => Int
+  -> m a
+  -> m (PrimArray a)
+replicatePrimArrayP sz f = do
+  marr <- newPrimArray sz
+  let go !ix = if ix < sz
+        then do
+          b <- f
+          writePrimArray marr ix b
+          go (ix + 1)
+        else return ()
+  go 0
+  unsafeFreezePrimArray marr
+
+
+-- | Map over the elements of a primitive array.
+{-# INLINE mapPrimArray #-}
+mapPrimArray :: (Prim a, Prim b)
+  => (a -> b)
+  -> PrimArray a
+  -> PrimArray b
+mapPrimArray f arr = runST $ do
+  let !sz = sizeofPrimArray arr
+  marr <- newPrimArray sz
+  let go !ix = if ix < sz
+        then do
+          let b = f (indexPrimArray arr ix)
+          writePrimArray marr ix b
+          go (ix + 1)
+        else return ()
+  go 0
+  unsafeFreezePrimArray marr
+
+-- | Indexed map over the elements of a primitive array.
+{-# INLINE imapPrimArray #-}
+imapPrimArray :: (Prim a, Prim b)
+  => (Int -> a -> b)
+  -> PrimArray a
+  -> PrimArray b
+imapPrimArray f arr = runST $ do
+  let !sz = sizeofPrimArray arr
+  marr <- newPrimArray sz
+  let go !ix = if ix < sz
+        then do
+          let b = f ix (indexPrimArray arr ix)
+          writePrimArray marr ix b
+          go (ix + 1)
+        else return ()
+  go 0
+  unsafeFreezePrimArray marr
+
+-- | Filter elements of a primitive array according to a predicate.
+{-# INLINE filterPrimArray #-}
+filterPrimArray :: Prim a
+  => (a -> Bool)
+  -> PrimArray a
+  -> PrimArray a
+filterPrimArray p arr = runST $ do
+  let !sz = sizeofPrimArray arr
+  marr <- newPrimArray sz
+  let go !ixSrc !ixDst = if ixSrc < sz
+        then do
+          let !a = indexPrimArray arr ixSrc
+          if p a
+            then do
+              writePrimArray marr ixDst a
+              go (ixSrc + 1) (ixDst + 1)
+            else go (ixSrc + 1) ixDst
+        else return ixDst
+  dstLen <- go 0 0
+  marr' <- resizeMutablePrimArray marr dstLen
+  unsafeFreezePrimArray marr'
+
+-- | Filter the primitive array, keeping the elements for which the monadic
+-- predicate evaluates true.
+filterPrimArrayA ::
+     (Applicative f, Prim a)
+  => (a -> f Bool) -- ^ mapping function
+  -> PrimArray a -- ^ primitive array
+  -> f (PrimArray a)
+filterPrimArrayA f = \ !ary ->
+  let
+    !len = sizeofPrimArray ary
+    go !ixSrc
+      | ixSrc == len = pure $ IxSTA $ \ixDst _ -> return ixDst
+      | otherwise = let x = indexPrimArray ary ixSrc in
+          liftA2
+            (\keep (IxSTA m) -> IxSTA $ \ixDst mary -> if keep
+              then writePrimArray (MutablePrimArray mary) ixDst x >> m (ixDst + 1) mary
+              else m ixDst mary
+            )
+            (f x)
+            (go (ixSrc + 1))
+  in if len == 0
+     then pure emptyPrimArray
+     else runIxSTA len <$> go 0
+
+-- | Map over the primitive array, keeping the elements for which the applicative
+-- predicate provides a 'Just'.
+mapMaybePrimArrayA ::
+     (Applicative f, Prim a, Prim b)
+  => (a -> f (Maybe b)) -- ^ mapping function
+  -> PrimArray a -- ^ primitive array
+  -> f (PrimArray b)
+mapMaybePrimArrayA f = \ !ary ->
+  let
+    !len = sizeofPrimArray ary
+    go !ixSrc
+      | ixSrc == len = pure $ IxSTA $ \ixDst _ -> return ixDst
+      | otherwise = let x = indexPrimArray ary ixSrc in
+          liftA2
+            (\mb (IxSTA m) -> IxSTA $ \ixDst mary -> case mb of
+              Just b -> writePrimArray (MutablePrimArray mary) ixDst b >> m (ixDst + 1) mary
+              Nothing -> m ixDst mary
+            )
+            (f x)
+            (go (ixSrc + 1))
+  in if len == 0
+     then pure emptyPrimArray
+     else runIxSTA len <$> go 0
+
+-- | Map over a primitive array, optionally discarding some elements. This
+--   has the same behavior as @Data.Maybe.mapMaybe@.
+{-# INLINE mapMaybePrimArray #-}
+mapMaybePrimArray :: (Prim a, Prim b)
+  => (a -> Maybe b)
+  -> PrimArray a
+  -> PrimArray b
+mapMaybePrimArray p arr = runST $ do
+  let !sz = sizeofPrimArray arr
+  marr <- newPrimArray sz
+  let go !ixSrc !ixDst = if ixSrc < sz
+        then do
+          let !a = indexPrimArray arr ixSrc
+          case p a of
+            Just b -> do
+              writePrimArray marr ixDst b
+              go (ixSrc + 1) (ixDst + 1)
+            Nothing -> go (ixSrc + 1) ixDst
+        else return ixDst
+  dstLen <- go 0 0
+  marr' <- resizeMutablePrimArray marr dstLen
+  unsafeFreezePrimArray marr'
+
+
+-- | Traverse a primitive array. The traversal performs all of the applicative
+-- effects /before/ forcing the resulting values and writing them to the new
+-- primitive array. Consequently:
+--
+-- >>> traversePrimArray (\x -> print x $> bool x undefined (x == 2)) (fromList [1, 2, 3 :: Int])
+-- 1
+-- 2
+-- 3
+-- *** Exception: Prelude.undefined
+--
+-- The function 'traversePrimArrayP' always outperforms this function, but it
+-- requires a 'PrimAffineMonad' constraint, and it forces the values as
+-- it performs the effects.
+traversePrimArray ::
+     (Applicative f, Prim a, Prim b)
+  => (a -> f b) -- ^ mapping function
+  -> PrimArray a -- ^ primitive array
+  -> f (PrimArray b)
+traversePrimArray f = \ !ary ->
+  let
+    !len = sizeofPrimArray ary
+    go !i
+      | i == len = pure $ STA $ \mary -> unsafeFreezePrimArray (MutablePrimArray mary)
+      | x <- indexPrimArray ary i
+      = liftA2 (\b (STA m) -> STA $ \mary ->
+                  writePrimArray (MutablePrimArray mary) i b >> m mary)
+               (f x) (go (i + 1))
+  in if len == 0
+     then pure emptyPrimArray
+     else runSTA len <$> go 0
+
+-- | Traverse a primitive array with the index of each element.
+itraversePrimArray ::
+     (Applicative f, Prim a, Prim b)
+  => (Int -> a -> f b)
+  -> PrimArray a
+  -> f (PrimArray b)
+itraversePrimArray f = \ !ary ->
+  let
+    !len = sizeofPrimArray ary
+    go !i
+      | i == len = pure $ STA $ \mary -> unsafeFreezePrimArray (MutablePrimArray mary)
+      | x <- indexPrimArray ary i
+      = liftA2 (\b (STA m) -> STA $ \mary ->
+                  writePrimArray (MutablePrimArray mary) i b >> m mary)
+               (f i x) (go (i + 1))
+  in if len == 0
+     then pure emptyPrimArray
+     else runSTA len <$> go 0
+
+-- | Traverse a primitive array with the indices. The traversal forces the
+-- resulting values and writes them to the new primitive array as it performs
+-- the monadic effects.
+{-# INLINE itraversePrimArrayP #-}
+itraversePrimArrayP :: (Prim a, Prim b, PrimMonad m)
+  => (Int -> a -> m b)
+  -> PrimArray a
+  -> m (PrimArray b)
+itraversePrimArrayP f arr = do
+  let !sz = sizeofPrimArray arr
+  marr <- newPrimArray sz
+  let go !ix
+        | ix < sz = do
+            writePrimArray marr ix =<< f ix (indexPrimArray arr ix)
+            go (ix + 1)
+        | otherwise = return ()
+  go 0
+  unsafeFreezePrimArray marr
+
+-- | Generate a primitive array.
+{-# INLINE generatePrimArray #-}
+generatePrimArray :: Prim a
+  => Int -- ^ length
+  -> (Int -> a) -- ^ element from index
+  -> PrimArray a
+generatePrimArray len f = runST $ do
+  marr <- newPrimArray len
+  let go !ix = if ix < len
+        then do
+          writePrimArray marr ix (f ix)
+          go (ix + 1)
+        else return ()
+  go 0
+  unsafeFreezePrimArray marr
+
+-- | Create a primitive array by copying the element the given
+-- number of times.
+{-# INLINE replicatePrimArray #-}
+replicatePrimArray :: Prim a
+  => Int -- ^ length
+  -> a -- ^ element
+  -> PrimArray a
+replicatePrimArray len a = runST $ do
+  marr <- newPrimArray len
+  setPrimArray marr 0 len a
+  unsafeFreezePrimArray marr
+
+-- | Generate a primitive array by evaluating the applicative generator
+-- function at each index.
+{-# INLINE generatePrimArrayA #-}
+generatePrimArrayA ::
+     (Applicative f, Prim a)
+  => Int -- ^ length
+  -> (Int -> f a) -- ^ element from index
+  -> f (PrimArray a)
+generatePrimArrayA len f =
+  let
+    go !i
+      | i == len = pure $ STA $ \mary -> unsafeFreezePrimArray (MutablePrimArray mary)
+      | otherwise
+      = liftA2 (\b (STA m) -> STA $ \mary ->
+                  writePrimArray (MutablePrimArray mary) i b >> m mary)
+               (f i) (go (i + 1))
+  in if len == 0
+     then pure emptyPrimArray
+     else runSTA len <$> go 0
+
+-- | Execute the applicative action the given number of times and store the
+-- results in a vector.
+{-# INLINE replicatePrimArrayA #-}
+replicatePrimArrayA ::
+     (Applicative f, Prim a)
+  => Int -- ^ length
+  -> f a -- ^ applicative element producer
+  -> f (PrimArray a)
+replicatePrimArrayA len f =
+  let
+    go !i
+      | i == len = pure $ STA $ \mary -> unsafeFreezePrimArray (MutablePrimArray mary)
+      | otherwise
+      = liftA2 (\b (STA m) -> STA $ \mary ->
+                  writePrimArray (MutablePrimArray mary) i b >> m mary)
+               f (go (i + 1))
+  in if len == 0
+     then pure emptyPrimArray
+     else runSTA len <$> go 0
+
+-- | Traverse the primitive array, discarding the results. There
+-- is no 'PrimMonad' variant of this function since it would not provide
+-- any performance benefit.
+traversePrimArray_ ::
+     (Applicative f, Prim a)
+  => (a -> f b)
+  -> PrimArray a
+  -> f ()
+traversePrimArray_ f a = go 0 where
+  !sz = sizeofPrimArray a
+  go !ix = if ix < sz
+    then f (indexPrimArray a ix) *> go (ix + 1)
+    else pure ()
+
+-- | Traverse the primitive array with the indices, discarding the results.
+-- There is no 'PrimMonad' variant of this function since it would not
+-- provide any performance benefit.
+itraversePrimArray_ ::
+     (Applicative f, Prim a)
+  => (Int -> a -> f b)
+  -> PrimArray a
+  -> f ()
+itraversePrimArray_ f a = go 0 where
+  !sz = sizeofPrimArray a
+  go !ix = if ix < sz
+    then f ix (indexPrimArray a ix) *> go (ix + 1)
+    else pure ()
+
+newtype IxSTA a = IxSTA {_runIxSTA :: forall s. Int -> MutableByteArray# s -> ST s Int}
+
+runIxSTA :: forall a. Prim a
+  => Int -- maximum possible size
+  -> IxSTA a
+  -> PrimArray a
+runIxSTA !szUpper = \ (IxSTA m) -> runST $ do
+  ar :: MutablePrimArray s a <- newPrimArray szUpper
+  sz <- m 0 (unMutablePrimArray ar)
+  ar' <- resizeMutablePrimArray ar sz
+  unsafeFreezePrimArray ar'
+{-# INLINE runIxSTA #-}
+
+newtype STA a = STA {_runSTA :: forall s. MutableByteArray# s -> ST s (PrimArray a)}
+
+runSTA :: forall a. Prim a => Int -> STA a -> PrimArray a
+runSTA !sz = \ (STA m) -> runST $ newPrimArray sz >>= \ (ar :: MutablePrimArray s a) -> m (unMutablePrimArray ar)
+{-# INLINE runSTA #-}
+
+unMutablePrimArray :: MutablePrimArray s a -> MutableByteArray# s
+unMutablePrimArray (MutablePrimArray m) = m
+
+{- $effectfulMapCreate
+The naming conventions adopted in this section are explained in the
+documentation of the @Data.Primitive@ module.
+-}
+
+
diff --git a/Data/Primitive/Ptr.hs b/Data/Primitive/Ptr.hs
new file mode 100644
--- /dev/null
+++ b/Data/Primitive/Ptr.hs
@@ -0,0 +1,125 @@
+{-# LANGUAGE CPP #-}
+{-# LANGUAGE MagicHash #-}
+{-# LANGUAGE UnboxedTuples #-}
+{-# LANGUAGE ScopedTypeVariables #-}
+
+-- |
+-- Module      : Data.Primitive.Ptr
+-- Copyright   : (c) Roman Leshchinskiy 2009-2012
+-- License     : BSD-style
+--
+-- Maintainer  : Roman Leshchinskiy <rl@cse.unsw.edu.au>
+-- Portability : non-portable
+--
+-- Primitive operations on machine addresses
+--
+-- @since 0.6.4.0
+
+module Data.Primitive.Ptr (
+  -- * Types
+  Ptr(..),
+
+  -- * Address arithmetic
+  nullPtr, advancePtr, subtractPtr,
+
+  -- * Element access
+  indexOffPtr, readOffPtr, writeOffPtr,
+
+  -- * Block operations
+  copyPtr, movePtr, setPtr
+
+#if __GLASGOW_HASKELL__ >= 708
+  , copyPtrToMutablePrimArray
+#endif
+) where
+
+import Control.Monad.Primitive
+import Data.Primitive.Types
+#if __GLASGOW_HASKELL__ >= 708
+import Data.Primitive.PrimArray (MutablePrimArray(..))
+#endif
+
+import GHC.Base ( Int(..) )
+import GHC.Prim
+
+import GHC.Ptr
+import Foreign.Marshal.Utils
+
+
+-- | Offset a pointer by the given number of elements.
+advancePtr :: forall a. Prim a => Ptr a -> Int -> Ptr a
+{-# INLINE advancePtr #-}
+advancePtr (Ptr a#) (I# i#) = Ptr (plusAddr# a# (i# *# sizeOf# (undefined :: a)))
+
+-- | Subtract a pointer from another pointer. The result represents
+--   the number of elements of type @a@ that fit in the contiguous
+--   memory range bounded by these two pointers.
+subtractPtr :: forall a. Prim a => Ptr a -> Ptr a -> Int
+{-# INLINE subtractPtr #-}
+subtractPtr (Ptr a#) (Ptr b#) = I# (quotInt# (minusAddr# a# b#) (sizeOf# (undefined :: a)))
+
+-- | Read a value from a memory position given by a pointer and an offset.
+-- The memory block the address refers to must be immutable. The offset is in
+-- elements of type @a@ rather than in bytes.
+indexOffPtr :: Prim a => Ptr a -> Int -> a
+{-# INLINE indexOffPtr #-}
+indexOffPtr (Ptr addr#) (I# i#) = indexOffAddr# addr# i#
+
+-- | Read a value from a memory position given by an address and an offset.
+-- The offset is in elements of type @a@ rather than in bytes.
+readOffPtr :: (Prim a, PrimMonad m) => Ptr a -> Int -> m a
+{-# INLINE readOffPtr #-}
+readOffPtr (Ptr addr#) (I# i#) = primitive (readOffAddr# addr# i#)
+
+-- | Write a value to a memory position given by an address and an offset.
+-- The offset is in elements of type @a@ rather than in bytes.
+writeOffPtr :: (Prim a, PrimMonad m) => Ptr a -> Int -> a -> m ()
+{-# INLINE writeOffPtr #-}
+writeOffPtr (Ptr addr#) (I# i#) x = primitive_ (writeOffAddr# addr# i# x)
+
+-- | Copy the given number of elements from the second 'Ptr' to the first. The
+-- areas may not overlap.
+copyPtr :: forall m a. (PrimMonad m, Prim a)
+  => Ptr a -- ^ destination pointer
+  -> Ptr a -- ^ source pointer
+  -> Int -- ^ number of elements
+  -> m ()
+{-# INLINE copyPtr #-}
+copyPtr (Ptr dst#) (Ptr src#) n
+  = unsafePrimToPrim $ copyBytes (Ptr dst#) (Ptr src#) (n * sizeOf (undefined :: a))
+
+-- | Copy the given number of elements from the second 'Ptr' to the first. The
+-- areas may overlap.
+movePtr :: forall m a. (PrimMonad m, Prim a)
+  => Ptr a -- ^ destination address
+  -> Ptr a -- ^ source address
+  -> Int -- ^ number of elements
+  -> m ()
+{-# INLINE movePtr #-}
+movePtr (Ptr dst#) (Ptr src#) n
+  = unsafePrimToPrim $ moveBytes (Ptr dst#) (Ptr src#) (n * sizeOf (undefined :: a))
+
+-- | Fill a memory block with the given value. The length is in
+-- elements of type @a@ rather than in bytes.
+setPtr :: (Prim a, PrimMonad m) => Ptr a -> Int -> a -> m ()
+{-# INLINE setPtr #-}
+setPtr (Ptr addr#) (I# n#) x = primitive_ (setOffAddr# addr# 0# n# x)
+
+
+#if __GLASGOW_HASKELL__ >= 708
+-- | Copy from a pointer to a mutable primitive array.
+-- The offset and length are given in elements of type @a@.
+-- This function is only available when building with GHC 7.8
+-- or newer.
+copyPtrToMutablePrimArray :: forall m a. (PrimMonad m, Prim a)
+  => MutablePrimArray (PrimState m) a -- ^ destination array
+  -> Int -- ^ destination offset
+  -> Ptr a -- ^ source pointer
+  -> Int -- ^ number of elements
+  -> m ()
+{-# INLINE copyPtrToMutablePrimArray #-}
+copyPtrToMutablePrimArray (MutablePrimArray ba#) (I# doff#) (Ptr addr#) (I# n#) = 
+  primitive_ (copyAddrToByteArray# addr# ba# (doff# *# siz#) (n# *# siz#))
+  where
+  siz# = sizeOf# (undefined :: a)
+#endif
diff --git a/Data/Primitive/SmallArray.hs b/Data/Primitive/SmallArray.hs
--- a/Data/Primitive/SmallArray.hs
+++ b/Data/Primitive/SmallArray.hs
@@ -6,6 +6,7 @@
 {-# LANGUAGE DeriveTraversable #-}
 {-# LANGUAGE DeriveDataTypeable #-}
 {-# LANGUAGE GeneralizedNewtypeDeriving #-}
+{-# LANGUAGE BangPatterns #-}
 
