@@ -31,37 +31,57 @@ module Streamly.Internal.Data.MutArray.Type
3131 -- ** Type
3232 -- $arrayNotes
3333 MutArray (..)
34+
35+ -- ** Conversion
36+ -- *** Pinned and Unpinned
3437 , pin
3538 , unpin
3639 , isPinned
3740
38− -- -- * Constructing and Writing
41+ -- ** Casting
42+ , cast
43+ , castUnsafe -- XXX unsafeCast
44+ , asBytes
45+ , unsafePinnedAsPtr
46+ , unsafeAsPtr
47+
3948 -- ** Construction
40− , nil
49+ , empty
50+
51+ -- *** New
52+ -- | New arrays are always empty arrays with some reserve capacity to
53+ -- extend the length without reallocating.
54+ , emptyOf
55+ , newArrayWith -- emptyAlignedWith
56+ , pinnedEmptyOf
57+ , pinnedNewAligned -- XXX not required
58+ -- , new -- uninitialized array of specified length
59+
60+ -- *** Cloning
61+ , clone
62+ , pinnedClone
4163
42− -- *** Uninitialized Arrays
43− , pinnedNew
44− , pinnedNewBytes
45− , pinnedNewAligned
46− , new
47− , newArrayWith
64+ -- *** Slicing
65+ -- | Get a subarray without copying
66+ , getSliceUnsafe -- XXX unsafeGetSlice
67+ , getSlice
68+ , splitAt -- XXX should be able to express using getSlice
69+ , breakOn
4870
49− -- *** From streams
71+ -- *** Stream Folds
5072 , ArrayUnsafe (..)
51− , writeNWithUnsafe
52− , writeNWith
53− , writeNUnsafe
54− , pinnedWriteNUnsafe
55− , writeN
56− , pinnedWriteN
57− , pinnedWriteNAligned
58−
59− , writeWith
60− , write
61− , pinnedWrite
62−
63− , writeRevN
64− -- , writeRev
73+ , unsafeCreateOfWith
74+ , unsafeCreateOf
75+ , unsafePinnedCreateOf
76+ , pinnedCreateOf
77+ , createOfWith
78+ , createOf
79+ , revCreateOf
80+
81+ , pinnedCreate
82+ , createWith
83+ , create
84+ -- , revCreate
6585
6686 -- *** From containers
6787 , fromListN
@@ -77,11 +97,11 @@ module Streamly.Internal.Data.MutArray.Type
7797 , fromByteStr#
7898 , fromPtrN
7999 , fromChunksK
80− , fromChunksRealloced
100+ , fromChunksRealloced -- fromSmallChunks
81101
82102 -- ** Random writes
83103 , putIndex
84− , putIndexUnsafe
104+ , putIndexUnsafe -- XXX unsafePutIndex
85105 , putIndices
86106 -- , putFromThenTo
87107 -- , putFrom -- start writing at the given position
@@ -89,77 +109,70 @@ module Streamly.Internal.Data.MutArray.Type
89109 -- , putFromTo
90110 -- , putFromRev
91111 -- , putUptoRev
92− , modifyIndexUnsafe
112+ , modifyIndexUnsafe -- XXX unsafeModifyIndex
93113 , modifyIndex
94114 , modifyIndices
95115 , modify
96116 , swapIndices
97117 , unsafeSwapIndices
98118
99− -- ** Growing and Shrinking
100− -- | Arrays grow only at the end, though it is possible to grow on both sides
101− -- and therefore have a cons as well as snoc. But that will require two
102− -- bounds in the array representation.
103−
104− -- *** Appending elements
105− , snocWith
106− , snoc
107− , snocLinear
108− , snocMay
109− , snocUnsafe
110−
111− -- *** Appending streams
112− , writeAppendNUnsafe
113− , writeAppendN
114− , writeAppendWith
115− , writeAppend
119+ -- ** Reading
116120
117− -- ** Eliminating and Reading
118−
119− -- *** To streams
120− , reader
121− , readerRevWith
122− , readerRev
121+ -- *** Indexing
122+ , getIndex
123+ , getIndexUnsafe -- XXX unsafeGetIndex
124+ -- , getFromThenTo
125+ , getIndexRev -- getRevIndex?