 -- |
 -- Module : Data.Primitive.SmallArray
@@ -45,14 +46,20 @@
   , copySmallMutableArray
   , indexSmallArray
   , indexSmallArrayM
+  , indexSmallArray##
   , cloneSmallArray
   , cloneSmallMutableArray
   , freezeSmallArray
   , unsafeFreezeSmallArray
   , thawSmallArray
+  , runSmallArray
   , unsafeThawSmallArray
   , sizeofSmallArray
   , sizeofSmallMutableArray
+  , smallArrayFromList
+  , smallArrayFromListN
+  , mapSmallArray'
+  , traverseSmallArrayP
   ) where
 
 
@@ -70,25 +77,34 @@
 import Control.Monad.Fix
 import Control.Monad.Primitive
 import Control.Monad.ST
-#if MIN_VERSION_base(4,4,0)
 import Control.Monad.Zip
-#endif
 import Data.Data
-import Data.Foldable
+import Data.Foldable as Foldable
 import Data.Functor.Identity
+#if !(MIN_VERSION_base(4,10,0))
 import Data.Monoid
+#endif
 #if MIN_VERSION_base(4,9,0)
+import qualified GHC.ST as GHCST
 import qualified Data.Semigroup as Sem
 #endif
-import Text.ParserCombinators.ReadPrec
-import Text.Read
-import Text.Read.Lex
+import Text.ParserCombinators.ReadP
+#if MIN_VERSION_base(4,10,0)
+import GHC.Exts (runRW#)
+#elif MIN_VERSION_base(4,9,0)
+import GHC.Base (runRW#)
+#endif
 
 #if !(HAVE_SMALL_ARRAY)
 import Data.Primitive.Array
 import Data.Traversable
+import qualified Data.Primitive.Array as Array
 #endif
 
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+import Data.Functor.Classes (Eq1(..),Ord1(..),Show1(..),Read1(..))
+#endif
+
 #if HAVE_SMALL_ARRAY
 data SmallArray a = SmallArray (SmallArray# a)
   deriving Typeable
@@ -105,12 +121,16 @@
   , Alternative
   , Monad
   , MonadPlus
-#if MIN_VERSION_base(4,4,0)
   , MonadZip
-#endif
   , MonadFix
   , Monoid
   , Typeable
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+  , Eq1
+  , Ord1
+  , Show1
+  , Read1
+#endif
   )
 
 #if MIN_VERSION_base(4,7,0)
@@ -118,7 +138,7 @@
   type Item (SmallArray a) = a
   fromListN n l = SmallArray (fromListN n l)
   fromList l = SmallArray (fromList l)
-  toList (SmallArray a) = toList a
+  toList a = Foldable.toList a
 #endif
 #endif
 
@@ -223,6 +243,17 @@
 #endif
 {-# INLINE indexSmallArray #-}
 
+-- | Read a value from the immutable array at the given index, returning
+-- the result in an unboxed unary tuple. This is currently used to implement
+-- folds.
+indexSmallArray## :: SmallArray a -> Int -> (# a #)
+#if HAVE_SMALL_ARRAY
+indexSmallArray## (SmallArray ary) (I# i) = indexSmallArray# ary i
+#else
+indexSmallArray## (SmallArray a) = indexArray## a
+#endif
+{-# INLINE indexSmallArray## #-}
+
 -- | Create a copy of a slice of an immutable array.
 cloneSmallArray
   :: SmallArray a -- ^ source
@@ -378,7 +409,106 @@
 #endif
 {-# INLINE sizeofSmallMutableArray #-}
 
+-- | This is the fastest, most straightforward way to traverse
+-- an array, but it only works correctly with a sufficiently
+-- "affine" 'PrimMonad' instance. In particular, it must only produce
+-- *one* result array. 'Control.Monad.Trans.List.ListT'-transformed
+-- monads, for example, will not work right at all.
+traverseSmallArrayP
+  :: PrimMonad m
+  => (a -> m b)
+  -> SmallArray a
+  -> m (SmallArray b)
 #if HAVE_SMALL_ARRAY
+traverseSmallArrayP f = \ !ary ->
+  let
+    !sz = sizeofSmallArray ary
+    go !i !mary
+      | i == sz
+      = unsafeFreezeSmallArray mary
+      | otherwise
+      = do
+          a <- indexSmallArrayM ary i
+          b <- f a
+          writeSmallArray mary i b
+          go (i + 1) mary
+  in do
+    mary <- newSmallArray sz badTraverseValue
+    go 0 mary
+#else
+traverseSmallArrayP f (SmallArray ar) = SmallArray `liftM` traverseArrayP f ar
+#endif
+{-# INLINE traverseSmallArrayP #-}
+
+-- | Strict map over the elements of the array.
+mapSmallArray' :: (a -> b) -> SmallArray a -> SmallArray b
+#if HAVE_SMALL_ARRAY
+mapSmallArray' f sa = createSmallArray (length sa) (die "mapSmallArray'" "impossible") $ \smb ->
+  fix ? 0 $ \go i ->
+    when (i < length sa) $ do
+      x <- indexSmallArrayM sa i
+      let !y = f x
+      writeSmallArray smb i y *> go (i+1)
+#else
+mapSmallArray' f (SmallArray ar) = SmallArray (mapArray' f ar)
+#endif
+{-# INLINE mapSmallArray' #-}
+
+#ifndef HAVE_SMALL_ARRAY
+runSmallArray
+  :: (forall s. ST s (SmallMutableArray s a))
+  -> SmallArray a
+runSmallArray m = SmallArray $ runArray $
+  m >>= \(SmallMutableArray mary) -> return mary
+
+#elif !MIN_VERSION_base(4,9,0)
+runSmallArray
+  :: (forall s. ST s (SmallMutableArray s a))
+  -> SmallArray a
+runSmallArray m = runST $ m >>= unsafeFreezeSmallArray
+
+#else
+-- This low-level business is designed to work with GHC's worker-wrapper
+-- transformation. A lot of the time, we don't actually need an Array
+-- constructor. By putting it on the outside, and being careful about
+-- how we special-case the empty array, we can make GHC smarter about this.
+-- The only downside is that separately created 0-length arrays won't share
+-- their Array constructors, although they'll share their underlying
+-- Array#s.
+runSmallArray
+  :: (forall s. ST s (SmallMutableArray s a))
+  -> SmallArray a
+runSmallArray m = SmallArray (runSmallArray# m)
+
+runSmallArray#
+  :: (forall s. ST s (SmallMutableArray s a))
+  -> SmallArray# a
+runSmallArray# m = case runRW# $ \s ->
+  case unST m s of { (# s', SmallMutableArray mary# #) ->
+  unsafeFreezeSmallArray# mary# s'} of (# _, ary# #) -> ary#
+
+unST :: ST s a -> State# s -> (# State# s, a #)
+unST (GHCST.ST f) = f
+
+#endif
+
+#if HAVE_SMALL_ARRAY
+-- See the comment on runSmallArray for why we use emptySmallArray#.
+createSmallArray
+  :: Int
+  -> a
+  -> (forall s. SmallMutableArray s a -> ST s ())
+  -> SmallArray a
+createSmallArray 0 _ _ = SmallArray (emptySmallArray# (# #))
+createSmallArray n x f = runSmallArray $ do
+  mary <- newSmallArray n x
+  f mary
+  pure mary
+
+emptySmallArray# :: (# #) -> SmallArray# a
+emptySmallArray# _ = case emptySmallArray of SmallArray ar -> ar
+{-# NOINLINE emptySmallArray# #-}
+
 die :: String -> String -> a
 die fun problem = error $ "Data.Primitive.SmallArray." ++ fun ++ ": " ++ problem
 
@@ -388,12 +518,6 @@
             >>= unsafeFreezeSmallArray
 {-# NOINLINE emptySmallArray #-}
 
-createSmallArray
-  :: Int -> a -> (forall s. SmallMutableArray s a -> ST s ()) -> SmallArray a
-createSmallArray 0 _ _ = emptySmallArray
-createSmallArray i x k =
-  runST $ newSmallArray i x >>= \sa -> k sa *> unsafeFreezeSmallArray sa
-{-# INLINE createSmallArray #-}
 
 infixl 1 ?
 (?) :: (a -> b -> c) -> (b -> a -> c)
@@ -403,82 +527,197 @@
 noOp :: a -> ST s ()
 noOp = const $ pure ()
 
+smallArrayLiftEq :: (a -> b -> Bool) -> SmallArray a -> SmallArray b -> Bool
+smallArrayLiftEq p sa1 sa2 = length sa1 == length sa2 && loop (length sa1 - 1)
+  where
+  loop i
+    | i < 0
+    = True
+    | (# x #) <- indexSmallArray## sa1 i
+    , (# y #) <- indexSmallArray## sa2 i
+    = p x y && loop (i-1)
+
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+-- | @since 0.6.4.0
+instance Eq1 SmallArray where
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,5,0)
+  liftEq = smallArrayLiftEq
+#else
+  eq1 = smallArrayLiftEq (==)
+#endif
+#endif
+
 instance Eq a => Eq (SmallArray a) where
-  sa1 == sa2 = length sa1 == length sa2 && loop (length sa1 - 1)
-   where
-   loop i
-     | i < 0     = True
-     | otherwise = indexSmallArray sa1 i == indexSmallArray sa2 i && loop (i-1)
+  sa1 == sa2 = smallArrayLiftEq (==) sa1 sa2
 
 instance Eq (SmallMutableArray s a) where
   SmallMutableArray sma1# == SmallMutableArray sma2# =
     isTrue# (sameSmallMutableArray# sma1# sma2#)
 
+smallArrayLiftCompare :: (a -> b -> Ordering) -> SmallArray a -> SmallArray b -> Ordering
+smallArrayLiftCompare elemCompare a1 a2 = loop 0
+  where
+  mn = length a1 `min` length a2
+  loop i
+    | i < mn
+    , (# x1 #) <- indexSmallArray## a1 i
+    , (# x2 #) <- indexSmallArray## a2 i
+    = elemCompare x1 x2 `mappend` loop (i+1)
+    | otherwise = compare (length a1) (length a2)
+
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+-- | @since 0.6.4.0
+instance Ord1 SmallArray where
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,5,0)
+  liftCompare = smallArrayLiftCompare
+#else
+  compare1 = smallArrayLiftCompare compare
+#endif
+#endif
+
+-- | Lexicographic ordering. Subject to change between major versions.
 instance Ord a => Ord (SmallArray a) where
-  compare sl sr = fix ? 0 $ \go i ->
-    if i < l
-      then compare (indexSmallArray sl i) (indexSmallArray sr i) <> go (i+1)
-      else compare (length sl) (length sr)
-   where l = length sl `min` length sr
+  compare sa1 sa2 = smallArrayLiftCompare compare sa1 sa2
 
 instance Foldable SmallArray where
-  foldr f z sa = fix ? 0 $ \go i ->
-    if i < length sa
-      then f (indexSmallArray sa i) (go $ i+1)
-      else z
+  -- Note: we perform the array lookups eagerly so we won't
+  -- create thunks to perform lookups even if GHC can't see
+  -- that the folding function is strict.
+  foldr f = \z !ary ->
+    let
+      !sz = sizeofSmallArray ary
+      go i
+        | i == sz = z
+        | (# x #) <- indexSmallArray## ary i
+        = f x (go (i+1))
+    in go 0
   {-# INLINE foldr #-}
-
-  foldr' f z sa = fix ? z ? length sa - 1 $ \go acc i ->
-    if i < 0
-      then acc
-      else go (f (indexSmallArray sa i) acc) (i-1)
-  {-# INLINE foldr' #-}
-
-  foldl f z sa = fix ? length sa - 1 $ \go i ->
-    if i < 0
-      then z
-      else f (go $ i-1) $ indexSmallArray sa i
+  foldl f = \z !ary ->
+    let
+      go i
+        | i < 0 = z
+        | (# x #) <- indexSmallArray## ary i
+        = f (go (i-1)) x
+    in go (sizeofSmallArray ary - 1)
   {-# INLINE foldl #-}
-
-  foldl' f z sa = fix ? z ? 0 $ \go acc i ->
-    if i < length sa
-      then go (f acc $ indexSmallArray sa i) (i+1)
-      else acc
-  {-# INLINE foldl' #-}
-
-  foldr1 f sa
-    | sz == 0   = die "foldr1" "empty list"
-    | otherwise = fix ? 0 $ \go i ->
-        if i < sz-1
-          then f (indexSmallArray sa i) (go $ i+1)
-          else indexSmallArray sa $ sz-1
-   where sz = sizeofSmallArray sa
+  foldr1 f = \ !ary ->
+    let
+      !sz = sizeofSmallArray ary - 1
+      go i =
+        case indexSmallArray## ary i of
+          (# x #) | i == sz -> x
+                  | otherwise -> f x (go (i+1))
+    in if sz < 0
+       then die "foldr1" "Empty SmallArray"
+       else go 0
   {-# INLINE foldr1 #-}
-
-  foldl1 f sa
-    | sz == 0   = die "foldl1" "empty list"
-    | otherwise = fix ? sz-1 $ \go i ->
-        if i < 1
-        then indexSmallArray sa 0
-        else f (go $ i-1) (indexSmallArray sa i)
-   where sz = sizeofSmallArray sa
+  foldl1 f = \ !ary ->
+    let
+      !sz = sizeofSmallArray ary - 1
+      go i =
+        case indexSmallArray## ary i of
+          (# x #) | i == 0 -> x
+                  | otherwise -> f (go (i - 1)) x
+    in if sz < 0
+       then die "foldl1" "Empty SmallArray"
+       else go sz
   {-# INLINE foldl1 #-}
-
-  null sa = sizeofSmallArray sa == 0
+  foldr' f = \z !ary ->
+    let
+      go i !acc
+        | i == -1 = acc
+        | (# x #) <- indexSmallArray## ary i
+        = go (i-1) (f x acc)
+    in go (sizeofSmallArray ary - 1) z
+  {-# INLINE foldr' #-}
+  foldl' f = \z !ary ->
+    let
+      !sz = sizeofSmallArray ary
+      go i !acc
+        | i == sz = acc
+        | (# x #) <- indexSmallArray## ary i
+        = go (i+1) (f acc x)
+    in go 0 z
+  {-# INLINE foldl' #-}
+  null a = sizeofSmallArray a == 0
   {-# INLINE null #-}
-
   length = sizeofSmallArray
   {-# INLINE length #-}
+  maximum ary | sz == 0   = die "maximum" "Empty SmallArray"
+              | (# frst #) <- indexSmallArray## ary 0
+              = go 1 frst
+   where
+     sz = sizeofSmallArray ary
+     go i !e
+       | i == sz = e
+       | (# x #) <- indexSmallArray## ary i
+       = go (i+1) (max e x)
+  {-# INLINE maximum #-}
+  minimum ary | sz == 0   = die "minimum" "Empty SmallArray"
+              | (# frst #) <- indexSmallArray## ary 0
+              = go 1 frst
+   where sz = sizeofSmallArray ary
+         go i !e
+           | i == sz = e
+           | (# x #) <- indexSmallArray## ary i
+           = go (i+1) (min e x)
+  {-# INLINE minimum #-}
+  sum = foldl' (+) 0
+  {-# INLINE sum #-}
+  product = foldl' (*) 1
+  {-# INLINE product #-}
 
+newtype STA a = STA {_runSTA :: forall s. SmallMutableArray# s a -> ST s (SmallArray a)}
+
+runSTA :: Int -> STA a -> SmallArray a
+runSTA !sz = \ (STA m) -> runST $ newSmallArray_ sz >>=
+                        \ (SmallMutableArray ar#) -> m ar#
+{-# INLINE runSTA #-}
+
+newSmallArray_ :: Int -> ST s (SmallMutableArray s a)
+newSmallArray_ !n = newSmallArray n badTraverseValue
+
+badTraverseValue :: a
+badTraverseValue = die "traverse" "bad indexing"
+{-# NOINLINE badTraverseValue #-}
+
 instance Traversable SmallArray where
-  traverse f sa = fromListN l <$> traverse (f . indexSmallArray sa) [0..l-1]
-   where l = length sa
+  traverse f = traverseSmallArray f
+  {-# INLINE traverse #-}
 
+traverseSmallArray
+  :: Applicative f
+  => (a -> f b) -> SmallArray a -> f (SmallArray b)
+traverseSmallArray f = \ !ary ->
+  let
+    !len = sizeofSmallArray ary
+    go !i
+      | i == len
+      = pure $ STA $ \mary -> unsafeFreezeSmallArray (SmallMutableArray mary)
+      | (# x #) <- indexSmallArray## ary i
+      = liftA2 (\b (STA m) -> STA $ \mary ->
+                  writeSmallArray (SmallMutableArray mary) i b >> m mary)
+               (f x) (go (i + 1))
+  in if len == 0
+     then pure emptySmallArray
+     else runSTA len <$> go 0
+{-# INLINE [1] traverseSmallArray #-}
+
+{-# RULES
+"traverse/ST" forall (f :: a -> ST s b). traverseSmallArray f = traverseSmallArrayP f
+"traverse/IO" forall (f :: a -> IO b). traverseSmallArray f = traverseSmallArrayP f
+"traverse/Id" forall (f :: a -> Identity b). traverseSmallArray f =
+   (coerce :: (SmallArray a -> SmallArray (Identity b))
+           -> SmallArray a -> Identity (SmallArray b)) (fmap f)
+ #-}
+
+
 instance Functor SmallArray where
   fmap f sa = createSmallArray (length sa) (die "fmap" "impossible") $ \smb ->
     fix ? 0 $ \go i ->
-      when (i < length sa) $
-        writeSmallArray smb i (f $ indexSmallArray sa i) *> go (i+1)
+      when (i < length sa) $ do
+        x <- indexSmallArrayM sa i
+        writeSmallArray smb i (f x) *> go (i+1)
   {-# INLINE fmap #-}
 
   x <$ sa = createSmallArray (length sa) x noOp
@@ -493,26 +732,29 @@
    where
    la = length sa ; lb = length sb
 
-  sa <* sb = createSmallArray (la*lb) (indexSmallArray sa $ la-1) $ \sma ->
-    fix ? 0 $ \outer i -> when (i < la-1) $ do
-      let a = indexSmallArray sa i
-      fix ? 0 $ \inner j ->
-        when (j < lb) $
-          writeSmallArray sma (la*i + j) a *> inner (j+1)
-      outer $ i+1
-   where
-   la = length sa ; lb = length sb
+  a <* b = createSmallArray (sza*szb) (die "<*" "impossible") $ \ma ->
+    let fill off i e = when (i < szb) $
+                         writeSmallArray ma (off+i) e >> fill off (i+1) e
+        go i = when (i < sza) $ do
+                 x <- indexSmallArrayM a i
+                 fill (i*szb) 0 x
+                 go (i+1)
+     in go 0
+   where sza = sizeofSmallArray a ; szb = sizeofSmallArray b
 