126+ , indexReader
127+ , indexReaderWith
123128
124− -- *** To containers
125− , toStreamWith
126− , toStreamRevWith
127− , toStreamKWith
128− , toStreamKRevWith
129+ -- *** To Streams
129130 , read
130131 , readRev
132+ , toStreamWith
133+ , toStreamRevWith
131134 , toStreamK
135+ , toStreamKWith
132136 , toStreamKRev
137+ , toStreamKRevWith
138+
139+ -- *** To Containers
133140 , toList
134141
142+ -- *** Unfolds
135143 -- experimental
136144 , producerWith
137145 , producer
138146
139− -- *** Random reads
140− , getIndex
141− , getIndexUnsafe
142− , getIndices
143− , getIndicesWith
144− -- , getFromThenTo
145− , getIndexRev
146−
147− -- ** Memory Management
148− , blockSize
149− , arrayChunkBytes
150− , allocBytesToElemCount
151− , realloc
152− , resize
153− , resizeExp
154− , rightSize
147+ , reader
148+ , readerRevWith
149+ , readerRev
155150
156− -- ** Size
151+ -- ** Size and Capacity
152+ -- *** Size
157153 , length
158154 , byteLength
155+
156+ -- *** Capacity
159157 -- , capacity
160158 , byteCapacity
161159 , bytesFree
162160
161+ -- *** Capacity Management
162+ , blockSize
163+ , arrayChunkBytes
164+ , allocBytesToElemCount
165+ , realloc -- this is "resize"
166+ , grow
167+ , growExp
168+ , rightSize
169+
170+ -- ** Folding
171+ , foldl'
172+ , foldr
173+ , byteCmp
174+ , byteEq
175+
163176 -- ** In-place Mutation Algorithms
164177 , strip
165178 , reverse
@@ -170,59 +183,82 @@ module Streamly.Internal.Data.MutArray.Type
170183 , mergeBy
171184 , bubble
172185
173− -- ** Casting
174− , cast
175− , castUnsafe
176− , asBytes
177− , asPtrUnsafe
178− , asUnpinnedPtrUnsafe
186+ -- ** Growing and Shrinking
187+ -- | Arrays grow only at the end, though it is possible to grow on both
188+ -- sides and therefore have a cons as well as snoc. But that will require
189+ -- both lower and upper bound in the array representation.
179190
180− -- ** Folding
181− , foldl'
182− , foldr
183− , cmp
184− , byteCmp
185− , byteEq
191+ -- *** Appending elements
192+ , snocWith
193+ , snoc
194+ , snocLinear
195+ , snocMay
196+ , snocUnsafe -- XXX unsafeSnoc
197+
198+ -- *** Appending streams
199+ , unsafeAppendN
200+ , appendN
201+ , appendWith
202+ , append
203+
204+ -- *** Appending arrays
205+ , spliceCopy
206+ , spliceWith
207+ , splice
208+ , spliceExp
209+ , spliceUnsafe -- XXX unsafeSplice
210+ -- , putSlice
211+ -- , appendSlice
212+ -- , appendSliceFrom
213+
214+ -- ** Serialization using Unbox
215+ , pokeAppend
216+ , pokeAppendMay
217+ , pokeSkipUnsafe -- unsafePokeSkip
218+
219+ -- ** Deserialization using Unbox
220+ , peekUncons
221+ , peekUnconsUnsafe -- unsafePeekUncons
222+ , peekSkipUnsafe -- unsafePeekSkip
186223
187− -- ** Arrays of arrays
224+ -- Arrays of arrays
188225 -- We can add dimensionality parameter to the array type to get
189226 -- multidimensional arrays. Multidimensional arrays would just be a
190227 -- convenience wrapper on top of single dimensional arrays.
191228
192− -- | Operations dealing with multiple arrays, streams of arrays or
193− -- multidimensional array representations.
194−
195− -- *** Construct from streams
229+ -- ** Streams of Arrays
230+ -- *** Chunk
231+ -- | Group a stream into arrays.
196232 , chunksOf
197233 , pinnedChunksOf
198234 , buildChunks
199235
200− -- *** Eliminate to streams
201− , concatChunks
202− , concatChunksRev
236+ -- *** Split
237+ -- | Split an array into slices.
203238
204− -- *** Construct from arrays
205− -- get chunks without copying
206− , getSliceUnsafe
207− , getSlice
208239 -- , getSlicesFromLenN
209− , splitAt -- XXX should be able to express using getSlice
210240 , splitOn
211− , breakOn
212−
213− -- ** Cloning arrays
214− , clone
215− , pinnedClone
216−
217− -- ** Appending arrays
218− , spliceCopy
219− , spliceWith
220− , splice
221− , spliceExp
222− , spliceUnsafe
223− -- , putSlice
224− -- , appendSlice
225− -- , appendSliceFrom
241+ -- , slicesOf
242+
243+ -- *** Concat
244+ -- | Append the arrays in a stream to form a stream of elements.
245+ , concatWith
246+ , concatRevWith
247+ , concat
248+ , concatRev
249+
250+ -- *** Compact
251+ -- | Append the arrays in a stream to form a stream of larger arrays.
252+ , SpliceState (..)