-  sf <*> sx = createSmallArray (lf*lx) (die "<*>" "impossible") $ \smb ->
-    fix ? 0 $ \outer i -> when (i < lf) $ do
-      let f = indexSmallArray sf i
-      fix ? 0 $ \inner j ->
-        when (j < lx) $
-          writeSmallArray smb (lf*i + j) (f $ indexSmallArray sx j)
-            *> inner (j+1)
-      outer $ i+1
-   where
-   lf = length sf ; lx = length sx
+  ab <*> a = createSmallArray (szab*sza) (die "<*>" "impossible") $ \mb ->
+    let go1 i = when (i < szab) $
+            do
+              f <- indexSmallArrayM ab i
+              go2 (i*sza) f 0
+              go1 (i+1)
+        go2 off f j = when (j < sza) $
+            do
+              x <- indexSmallArrayM a j
+              writeSmallArray mb (off + j) (f x)
+              go2 off f (j + 1)
+    in go1 0
+   where szab = sizeofSmallArray ab ; sza = sizeofSmallArray a
 
 instance Alternative SmallArray where
   empty = emptySmallArray
@@ -528,20 +770,41 @@
   some sa | null sa   = emptySmallArray
           | otherwise = die "some" "infinite arrays are not well defined"
 
+data ArrayStack a
+  = PushArray !(SmallArray a) !(ArrayStack a)
+  | EmptyStack
+-- TODO: This isn't terribly efficient. It would be better to wrap
+-- ArrayStack with a type like
+--
+-- data NES s a = NES !Int !(SmallMutableArray s a) !(ArrayStack a)
+--
+-- We'd copy incoming arrays into the mutable array until we would
+-- overflow it. Then we'd freeze it, push it on the stack, and continue.
+-- Any sufficiently large incoming arrays would go straight on the stack.
+-- Such a scheme would make the stack much more compact in the case
+-- of many small arrays.
+
 instance Monad SmallArray where
   return = pure
   (>>) = (*>)
 
-  sa >>= f = collect 0 [] (la-1)
+  sa >>= f = collect 0 EmptyStack (la-1)
    where
    la = length sa
    collect sz stk i
      | i < 0 = createSmallArray sz (die ">>=" "impossible") $ fill 0 stk
-     | otherwise = let sb = f $ indexSmallArray sa i in
-         collect (sz + length sb) (sb:stk) (i-1)
+     | (# x #) <- indexSmallArray## sa i
+     , let sb = f x
+           lsb = length sb
+       -- If we don't perform this check, we could end up allocating
+       -- a stack full of empty arrays if someone is filtering most
+       -- things out. So we refrain from pushing empty arrays.
+     = if lsb == 0
+       then collect sz stk (i-1)
+       else collect (sz + lsb) (PushArray sb stk) (i-1)
 
-   fill _   [      ] _   = return ()
-   fill off (sb:sbs) smb =
+   fill _ EmptyStack _ = return ()
+   fill off (PushArray sb sbs) smb =
      copySmallArray smb off sb 0 (length sb)
        *> fill (off + length sb) sbs smb
 
@@ -554,9 +817,11 @@
 zipW :: String -> (a -> b -> c) -> SmallArray a -> SmallArray b -> SmallArray c
 zipW nm = \f sa sb -> let mn = length sa `min` length sb in
   createSmallArray mn (die nm "impossible") $ \mc ->
-    fix ? 0 $ \go i -> when (i < mn) $
-      writeSmallArray mc i (f (indexSmallArray sa i) (indexSmallArray sb i))
-        *> go (i+1)
+    fix ? 0 $ \go i -> when (i < mn) $ do
+      x <- indexSmallArrayM sa i
+      y <- indexSmallArrayM sb i
+      writeSmallArray mc i (f x y)
+      go (i+1)
 {-# INLINE zipW #-}
 
 instance MonadZip SmallArray where
@@ -576,9 +841,17 @@
         <*> unsafeFreezeSmallArray smb
 
 instance MonadFix SmallArray where
-  mfix f = fromList . mfix $ toList . f
+  mfix f = createSmallArray (sizeofSmallArray (f err))
+                            (die "mfix" "impossible") $ flip fix 0 $
+    \r !i !mary -> when (i < sz) $ do
+                      writeSmallArray mary i (fix (\xi -> f xi `indexSmallArray` i))
+                      r (i + 1) mary
+    where
+      sz = sizeofSmallArray (f err)
+      err = error "mfix for Data.Primitive.SmallArray applied to strict function."
 
 #if MIN_VERSION_base(4,9,0)
+-- | @since 0.6.3.0
 instance Sem.Semigroup (SmallArray a) where
   (<>) = (<|>)
   sconcat = mconcat . toList
@@ -589,34 +862,65 @@
 #if !(MIN_VERSION_base(4,11,0))
   mappend = (<|>)
 #endif
-  mconcat sas = createSmallArray n (die "mconcat" "impossible") $ \sma ->
-    fix ? 0 ? sas $ \go off l -> case l of
-      [] -> return ()
-      sa:stk -> copySmallArray sma off sa 0 (length sa) *> go (off+1) stk
-   where n = sum . fmap length $ sas
+  mconcat l = createSmallArray n (die "mconcat" "impossible") $ \ma ->
+    let go !_  [    ] = return ()
+        go off (a:as) =
+          copySmallArray ma off a 0 (sizeofSmallArray a) >> go (off + sizeofSmallArray a) as
+     in go 0 l
+   where n = sum . fmap length $ l
 
 instance IsList (SmallArray a) where
   type Item (SmallArray a) = a
-  fromListN n l =
-    createSmallArray n (die "fromListN" "mismatched size and list") $ \sma ->
-      fix ? 0 ? l $ \go i li -> case li of
-        [] -> pure ()
-        x:xs -> writeSmallArray sma i x *> go (i+1) xs
-  fromList l = fromListN (length l) l
-  toList sa = indexSmallArray sa <$> [0 .. length sa - 1]
+  fromListN = smallArrayFromListN
+  fromList = smallArrayFromList
+  toList = Foldable.toList
 
+smallArrayLiftShowsPrec :: (Int -> a -> ShowS) -> ([a] -> ShowS) -> Int -> SmallArray a -> ShowS
+smallArrayLiftShowsPrec elemShowsPrec elemListShowsPrec p sa = showParen (p > 10) $
+  showString "fromListN " . shows (length sa) . showString " "
+    . listLiftShowsPrec elemShowsPrec elemListShowsPrec 11 (toList sa)
+
+-- this need to be included for older ghcs
+listLiftShowsPrec :: (Int -> a -> ShowS) -> ([a] -> ShowS) -> Int -> [a] -> ShowS
+listLiftShowsPrec _ sl _ = sl
+
 instance Show a => Show (SmallArray a) where
-  showsPrec p sa = showParen (p > 10) $
-    showString "fromListN " . shows (length sa) . showString " "
-      . shows (toList sa)
+  showsPrec p sa = smallArrayLiftShowsPrec showsPrec showList p sa
 
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+-- | @since 0.6.4.0
+instance Show1 SmallArray where
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,5,0)
+  liftShowsPrec = smallArrayLiftShowsPrec
+#else
+  showsPrec1 = smallArrayLiftShowsPrec showsPrec showList
+#endif
+#endif
+
+smallArrayLiftReadsPrec :: (Int -> ReadS a) -> ReadS [a] -> Int -> ReadS (SmallArray a)
+smallArrayLiftReadsPrec _ listReadsPrec p = readParen (p > 10) . readP_to_S $ do
+  () <$ string "fromListN"
+  skipSpaces
+  n <- readS_to_P reads
+  skipSpaces
+  l <- readS_to_P listReadsPrec
+  return $ smallArrayFromListN n l
+
 instance Read a => Read (SmallArray a) where
-  readPrec = parens . prec 10 $ do
-    Symbol "fromListN" <- lexP
-    Number nu <- lexP
-    n <- maybe empty pure $ numberToInteger nu
-    fromListN (fromIntegral n) <$> readPrec
+  readsPrec = smallArrayLiftReadsPrec readsPrec readList
 
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+-- | @since 0.6.4.0
+instance Read1 SmallArray where
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,5,0)
+  liftReadsPrec = smallArrayLiftReadsPrec
+#else
+  readsPrec1 = smallArrayLiftReadsPrec readsPrec readList
+#endif
+#endif
+
+
+
 smallArrayDataType :: DataType
 smallArrayDataType =
   mkDataType "Data.Primitive.SmallArray.SmallArray" [fromListConstr]
@@ -637,3 +941,27 @@
   gunfold _ _ = die "gunfold" "SmallMutableArray"
   dataTypeOf _ = mkNoRepType "Data.Primitive.SmallArray.SmallMutableArray"
 #endif
+
+-- | Create a 'SmallArray' from a list of a known length. If the length
+--   of the list does not match the given length, this throws an exception.
+smallArrayFromListN :: Int -> [a] -> SmallArray a
+#if HAVE_SMALL_ARRAY
+smallArrayFromListN n l =
+  createSmallArray n
+      (die "smallArrayFromListN" "uninitialized element") $ \sma ->
+  let go !ix [] = if ix == n
+        then return ()
+        else die "smallArrayFromListN" "list length less than specified size"
+      go !ix (x : xs) = if ix < n
+        then do
+          writeSmallArray sma ix x
+          go (ix+1) xs
+        else die "smallArrayFromListN" "list length greater than specified size"
+  in go 0 l
+#else
+smallArrayFromListN n l = SmallArray (Array.fromListN n l)
+#endif
+
+-- | Create a 'SmallArray' from a list.
+smallArrayFromList :: [a] -> SmallArray a
+smallArrayFromList l = smallArrayFromListN (length l) l
diff --git a/Data/Primitive/Types.hs b/Data/Primitive/Types.hs
--- a/Data/Primitive/Types.hs
+++ b/Data/Primitive/Types.hs
@@ -1,5 +1,12 @@
 {-# LANGUAGE CPP, UnboxedTuples, MagicHash, DeriveDataTypeable #-}
+{-# LANGUAGE GeneralizedNewtypeDeriving, StandaloneDeriving #-}
+{-# LANGUAGE ScopedTypeVariables #-}
+#if __GLASGOW_HASKELL__ >= 800
+{-# LANGUAGE TypeInType #-}
+#endif
 
+#include "HsBaseConfig.h"
+
 -- |
 -- Module      : Data.Primitive.Types
 -- Copyright   : (c) Roman Leshchinskiy 2009-2012
@@ -13,14 +20,17 @@
 
 module Data.Primitive.Types (
   Prim(..),
-  sizeOf, alignment,
+  sizeOf, alignment, defaultSetByteArray#, defaultSetOffAddr#,
 
   Addr(..),
+  PrimStorable(..)
 ) where
 
 import Control.Monad.Primitive
 import Data.Primitive.MachDeps
 import Data.Primitive.Internal.Operations
+import Foreign.C.Types
+import System.Posix.Types
 
 import GHC.Base (
     Int(..), Char(..),
@@ -47,10 +57,18 @@
 import Data.Typeable ( Typeable )
 import Data.Data ( Data(..) )
 import Data.Primitive.Internal.Compat ( isTrue#, mkNoRepType )
+import Foreign.Storable (Storable)
+import Numeric
 
+import qualified Foreign.Storable as FS
+
 -- | A machine address
 data Addr = Addr Addr# deriving ( Typeable )
 
+instance Show Addr where
+  showsPrec _ (Addr a) =
+    showString "0x" . showHex (fromIntegral (I# (addr2Int# a)) :: Word)
+
 instance Eq Addr where
   Addr a# == Addr b# = isTrue# (eqAddr# a# b#)
   Addr a# /= Addr b# = isTrue# (neAddr# a# b#)
@@ -110,13 +128,87 @@
   setOffAddr# :: Addr# -> Int# -> Int# -> a -> State# s -> State# s
 
 -- | Size of values of type @a@. The argument is not used.
+--
+-- This function has existed since 0.1, but was moved from 'Data.Primitive'
+-- to 'Data.Primitive.Types' in version 0.6.3.0
 sizeOf :: Prim a => a -> Int
 sizeOf x = I# (sizeOf# x)
 
 -- | Alignment of values of type @a@. The argument is not used.
+--
+-- This function has existed since 0.1, but was moved from 'Data.Primitive'
+-- to 'Data.Primitive.Types' in version 0.6.3.0
 alignment :: Prim a => a -> Int
 alignment x = I# (alignment# x)
 
+-- | An implementation of 'setByteArray#' that calls 'writeByteArray#'
+-- to set each element. This is helpful when writing a 'Prim' instance
+-- for a multi-word data type for which there is no cpu-accelerated way
+-- to broadcast a value to contiguous memory. It is typically used
+-- alongside 'defaultSetOffAddr#'. For example:
+--
+-- > data Trip = Trip Int Int Int
+-- >
+-- > instance Prim Trip
+-- >   sizeOf# _ = 3# *# sizeOf# (undefined :: Int)
+-- >   alignment# _ = alignment# (undefined :: Int)
+-- >   indexByteArray# arr# i# = ...
+-- >   readByteArray# arr# i# = ...
+-- >   writeByteArray# arr# i# (Trip a b c) =
+-- >     \s0 -> case writeByteArray# arr# (3# *# i#) a s0 of
+-- >        s1 -> case writeByteArray# arr# ((3# *# i#) +# 1#) b s1 of
+-- >          s2 -> case writeByteArray# arr# ((3# *# i#) +# 2# ) c s2 of
+-- >            s3 -> s3
+-- >   setByteArray# = defaultSetByteArray#
+-- >   indexOffAddr# addr# i# = ...
+-- >   readOffAddr# addr# i# = ...
+-- >   writeOffAddr# addr# i# (Trip a b c) =
+-- >     \s0 -> case writeOffAddr# addr# (3# *# i#) a s0 of
+-- >        s1 -> case writeOffAddr# addr# ((3# *# i#) +# 1#) b s1 of
+-- >          s2 -> case writeOffAddr# addr# ((3# *# i#) +# 2# ) c s2 of
+-- >            s3 -> s3
+-- >   setOffAddr# = defaultSetOffAddr#
+defaultSetByteArray# :: Prim a => MutableByteArray# s -> Int# -> Int# -> a -> State# s -> State# s
+defaultSetByteArray# arr# i# len# ident = go 0#
+  where
+  go ix# s0 = if isTrue# (ix# <# len#)
+    then case writeByteArray# arr# (i# +# ix#) ident s0 of
+      s1 -> go (ix# +# 1#) s1
+    else s0
+
+-- | An implementation of 'setOffAddr#' that calls 'writeOffAddr#'
+-- to set each element. The documentation of 'defaultSetByteArray#'
+-- provides an example of how to use this.
+defaultSetOffAddr# :: Prim a => Addr# -> Int# -> Int# -> a -> State# s -> State# s
+defaultSetOffAddr# addr# i# len# ident = go 0#
+  where
+  go ix# s0 = if isTrue# (ix# <# len#)
+    then case writeOffAddr# addr# (i# +# ix#) ident s0 of
+      s1 -> go (ix# +# 1#) s1
+    else s0
+
+-- | Newtype that uses a 'Prim' instance to give rise to a 'Storable' instance.
+-- This type is intended to be used with the @DerivingVia@ extension available
+-- in GHC 8.6 and up. For example, consider a user-defined 'Prim' instance for
+-- a multi-word data type.
+--
+-- > data Uuid = Uuid Word64 Word64
+-- >   deriving Storable via (PrimStorable Uuid)
+-- > instance Prim Uuid where ...
+--
+-- Writing the 'Prim' instance is tedious and unavoidable, but the 'Storable'
+-- instance comes for free once the 'Prim' instance is written.
+newtype PrimStorable a = PrimStorable { getPrimStorable :: a }
+
+instance Prim a => Storable (PrimStorable a) where
+  sizeOf _ = sizeOf (undefined :: a)
+  alignment _ = alignment (undefined :: a)
+  peekElemOff (Ptr addr#) (I# i#) =
+    primitive $ \s0# -> case readOffAddr# addr# i# s0# of
+      (# s1, x #) -> (# s1, PrimStorable x #)
+  pokeElemOff (Ptr addr#) (I# i#) (PrimStorable a) = primitive_ $ \s# ->
+    writeOffAddr# addr# i# a s#
+
 #define derivePrim(ty, ctr, sz, align, idx_arr, rd_arr, wr_arr, set_arr, idx_addr, rd_addr, wr_addr, set_addr) \
 instance Prim (ty) where {                                      \
   sizeOf# _ = unI# sz                                           \
@@ -205,3 +297,99 @@
 derivePrim(FunPtr a, FunPtr, sIZEOF_PTR, aLIGNMENT_PTR,
            indexAddrArray#, readAddrArray#, writeAddrArray#, setAddrArray#,
            indexAddrOffAddr#, readAddrOffAddr#, writeAddrOffAddr#, setAddrOffAddr#)
+
+-- Prim instances for newtypes in Foreign.C.Types
+deriving instance Prim CChar
+deriving instance Prim CSChar
+deriving instance Prim CUChar
+deriving instance Prim CShort
+deriving instance Prim CUShort
+deriving instance Prim CInt
+deriving instance Prim CUInt
+deriving instance Prim CLong
+deriving instance Prim CULong
+deriving instance Prim CPtrdiff
+deriving instance Prim CSize
+deriving instance Prim CWchar
+deriving instance Prim CSigAtomic
+deriving instance Prim CLLong
+deriving instance Prim CULLong
+#if MIN_VERSION_base(4,10,0)
+deriving instance Prim CBool
+#endif
+deriving instance Prim CIntPtr
+deriving instance Prim CUIntPtr
+deriving instance Prim CIntMax
+deriving instance Prim CUIntMax
+deriving instance Prim CClock
+deriving instance Prim CTime
+deriving instance Prim CUSeconds
+deriving instance Prim CSUSeconds
+deriving instance Prim CFloat
+deriving instance Prim CDouble
+
+-- Prim instances for newtypes in System.Posix.Types
+#if defined(HTYPE_DEV_T)
+deriving instance Prim CDev
+#endif
+#if defined(HTYPE_INO_T)
+deriving instance Prim CIno
+#endif
+#if defined(HTYPE_MODE_T)
+deriving instance Prim CMode
+#endif
+#if defined(HTYPE_OFF_T)
+deriving instance Prim COff
+#endif
+#if defined(HTYPE_PID_T)
+deriving instance Prim CPid
+#endif
+#if defined(HTYPE_SSIZE_T)
+deriving instance Prim CSsize
+#endif
+#if defined(HTYPE_GID_T)
+deriving instance Prim CGid
+#endif
+#if defined(HTYPE_NLINK_T)
+deriving instance Prim CNlink
+#endif
+#if defined(HTYPE_UID_T)
+deriving instance Prim CUid
+#endif
+#if defined(HTYPE_CC_T)
+deriving instance Prim CCc
+#endif
+#if defined(HTYPE_SPEED_T)
+deriving instance Prim CSpeed
+#endif
+#if defined(HTYPE_TCFLAG_T)
+deriving instance Prim CTcflag
+#endif
+#if defined(HTYPE_RLIM_T)
+deriving instance Prim CRLim
+#endif
+#if defined(HTYPE_BLKSIZE_T)
+deriving instance Prim CBlkSize
+#endif
+#if defined(HTYPE_BLKCNT_T)
+deriving instance Prim CBlkCnt
+#endif
+#if defined(HTYPE_CLOCKID_T)
+deriving instance Prim CClockId
+#endif
+#if defined(HTYPE_FSBLKCNT_T)
+deriving instance Prim CFsBlkCnt
+#endif
+#if defined(HTYPE_FSFILCNT_T)
+deriving instance Prim CFsFilCnt
+#endif
+#if defined(HTYPE_ID_T)
+deriving instance Prim CId
+#endif
+#if defined(HTYPE_KEY_T)
+deriving instance Prim CKey
+#endif
+#if defined(HTYPE_TIMER_T)
+deriving instance Prim CTimer
+#endif
+deriving instance Prim Fd
diff --git a/Data/Primitive/UnliftedArray.hs b/Data/Primitive/UnliftedArray.hs
--- a/Data/Primitive/UnliftedArray.hs
+++ b/Data/Primitive/UnliftedArray.hs
@@ -1,6 +1,11 @@
+{-# Language BangPatterns #-}
+{-# Language CPP #-}
+{-# Language DeriveDataTypeable #-}
 {-# Language MagicHash #-}
+{-# Language RankNTypes #-}
+{-# Language ScopedTypeVariables #-}
+{-# Language TypeFamilies #-}
 {-# Language UnboxedTuples #-}
-{-# Language DeriveDataTypeable #-}
 
 -- |
 -- Module      : Data.Primitive.UnliftedArray
@@ -39,9 +44,11 @@
 -- that are eligible to be stored.
 
 module Data.Primitive.UnliftedArray
-  ( UnliftedArray(..)
+  ( -- * Types
+    UnliftedArray(..)
   , MutableUnliftedArray(..)
   , PrimUnlifted(..)
+    -- * Operations
   , unsafeNewUnliftedArray
   , newUnliftedArray
   , setUnliftedArray
@@ -54,33 +61,68 @@
   , unsafeFreezeUnliftedArray
   , freezeUnliftedArray
   , thawUnliftedArray
+  , runUnliftedArray
   , sameMutableUnliftedArray
   , copyUnliftedArray
   , copyMutableUnliftedArray
   , cloneUnliftedArray
   , cloneMutableUnliftedArray
+    -- * List Conversion
+  , unliftedArrayToList
+  , unliftedArrayFromList
+  , unliftedArrayFromListN
+    -- * Folding
+  , foldrUnliftedArray
+  , foldrUnliftedArray'
+  , foldlUnliftedArray
+  , foldlUnliftedArray'
+    -- * Mapping
+  , mapUnliftedArray
 -- Missing operations:
 --  , unsafeThawUnliftedArray
   ) where
 
 import Data.Typeable
+import Control.Applicative
 
 import GHC.Prim
-import GHC.Base (Int(..))
+import GHC.Base (Int(..),build)
 
 import Control.Monad.Primitive
 
-import Control.Monad.ST (runST)
+import Control.Monad.ST (runST,ST)
 
+import Data.Monoid (Monoid,mappend)
 import Data.Primitive.Internal.Compat ( isTrue# )
 
+import qualified Data.List as L
 import           Data.Primitive.Array (Array)
 import qualified Data.Primitive.Array as A
 import           Data.Primitive.ByteArray (ByteArray)
 import qualified Data.Primitive.ByteArray as BA
+import qualified Data.Primitive.PrimArray as PA
 import qualified Data.Primitive.SmallArray as SA
 import qualified Data.Primitive.MutVar as MV
+import qualified Data.Monoid
+import qualified GHC.MVar as GM (MVar(..))
+import qualified GHC.Conc as GC (TVar(..))
+import qualified GHC.Stable as GSP (StablePtr(..))
+import qualified GHC.Weak as GW (Weak(..))
+import qualified GHC.Conc.Sync as GCS (ThreadId(..))
+import qualified GHC.Exts as E
+import qualified GHC.ST as GHCST
 
+#if MIN_VERSION_base(4,9,0)
+import Data.Semigroup (Semigroup)
+import qualified Data.Semigroup
+#endif
+
+#if MIN_VERSION_base(4,10,0)
+import GHC.Exts (runRW#)
+#elif MIN_VERSION_base(4,9,0)
+import GHC.Base (runRW#)
+#endif
+
 -- | Immutable arrays that efficiently store types that are simple wrappers
 -- around unlifted primitive types. The values of the unlifted type are
 -- stored directly, eliminating a layer of indirection.
@@ -125,6 +167,16 @@
   toArrayArray# (BA.MutableByteArray mba#) = unsafeCoerce# mba#
   fromArrayArray# aa# = BA.MutableByteArray (unsafeCoerce# aa#)
 
+-- | @since 0.6.4.0
+instance PrimUnlifted (PA.PrimArray a) where
+  toArrayArray# (PA.PrimArray ba#) = unsafeCoerce# ba#
+  fromArrayArray# aa# = PA.PrimArray (unsafeCoerce# aa#)
+
+-- | @since 0.6.4.0
+instance PrimUnlifted (PA.MutablePrimArray s a) where
+  toArrayArray# (PA.MutablePrimArray mba#) = unsafeCoerce# mba#
+  fromArrayArray# aa# = PA.MutablePrimArray (unsafeCoerce# aa#)
+
 instance PrimUnlifted (SA.SmallArray a) where
   toArrayArray# (SA.SmallArray sa#) = unsafeCoerce# sa#
   fromArrayArray# aa# = SA.SmallArray (unsafeCoerce# aa#)
@@ -137,8 +189,38 @@
   toArrayArray# (MV.MutVar mv#) = unsafeCoerce# mv#
   fromArrayArray# aa# = MV.MutVar (unsafeCoerce# aa#)
 
--- | Creates a new 'MutableUnliftedArray'. This function is unsafe, because it
--- allows access to the raw contents of the underlying 'ArrayArray#'.
+-- | @since 0.6.4.0
+instance PrimUnlifted (GM.MVar a) where
+  toArrayArray# (GM.MVar mv#) = unsafeCoerce# mv#
+  fromArrayArray# mv# = GM.MVar (unsafeCoerce# mv#)
+
+-- | @since 0.6.4.0
+instance PrimUnlifted (GC.TVar a) where
+  toArrayArray# (GC.TVar tv#) = unsafeCoerce# tv#
+  fromArrayArray# tv# = GC.TVar (unsafeCoerce# tv#)
+
+-- | @since 0.6.4.0
+instance PrimUnlifted (GSP.StablePtr a) where
+  toArrayArray# (GSP.StablePtr tv#) = unsafeCoerce# tv#
+  fromArrayArray# tv# = GSP.StablePtr (unsafeCoerce# tv#)
+
+-- | @since 0.6.4.0
+instance PrimUnlifted (GW.Weak a) where
+  toArrayArray# (GW.Weak tv#) = unsafeCoerce# tv#
+  fromArrayArray# tv# = GW.Weak (unsafeCoerce# tv#)
+
+-- | @since 0.6.4.0
+instance PrimUnlifted GCS.ThreadId where
+  toArrayArray# (GCS.ThreadId tv#) = unsafeCoerce# tv#
+  fromArrayArray# tv# = GCS.ThreadId (unsafeCoerce# tv#)
+
+die :: String -> String -> a
+die fun problem = error $ "Data.Primitive.UnliftedArray." ++ fun ++ ": " ++ problem
+
+-- | Creates a new 'MutableUnliftedArray'. This function is unsafe because it
+-- initializes all elements of the array as pointers to the array itself. Attempting
+-- to read one of these elements before writing to it is in effect an unsafe
+-- coercion from the @MutableUnliftedArray s a@ to the element type.
 unsafeNewUnliftedArray
   :: (PrimMonad m)
   => Int -- ^ size
@@ -329,6 +411,63 @@
   return dst
 {-# inline thawUnliftedArray #-}
 
+#if !MIN_VERSION_base(4,9,0)
+unsafeCreateUnliftedArray
+  :: Int
+  -> (forall s. MutableUnliftedArray s a -> ST s ())
+  -> UnliftedArray a
+unsafeCreateUnliftedArray 0 _ = emptyUnliftedArray
+unsafeCreateUnliftedArray n f = runUnliftedArray $ do
+  mary <- unsafeNewUnliftedArray n
+  f mary
+  pure mary
+
+-- | Execute a stateful computation and freeze the resulting array.
+runUnliftedArray
+  :: (forall s. ST s (MutableUnliftedArray s a))
+  -> UnliftedArray a
+runUnliftedArray m = runST $ m >>= unsafeFreezeUnliftedArray
+
+#else /* Below, runRW# is available. */
+
+-- This low-level business is designed to work with GHC's worker-wrapper
+-- transformation. A lot of the time, we don't actually need an Array
+-- constructor. By putting it on the outside, and being careful about
+-- how we special-case the empty array, we can make GHC smarter about this.
+-- The only downside is that separately created 0-length arrays won't share
+-- their Array constructors, although they'll share their underlying
+-- Array#s.
+unsafeCreateUnliftedArray
+  :: Int
+  -> (forall s. MutableUnliftedArray s a -> ST s ())
+  -> UnliftedArray a
+unsafeCreateUnliftedArray 0 _ = UnliftedArray (emptyArrayArray# (# #))
+unsafeCreateUnliftedArray n f = runUnliftedArray $ do
+  mary <- unsafeNewUnliftedArray n
+  f mary
+  pure mary
+
+-- | Execute a stateful computation and freeze the resulting array.
+runUnliftedArray
+  :: (forall s. ST s (MutableUnliftedArray s a))
+  -> UnliftedArray a
+runUnliftedArray m = UnliftedArray (runUnliftedArray# m)
+
+runUnliftedArray#
+  :: (forall s. ST s (MutableUnliftedArray s a))
+  -> ArrayArray#
+runUnliftedArray# m = case runRW# $ \s ->
+  case unST m s of { (# s', MutableUnliftedArray mary# #) ->
+  unsafeFreezeArrayArray# mary# s'} of (# _, ary# #) -> ary#
+
+unST :: ST s a -> State# s -> (# State# s, a #)
+unST (GHCST.ST f) = f
+
+emptyArrayArray# :: (# #) -> ArrayArray#
+emptyArrayArray# _ = case emptyUnliftedArray of UnliftedArray ar -> ar
+{-# NOINLINE emptyArrayArray# #-}
+#endif
+
 -- | Creates a copy of a portion of an 'UnliftedArray'
 cloneUnliftedArray
   :: UnliftedArray a -- ^ source
@@ -336,7 +475,7 @@
   -> Int -- ^ length
   -> UnliftedArray a
 cloneUnliftedArray src off len =
-  runST $ thawUnliftedArray src off len >>= unsafeFreezeUnliftedArray
+  runUnliftedArray (thawUnliftedArray src off len)
 {-# inline cloneUnliftedArray #-}
 
 -- | Creates a new 'MutableUnliftedArray' containing a copy of a portion of
@@ -363,3 +502,137 @@
    loop i
      | i < 0 = True
      | otherwise = indexUnliftedArray aa1 i == indexUnliftedArray aa2 i && loop (i-1)
+
+-- | Lexicographic ordering. Subject to change between major versions.
+--
+--   @since 0.6.4.0
+instance (Ord a, PrimUnlifted a) => Ord (UnliftedArray a) where
+  compare a1 a2 = loop 0
+    where
+    mn = sizeofUnliftedArray a1 `min` sizeofUnliftedArray a2
+    loop i
+      | i < mn
+      , x1 <- indexUnliftedArray a1 i
+      , x2 <- indexUnliftedArray a2 i
+      = compare x1 x2 `mappend` loop (i+1)
+      | otherwise = compare (sizeofUnliftedArray a1) (sizeofUnliftedArray a2)
+
+-- | @since 0.6.4.0
+instance (Show a, PrimUnlifted a) => Show (UnliftedArray a) where
+  showsPrec p a = showParen (p > 10) $
+    showString "fromListN " . shows (sizeofUnliftedArray a) . showString " "
+      . shows (unliftedArrayToList a)
+
+#if MIN_VERSION_base(4,9,0)
+-- | @since 0.6.4.0
+instance PrimUnlifted a => Semigroup (UnliftedArray a) where
+  (<>) = concatUnliftedArray
+#endif
+
+-- | @since 0.6.4.0
+instance PrimUnlifted a => Monoid (UnliftedArray a) where
+  mempty = emptyUnliftedArray
+#if !(MIN_VERSION_base(4,11,0))
+  mappend = concatUnliftedArray
+#endif
+
+emptyUnliftedArray :: UnliftedArray a
+emptyUnliftedArray = runUnliftedArray (unsafeNewUnliftedArray 0)
+{-# NOINLINE emptyUnliftedArray #-}
+
+concatUnliftedArray :: UnliftedArray a -> UnliftedArray a -> UnliftedArray a
+concatUnliftedArray x y = unsafeCreateUnliftedArray (sizeofUnliftedArray x + sizeofUnliftedArray y) $ \m -> do
+  copyUnliftedArray m 0 x 0 (sizeofUnliftedArray x)
+  copyUnliftedArray m (sizeofUnliftedArray x) y 0 (sizeofUnliftedArray y)
+
+-- | Lazy right-associated fold over the elements of an 'UnliftedArray'.
+{-# INLINE foldrUnliftedArray #-}
+foldrUnliftedArray :: forall a b. PrimUnlifted a => (a -> b -> b) -> b -> UnliftedArray a -> b
+foldrUnliftedArray f z arr = go 0
+  where
+    !sz = sizeofUnliftedArray arr
+    go !i
+      | sz > i = f (indexUnliftedArray arr i) (go (i+1))
+      | otherwise = z
+
+-- | Strict right-associated fold over the elements of an 'UnliftedArray.
+{-# INLINE foldrUnliftedArray' #-}
+foldrUnliftedArray' :: forall a b. PrimUnlifted a => (a -> b -> b) -> b -> UnliftedArray a -> b
+foldrUnliftedArray' f z0 arr = go (sizeofUnliftedArray arr - 1) z0
+  where
+    go !i !acc
+      | i < 0 = acc
+      | otherwise = go (i - 1) (f (indexUnliftedArray arr i) acc)
+
+-- | Lazy left-associated fold over the elements of an 'UnliftedArray'.
+{-# INLINE foldlUnliftedArray #-}
+foldlUnliftedArray :: forall a b. PrimUnlifted a => (b -> a -> b) -> b -> UnliftedArray a -> b
+foldlUnliftedArray f z arr = go (sizeofUnliftedArray arr - 1)
+  where
+    go !i
+      | i < 0 = z
+      | otherwise = f (go (i - 1)) (indexUnliftedArray arr i)
+
+-- | Strict left-associated fold over the elements of an 'UnliftedArray'.
+{-# INLINE foldlUnliftedArray' #-}
+foldlUnliftedArray' :: forall a b. PrimUnlifted a => (b -> a -> b) -> b -> UnliftedArray a -> b
+foldlUnliftedArray' f z0 arr = go 0 z0
+  where
+    !sz = sizeofUnliftedArray arr
+    go !i !acc
+      | i < sz = go (i + 1) (f acc (indexUnliftedArray arr i))
+      | otherwise = acc
+
+-- | Map over the elements of an 'UnliftedArray'.
+{-# INLINE mapUnliftedArray #-}
+mapUnliftedArray :: (PrimUnlifted a, PrimUnlifted b)
+  => (a -> b)
+  -> UnliftedArray a
+  -> UnliftedArray b
+mapUnliftedArray f arr = unsafeCreateUnliftedArray sz $ \marr -> do
+  let go !ix = if ix < sz
+        then do
+          let b = f (indexUnliftedArray arr ix)
+          writeUnliftedArray marr ix b
+          go (ix + 1)
+        else return ()
+  go 0
+  where
+  !sz = sizeofUnliftedArray arr
+
+-- | Convert the unlifted array to a list.
+{-# INLINE unliftedArrayToList #-}
+unliftedArrayToList :: PrimUnlifted a => UnliftedArray a -> [a]
+unliftedArrayToList xs = build (\c n -> foldrUnliftedArray c n xs)
+
+unliftedArrayFromList :: PrimUnlifted a => [a] -> UnliftedArray a
+unliftedArrayFromList xs = unliftedArrayFromListN (L.length xs) xs
+
+unliftedArrayFromListN :: forall a. PrimUnlifted a => Int -> [a] -> UnliftedArray a
+unliftedArrayFromListN len vs = unsafeCreateUnliftedArray len run where
+  run :: forall s. MutableUnliftedArray s a -> ST s ()
+  run arr = do
+    let go :: [a] -> Int -> ST s ()
+        go [] !ix = if ix == len
+          -- The size check is mandatory since failure to initialize all elements
+          -- introduces the possibility of a segfault happening when someone attempts
+          -- to read the unitialized element. See the docs for unsafeNewUnliftedArray.
+          then return ()
+          else die "unliftedArrayFromListN" "list length less than specified size"
+        go (a : as) !ix = if ix < len
+          then do
+            writeUnliftedArray arr ix a
+            go as (ix + 1)
+          else die "unliftedArrayFromListN" "list length greater than specified size"
+    go vs 0
+
+
+#if MIN_VERSION_base(4,7,0)
+-- | @since 0.6.4.0
+instance PrimUnlifted a => E.IsList (UnliftedArray a) where
+  type Item (UnliftedArray a) = a
+  fromList = unliftedArrayFromList
+  fromListN = unliftedArrayFromListN
+  toList = unliftedArrayToList
+#endif
+
diff --git a/changelog.md b/changelog.md
--- a/changelog.md
+++ b/changelog.md
@@ -1,3 +1,64 @@
+## Changes in version 0.6.4.0
+
+ * Introduce `Data.Primitive.PrimArray`, which offers types and function
+   for dealing with a `ByteArray` tagged with a phantom type variable for
+   tracking the element type.
+
+ * Implement `isByteArrayPinned` and `isMutableByteArrayPinned`.
+
+ * Add `Eq1`, `Ord1`, `Show1`, and `Read1` instances for `Array` and
+   `SmallArray`.
+
+ * Improve the test suite. This includes having property tests for
+   typeclasses from `base` such as `Eq`, `Ord`, `Functor`, `Applicative`,
+   `Monad`, `IsList`, `Monoid`, `Foldable`, and `Traversable`.
+
+ * Fix the broken `IsList` instance for `ByteArray`. The old definition
+   would allocate a byte array of the correct size and then leave the
+   memory unitialized instead of writing the list elements to it.
+
+ * Fix the broken `Functor` instance for `Array`. The old definition
+   would allocate an array of the correct size with thunks for erroring
+   installed at every index. It failed to replace these thunks with
+   the result of the function applied to the elements of the argument array.
+
+ * Fix the broken `Applicative` instances of `Array` and `SmallArray`.
+   The old implementation of `<*>` for `Array` failed to initialize
+   some elements but correctly initialized others in the resulting
+   `Array`. It is unclear what the old behavior of `<*>` was for
+   `SmallArray`, but it was incorrect.
+
+ * Fix the broken `Monad` instances for `Array` and `SmallArray`.
+
+ * Fix the implementation of `foldl1` in the `Foldable` instances for
+   `Array` and `SmallArray`. In both cases, the old implementation
+   simply returned the first element of the array and made no use of
+   the other elements in the array.
+
+ * Fix the implementation of `mconcat` in the `Monoid` instance for
+   `SmallArray`.
+ 
+ * Implement `Data.Primitive.Ptr`, implementations of `Ptr` functions
+   that require a `Prim` constraint instead of a `Storable` constraint.
+
+
+ * Add `PrimUnlifted` instances for `TVar` and `MVar`.
+
+ * Use `compareByteArrays#` for the `Eq` and `Ord` instances of
+   `ByteArray` when building with GHC 8.4 and newer.
+
+ * Add `Prim` instances for lots of types in `Foreign.C.Types` and
+   `System.Posix.Types`.
+
+ * Reexport `Data.Primitive.SmallArray` and `Data.Primitive.UnliftedArray`
+   from `Data.Primitive`.
+
+ * Add fold functions and map function to `Data.Primitive.UnliftedArray`.
+   Add typeclass instances for `IsList`, `Ord`, and `Show`.
+
+ * Add `defaultSetByteArray#` and `defaultSetOffAddr#` to
+   `Data.Primitive.Types`.
+
 ## Changes in version 0.6.3.0
 
  * Add `PrimMonad` instances for `ContT`, `AccumT`, and `SelectT` from
diff --git a/primitive.cabal b/primitive.cabal
--- a/primitive.cabal
+++ b/primitive.cabal
@@ -1,5 +1,5 @@
 Name:           primitive
-Version:        0.6.3.0
+Version:        0.6.4.0
 License:        BSD3
 License-File:   LICENSE
 
@@ -15,6 +15,9 @@
 Description:    This package provides various primitive memory-related operations.
 
 Extra-Source-Files: changelog.md
+                    test/*.hs
+                    test/LICENSE
+                    test/primitive-tests.cabal
 
 Tested-With:
   GHC == 7.4.2,
@@ -23,7 +26,7 @@
   GHC == 7.10.3,
   GHC == 8.0.2,
   GHC == 8.2.2,
-  GHC == 8.4.1
+  GHC == 8.4.2
 
 Library
   Default-Language: Haskell2010
@@ -38,10 +41,13 @@
         Data.Primitive.Types
         Data.Primitive.Array
         Data.Primitive.ByteArray
+        Data.Primitive.PrimArray
         Data.Primitive.SmallArray
         Data.Primitive.UnliftedArray
         Data.Primitive.Addr
+        Data.Primitive.Ptr
         Data.Primitive.MutVar
+        Data.Primitive.MVar
 
   Other-Modules:
         Data.Primitive.Internal.Compat
@@ -51,27 +57,16 @@
                , ghc-prim >= 0.2 && < 0.6
                , transformers >= 0.2 && < 0.6
 
-  Ghc-Options: -O2 -Wall
+  Ghc-Options: -O2
 
   Include-Dirs: cbits
   Install-Includes: primitive-memops.h
   includes: primitive-memops.h
   c-sources: cbits/primitive-memops.c
-  cc-options: -O3 -fomit-frame-pointer -Wall
   if !os(solaris)
       cc-options: -ftree-vectorize
   if arch(i386) || arch(x86_64)
       cc-options: -msse2
-
-test-suite test
-  Default-Language: Haskell2010
-  hs-source-dirs: test
-  main-is: main.hs
-  type: exitcode-stdio-1.0
-  build-depends: base
-               , ghc-prim
-               , primitive
-  ghc-options: -O2
 
 source-repository head
   type:     git
diff --git a/test/LICENSE b/test/LICENSE
new file mode 100644
--- /dev/null
+++ b/test/LICENSE
@@ -0,0 +1,30 @@
+Copyright (c) 2008-2009, Roman Leshchinskiy
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+- Redistributions of source code must retain the above copyright notice,
+this list of conditions and the following disclaimer.
+ 
+- Redistributions in binary form must reproduce the above copyright notice,
+this list of conditions and the following disclaimer in the documentation
+and/or other materials provided with the distribution.
+ 
+- Neither name of the University nor the names of its contributors may be
+used to endorse or promote products derived from this software without
+specific prior written permission. 
+
+THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY COURT OF THE UNIVERSITY OF
+GLASGOW AND THE CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
+INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+UNIVERSITY COURT OF THE UNIVERSITY OF GLASGOW OR THE CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
+DAMAGE.
+
diff --git a/test/main.hs b/test/main.hs
--- a/test/main.hs
+++ b/test/main.hs
@@ -1,26 +1,228 @@
-{-# LANGUAGE MagicHash, UnboxedTuples #-}
+{-# LANGUAGE BangPatterns #-}
+{-# LANGUAGE CPP #-}
+{-# LANGUAGE GeneralizedNewtypeDeriving #-}
+{-# LANGUAGE KindSignatures #-}
+{-# LANGUAGE MagicHash #-}
+{-# LANGUAGE UnboxedTuples #-}
+{-# LANGUAGE ScopedTypeVariables #-}
 
+import Control.Applicative
 import Control.Monad
+import Control.Monad.Fix (fix)
 import Control.Monad.Primitive
 import Control.Monad.ST
+import Data.Monoid
 import Data.Primitive
 import Data.Primitive.Array
 import Data.Primitive.ByteArray
 import Data.Primitive.Types
+import Data.Primitive.SmallArray
+import Data.Primitive.PrimArray
 import Data.Word
+import Data.Proxy (Proxy(..))
 import GHC.Int
 import GHC.IO
 import GHC.Prim
+import Data.Function (on)
+#if MIN_VERSION_base(4,9,0)
+import Data.Semigroup (stimes)
+#endif
 
--- Since we only have two test cases right now, I'm going to avoid the
--- issue of choosing a test framework for the moment. This also keeps the
--- package as a whole light on dependencies.
+import Test.Tasty (defaultMain,testGroup,TestTree)
+import Test.QuickCheck (Arbitrary,Arbitrary1,Gen,(===),CoArbitrary,Function)
+import qualified Test.Tasty.QuickCheck as TQC
+import qualified Test.QuickCheck as QC
+import qualified Test.QuickCheck.Classes as QCC
+import qualified Test.QuickCheck.Classes.IsList as QCCL
+import qualified Data.List as L
 
 main :: IO ()
 main = do
-    testArray
-    testByteArray
+  testArray
+  testByteArray
+  defaultMain $ testGroup "properties"
+    [ testGroup "Array"
+      [ lawsToTest (QCC.eqLaws (Proxy :: Proxy (Array Int)))
+      , lawsToTest (QCC.ordLaws (Proxy :: Proxy (Array Int)))
+      , lawsToTest (QCC.monoidLaws (Proxy :: Proxy (Array Int)))
+      , lawsToTest (QCC.showReadLaws (Proxy :: Proxy (Array Int)))
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+      , lawsToTest (QCC.functorLaws (Proxy1 :: Proxy1 Array))
+      , lawsToTest (QCC.applicativeLaws (Proxy1 :: Proxy1 Array))
+      , lawsToTest (QCC.monadLaws (Proxy1 :: Proxy1 Array))
+      , lawsToTest (QCC.foldableLaws (Proxy1 :: Proxy1 Array))
+      , lawsToTest (QCC.traversableLaws (Proxy1 :: Proxy1 Array))
+#endif
+#if MIN_VERSION_base(4,7,0)
+      , lawsToTest (QCC.isListLaws (Proxy :: Proxy (Array Int)))
+      , TQC.testProperty "mapArray'" (QCCL.mapProp int16 int32 mapArray')
+#endif
+      ]
+    , testGroup "SmallArray"
+      [ lawsToTest (QCC.eqLaws (Proxy :: Proxy (SmallArray Int)))
+      , lawsToTest (QCC.ordLaws (Proxy :: Proxy (SmallArray Int)))
+      , lawsToTest (QCC.monoidLaws (Proxy :: Proxy (SmallArray Int)))
+      , lawsToTest (QCC.showReadLaws (Proxy :: Proxy (Array Int)))
+#if MIN_VERSION_base(4,9,0) || MIN_VERSION_transformers(0,4,0)
+      , lawsToTest (QCC.functorLaws (Proxy1 :: Proxy1 SmallArray))
+      , lawsToTest (QCC.applicativeLaws (Proxy1 :: Proxy1 SmallArray))
+      , lawsToTest (QCC.monadLaws (Proxy1 :: Proxy1 SmallArray))
+      , lawsToTest (QCC.foldableLaws (Proxy1 :: Proxy1 SmallArray))
+      , lawsToTest (QCC.traversableLaws (Proxy1 :: Proxy1 SmallArray))
+#endif
+#if MIN_VERSION_base(4,7,0)
+      , lawsToTest (QCC.isListLaws (Proxy :: Proxy (SmallArray Int)))
+      , TQC.testProperty "mapSmallArray'" (QCCL.mapProp int16 int32 mapSmallArray')
+#endif
+      ]
+    , testGroup "ByteArray"
+      [ testGroup "Ordering"
+        [ TQC.testProperty "equality" byteArrayEqProp
+        , TQC.testProperty "compare" byteArrayCompareProp
+        ]
+      , testGroup "Resize"
+        [ TQC.testProperty "shrink" byteArrayShrinkProp
+        , TQC.testProperty "grow" byteArrayGrowProp
+        ]
+      , lawsToTest (QCC.eqLaws (Proxy :: Proxy ByteArray))
+      , lawsToTest (QCC.ordLaws (Proxy :: Proxy ByteArray))
+      , lawsToTest (QCC.showReadLaws (Proxy :: Proxy (Array Int)))
+#if MIN_VERSION_base(4,7,0)
+      , lawsToTest (QCC.isListLaws (Proxy :: Proxy ByteArray))
+#endif
+      ]
+    , testGroup "PrimArray"
+      [ lawsToTest (QCC.eqLaws (Proxy :: Proxy (PrimArray Word16)))
+      , lawsToTest (QCC.ordLaws (Proxy :: Proxy (PrimArray Word16)))
+      , lawsToTest (QCC.monoidLaws (Proxy :: Proxy (PrimArray Word16)))
+#if MIN_VERSION_base(4,7,0)
+      , lawsToTest (QCC.isListLaws (Proxy :: Proxy (PrimArray Word16)))
+      , TQC.testProperty "foldrPrimArray" (QCCL.foldrProp int16 foldrPrimArray)
+      , TQC.testProperty "foldrPrimArray'" (QCCL.foldrProp int16 foldrPrimArray')
+      , TQC.testProperty "foldlPrimArray" (QCCL.foldlProp int16 foldlPrimArray)
+      , TQC.testProperty "foldlPrimArray'" (QCCL.foldlProp int16 foldlPrimArray')
+      , TQC.testProperty "foldlPrimArrayM'" (QCCL.foldlMProp int16 foldlPrimArrayM')
+      , TQC.testProperty "mapPrimArray" (QCCL.mapProp int16 int32 mapPrimArray)
+      , TQC.testProperty "traversePrimArray" (QCCL.traverseProp int16 int32 traversePrimArray)
+      , TQC.testProperty "traversePrimArrayP" (QCCL.traverseProp int16 int32 traversePrimArrayP)
+      , TQC.testProperty "imapPrimArray" (QCCL.imapProp int16 int32 imapPrimArray)
+      , TQC.testProperty "itraversePrimArray" (QCCL.imapMProp int16 int32 itraversePrimArray)
+      , TQC.testProperty "itraversePrimArrayP" (QCCL.imapMProp int16 int32 itraversePrimArrayP)
+      , TQC.testProperty "generatePrimArray" (QCCL.generateProp int16 generatePrimArray)
+      , TQC.testProperty "generatePrimArrayA" (QCCL.generateMProp int16 generatePrimArrayA)
+      , TQC.testProperty "generatePrimArrayP" (QCCL.generateMProp int16 generatePrimArrayP)
+      , TQC.testProperty "replicatePrimArray" (QCCL.replicateProp int16 replicatePrimArray)
+      , TQC.testProperty "replicatePrimArrayA" (QCCL.replicateMProp int16 replicatePrimArrayA)
+      , TQC.testProperty "replicatePrimArrayP" (QCCL.replicateMProp int16 replicatePrimArrayP)
+      , TQC.testProperty "filterPrimArray" (QCCL.filterProp int16 filterPrimArray)
+      , TQC.testProperty "filterPrimArrayA" (QCCL.filterMProp int16 filterPrimArrayA)
+      , TQC.testProperty "filterPrimArrayP" (QCCL.filterMProp int16 filterPrimArrayP)
+      , TQC.testProperty "mapMaybePrimArray" (QCCL.mapMaybeProp int16 int32 mapMaybePrimArray)
+      , TQC.testProperty "mapMaybePrimArrayA" (QCCL.mapMaybeMProp int16 int32 mapMaybePrimArrayA)
+      , TQC.testProperty "mapMaybePrimArrayP" (QCCL.mapMaybeMProp int16 int32 mapMaybePrimArrayP)
+#endif
+      ]
+    , testGroup "UnliftedArray"
+      [ lawsToTest (QCC.eqLaws (Proxy :: Proxy (UnliftedArray (PrimArray Int16))))
+      , lawsToTest (QCC.ordLaws (Proxy :: Proxy (UnliftedArray (PrimArray Int16))))
+      , lawsToTest (QCC.monoidLaws (Proxy :: Proxy (UnliftedArray (PrimArray Int16))))
+#if MIN_VERSION_base(4,7,0)
+      , lawsToTest (QCC.isListLaws (Proxy :: Proxy (UnliftedArray (PrimArray Int16))))
+      , TQC.testProperty "mapUnliftedArray" (QCCL.mapProp arrInt16 arrInt32 mapUnliftedArray)
+      , TQC.testProperty "foldrUnliftedArray" (QCCL.foldrProp arrInt16 foldrUnliftedArray)
+      , TQC.testProperty "foldrUnliftedArray'" (QCCL.foldrProp arrInt16 foldrUnliftedArray')
+      , TQC.testProperty "foldlUnliftedArray" (QCCL.foldlProp arrInt16 foldlUnliftedArray)
+      , TQC.testProperty "foldlUnliftedArray'" (QCCL.foldlProp arrInt16 foldlUnliftedArray')
+#endif
+      ]
+    , testGroup "DefaultSetMethod"
+      [ lawsToTest (QCC.primLaws (Proxy :: Proxy DefaultSetMethod))
+      ]
+    -- , testGroup "PrimStorable"
+    --   [ lawsToTest (QCC.storableLaws (Proxy :: Proxy Derived))
+    --   ]
+    ]
 
+int16 :: Proxy Int16
+int16 = Proxy
+
+int32 :: Proxy Int32
+int32 = Proxy
+
+arrInt16 :: Proxy (PrimArray Int16)
+arrInt16 = Proxy
+
+arrInt32 :: Proxy (PrimArray Int16)
+arrInt32 = Proxy
+
+-- Tests that using resizeByteArray to shrink a byte array produces
+-- the same results as calling Data.List.take on the list that the
+-- byte array corresponds to.
+byteArrayShrinkProp :: QC.Property
+byteArrayShrinkProp = QC.property $ \(QC.NonNegative (n :: Int)) (QC.NonNegative (m :: Int)) ->
+  let large = max n m
+      small = min n m
+      xs = intsLessThan large
+      ys = byteArrayFromList xs
+      largeBytes = large * sizeOf (undefined :: Int)
+      smallBytes = small * sizeOf (undefined :: Int)
+      expected = byteArrayFromList (L.take small xs)
+      actual = runST $ do
+        mzs0 <- newByteArray largeBytes
+        copyByteArray mzs0 0 ys 0 largeBytes
+        mzs1 <- resizeMutableByteArray mzs0 smallBytes
+        unsafeFreezeByteArray mzs1
+   in expected === actual
+
+-- Tests that using resizeByteArray with copyByteArray (to fill in the
+-- new empty space) to grow a byte array produces the same results as
+-- calling Data.List.++ on the lists corresponding to the original
+-- byte array and the appended byte array.
+byteArrayGrowProp :: QC.Property
+byteArrayGrowProp = QC.property $ \(QC.NonNegative (n :: Int)) (QC.NonNegative (m :: Int)) ->
+  let large = max n m
+      small = min n m
+      xs1 = intsLessThan small
+      xs2 = intsLessThan (large - small)
+      ys1 = byteArrayFromList xs1
+      ys2 = byteArrayFromList xs2
+      largeBytes = large * sizeOf (undefined :: Int)
+      smallBytes = small * sizeOf (undefined :: Int)
+      expected = byteArrayFromList (xs1 ++ xs2)
+      actual = runST $ do
+        mzs0 <- newByteArray smallBytes
+        copyByteArray mzs0 0 ys1 0 smallBytes
+        mzs1 <- resizeMutableByteArray mzs0 largeBytes
+        copyByteArray mzs1 smallBytes ys2 0 ((large - small) * sizeOf (undefined :: Int))
+        unsafeFreezeByteArray mzs1
+   in expected === actual
+
+-- Provide the non-negative integers up to the bound. For example:
+--
+-- >>> intsLessThan 5
+-- [0,1,2,3,4]
+intsLessThan :: Int -> [Int]
+intsLessThan i = if i < 1
+  then []
+  else (i - 1) : intsLessThan (i - 1)
+  
+byteArrayCompareProp :: QC.Property
+byteArrayCompareProp = QC.property $ \(xs :: [Word8]) (ys :: [Word8]) ->
+  compareLengthFirst xs ys === compare (byteArrayFromList xs) (byteArrayFromList ys)
+
+byteArrayEqProp :: QC.Property
+byteArrayEqProp = QC.property $ \(xs :: [Word8]) (ys :: [Word8]) ->
+  (compareLengthFirst xs ys == EQ) === (byteArrayFromList xs == byteArrayFromList ys)
+
+compareLengthFirst :: [Word8] -> [Word8] -> Ordering
+compareLengthFirst xs ys = (compare `on` length) xs ys <> compare xs ys
+
+-- on GHC 7.4, Proxy is not polykinded, so we need this instead.
+data Proxy1 (f :: * -> *) = Proxy1
+
+lawsToTest :: QCC.Laws -> TestTree
+lawsToTest (QCC.Laws name pairs) = testGroup name (map (uncurry TQC.testProperty) pairs)
+
 testArray :: IO ()
 testArray = do
     arr <- newArray 1 'A'
@@ -41,15 +243,100 @@
     let arr1 = mkByteArray ([0xde, 0xad, 0xbe, 0xef] :: [Word8])
         arr2 = mkByteArray ([0xde, 0xad, 0xbe, 0xef] :: [Word8])
         arr3 = mkByteArray ([0xde, 0xad, 0xbe, 0xee] :: [Word8])
+        arr4 = mkByteArray ([0xde, 0xad, 0xbe, 0xdd] :: [Word8])
+        arr5 = mkByteArray ([0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xdd] :: [Word8])
     when (show arr1 /= "[0xde, 0xad, 0xbe, 0xef]") $
         fail $ "ByteArray Show incorrect: "++show arr1
     unless (arr1 > arr3) $
         fail $ "ByteArray Ord incorrect"
     unless (arr1 == arr2) $
         fail $ "ByteArray Eq incorrect"
+    unless (mappend arr1 arr4 == arr5) $
+        fail $ "ByteArray Monoid mappend incorrect"
+    unless (mappend arr1 (mappend arr3 arr4) == mappend (mappend arr1 arr3) arr4) $
+        fail $ "ByteArray Monoid mappend not associative"
+    unless (mconcat [arr1,arr2,arr3,arr4,arr5] == (arr1 <> arr2 <> arr3 <> arr4 <> arr5)) $
+        fail $ "ByteArray Monoid mconcat incorrect"
+#if MIN_VERSION_base(4,9,0)
+    unless (stimes (3 :: Int) arr4 == (arr4 <> arr4 <> arr4)) $
+        fail $ "ByteArray Semigroup stimes incorrect"
+#endif
 
 mkByteArray :: Prim a => [a] -> ByteArray
 mkByteArray xs = runST $ do
     marr <- newByteArray (length xs * sizeOf (head xs))
     sequence $ zipWith (writeByteArray marr) [0..] xs
     unsafeFreezeByteArray marr
+
+instance Arbitrary1 Array where
+  liftArbitrary elemGen = fmap fromList (QC.liftArbitrary elemGen)
+
+instance Arbitrary a => Arbitrary (Array a) where
+  arbitrary = fmap fromList QC.arbitrary
+
+instance Arbitrary1 SmallArray where
+  liftArbitrary elemGen = fmap smallArrayFromList (QC.liftArbitrary elemGen)
+
+instance Arbitrary a => Arbitrary (SmallArray a) where
+  arbitrary = fmap smallArrayFromList QC.arbitrary
+
+instance Arbitrary ByteArray where
+  arbitrary = do
+    xs <- QC.arbitrary :: Gen [Word8]
+    return $ runST $ do
+      a <- newByteArray (L.length xs)
+      iforM_ xs $ \ix x -> do
+        writeByteArray a ix x
+      unsafeFreezeByteArray a
+
+instance (Arbitrary a, Prim a) => Arbitrary (PrimArray a) where
+  arbitrary = do
+    xs <- QC.arbitrary :: Gen [a]
+    return $ runST $ do
+      a <- newPrimArray (L.length xs)
+      iforM_ xs $ \ix x -> do
+        writePrimArray a ix x
+      unsafeFreezePrimArray a
+
+instance (Arbitrary a, PrimUnlifted a) => Arbitrary (UnliftedArray a) where
+  arbitrary = do
+    xs <- QC.vector =<< QC.choose (0,3)
+    return (unliftedArrayFromList xs)
+
+instance (Prim a, CoArbitrary a) => CoArbitrary (PrimArray a) where
+  coarbitrary x = QC.coarbitrary (primArrayToList x)
+
+instance (Prim a, Function a) => Function (PrimArray a) where
+  function = QC.functionMap primArrayToList primArrayFromList
+
+iforM_ :: Monad m => [a] -> (Int -> a -> m b) -> m ()
+iforM_ xs0 f = go 0 xs0 where
+  go !_ [] = return ()
+  go !ix (x : xs) = f ix x >> go (ix + 1) xs
+
+newtype DefaultSetMethod = DefaultSetMethod Int16
+  deriving (Eq,Show,Arbitrary)
+
+instance Prim DefaultSetMethod where
+  sizeOf# _ = sizeOf# (undefined :: Int16)
+  alignment# _ = alignment# (undefined :: Int16)
+  indexByteArray# arr ix = DefaultSetMethod (indexByteArray# arr ix)
+  readByteArray# arr ix s0 = case readByteArray# arr ix s0 of
+    (# s1, n #) -> (# s1, DefaultSetMethod n #)
+  writeByteArray# arr ix (DefaultSetMethod n) s0 = writeByteArray# arr ix n s0
+  setByteArray# = defaultSetByteArray#
+  indexOffAddr# addr off = DefaultSetMethod (indexOffAddr# addr off)
+  readOffAddr# addr off s0 = case readOffAddr# addr off s0 of
+    (# s1, n #) -> (# s1, DefaultSetMethod n #)
+  writeOffAddr# addr off (DefaultSetMethod n) s0 = writeOffAddr# addr off n s0
+  setOffAddr# = defaultSetOffAddr#
+
+-- TODO: Uncomment this out when GHC 8.6 is release. Also, uncomment
+-- the corresponding PrimStorable test group above.
+--
+-- newtype Derived = Derived Int16
+--   deriving newtype (Prim)
+--   deriving Storable via (PrimStorable Derived)
+
+
+
diff --git a/test/primitive-tests.cabal b/test/primitive-tests.cabal
new file mode 100644
--- /dev/null
+++ b/test/primitive-tests.cabal
@@ -0,0 +1,45 @@
+Name:           primitive-tests
+Version:        0.1
+License:        BSD3
+License-File:   LICENSE
+
+Author:         Roman Leshchinskiy <rl@cse.unsw.edu.au>
+Maintainer:     libraries@haskell.org
+Copyright:      (c) Roman Leshchinskiy 2009-2012
+Homepage:       https://github.com/haskell/primitive
+Bug-Reports:    https://github.com/haskell/primitive/issues
+Category:       Data
+Synopsis:       primitive tests
+Cabal-Version:  >= 1.10
+Build-Type:     Simple
+Description:    @primitive@ tests
+
+Tested-With:
+  GHC == 7.4.2,
+  GHC == 7.6.3,
+  GHC == 7.8.4,
+  GHC == 7.10.3,
+  GHC == 8.0.2,
+  GHC == 8.2.2,
+  GHC == 8.4.2
+
+test-suite test
+  Default-Language: Haskell2010
+  hs-source-dirs: .
+  main-is: main.hs
+  type: exitcode-stdio-1.0
+  build-depends: base >= 4.5 && < 4.12
+               , ghc-prim
+               , primitive
+               , QuickCheck
+               , tasty
+               , tasty-quickcheck
+               , tagged
+               , transformers >= 0.3
+               , quickcheck-classes >= 0.4.11.1
+  ghc-options: -O2
+
+source-repository head
+  type:     git
+  location: https://github.com/haskell/primitive
+  subdir:   test