253+ , pCompactLE
254+ , pPinnedCompactLE
255+ , compactLeAs
256+ , fCompactGE
257+ , fPinnedCompactGE
258+ , lCompactGE
259+ , lPinnedCompactGE
260+ , compactGE
261+ , compactEQ
226262
227263 -- ** Utilities
228264 , roundUpToPower2
@@ -231,12 +267,37 @@ module Streamly.Internal.Data.MutArray.Type
231267 , c_memchr
232268
233269 -- * Deprecated
270+ , asPtrUnsafe
234271 , writeChunks
235272 , flattenArrays
236273 , flattenArraysRev
237274 , fromArrayStreamK
238275 , fromStreamDN
239276 , fromStreamD
277+ , cmp
278+ , getIndices
279+ , getIndicesWith
280+ , resize
281+ , resizeExp
282+ , nil
283+ , new
284+ , pinnedNew
285+ , pinnedNewBytes
286+ , writeAppendNUnsafe
287+ , writeAppendN
288+ , writeAppendWith
289+ , writeAppend
290+ , writeNWithUnsafe
291+ , writeNWith
292+ , writeNUnsafe
293+ , pinnedWriteNUnsafe
294+ , writeN
295+ , pinnedWriteN
296+ , pinnedWriteNAligned -- XXX not required
297+ , writeWith
298+ , write
299+ , pinnedWrite
300+ , writeRevN
240301 )
241302where
242303
@@ -247,6 +308,7 @@ where
247308
248309import Control.Monad (when, void)
249310import Control.Monad.IO.Class (MonadIO(..))
311+import Data.Bifunctor (first)
250312import Data.Bits (shiftR, (.|.), (.&.))
251313import Data.Functor.Identity (Identity(..))
252314import Data.Proxy (Proxy(..))
@@ -273,13 +335,17 @@ import GHC.Ptr (Ptr(..))
273335import Streamly.Internal.Data.Fold.Type (Fold(..))
274336import Streamly.Internal.Data.Producer.Type (Producer (..))
275337import Streamly.Internal.Data.Stream.Type (Stream)
338+import Streamly.Internal.Data.Parser.Type (Parser (..))
276339import Streamly.Internal.Data.StreamK.Type (StreamK)
277340import Streamly.Internal.Data.SVar.Type (adaptState, defState)
341+import Streamly.Internal.Data.Tuple.Strict (Tuple'(..))
278342import Streamly.Internal.Data.Unfold.Type (Unfold(..))
279343import Streamly.Internal.System.IO (arrayPayloadSize, defaultChunkSize)
280344
281345import qualified Streamly.Internal.Data.Fold.Type as FL
282346import qualified Streamly.Internal.Data.MutByteArray.Type as Unboxed
347+import qualified Streamly.Internal.Data.Parser.Type as Parser
348+-- import qualified Streamly.Internal.Data.Fold.Type as Fold
283349import qualified Streamly.Internal.Data.Producer as Producer
284350import qualified Streamly.Internal.Data.Stream.Type as D
285351import qualified Streamly.Internal.Data.Stream.Lift as D
@@ -287,7 +353,7 @@ import qualified Streamly.Internal.Data.StreamK.Type as K
287353import qualified Prelude
288354
289355import Prelude hiding
290− (Foldable(..), read, unlines, splitAt, reverse, truncate)
356+ (Foldable(..), concat, read, unlines, splitAt, reverse, truncate)
291357
292358#include "DocTestDataMutArray.hs"
293359
@@ -413,6 +479,8 @@ isPinned MutArray{..} = Unboxed.isPinned arrContents
413479-- we need more than that. See stg_pinnedNewAlignedByteArrayzh and
414480-- allocatePinned in GHC source.
415481
482+-- XXX Rename to emptyAlignedWith, alignSize should be first arg.
483+
416484-- | @newArrayWith allocator alignment count@ allocates a new array of zero
417485-- length and with a capacity to hold @count@ elements, using @allocator
418486-- size alignment@ as the memory allocator function.
@@ -436,12 +504,25 @@ newArrayWith alloc alignSize count = do
436504 , arrBound = size
437505 }
438506
507+-- For arrays "nil" sounds a bit odd. empty is better. The only problem with
508+-- empty is that it is also used by the Alternative type class. But assuming we
509+-- will mostly import the Array module qualified this should be fine.
510+
511+-- | Create an empty array.
512+empty ::
513+#ifdef DEVBUILD
514+ Unbox a =>
515+#endif
516+ MutArray a
517+empty = MutArray Unboxed.empty 0 0 0
518+
519+{-# DEPRECATED nil "Please use empty instead." #-}
439520nil ::
440521#ifdef DEVBUILD
441522 Unbox a =>
442523#endif
443524 MutArray a
444−nil = MutArray Unboxed.nil 0 0 0
525+nil = empty
445526
446527{-# INLINE newBytesAs #-}
447528newBytesAs :: MonadIO m =>
@@ -458,12 +539,13 @@ newBytesAs ps bytes = do
458539 , arrBound = bytes
459540 }
460541
461−-- | Allocates a pinned empty array that can hold 'count' items. The memory of
462−-- the array is uninitialized and the allocation is aligned as per the
463−-- 'Unboxed' instance of the type.
542+-- | Allocates a pinned empty array that with a reserved capacity of bytes.
543+-- The memory of the array is uninitialized and the allocation is aligned as
544+-- per the 'Unboxed' instance of the type.
464545--
465546-- /Pre-release/
466547{-# INLINE pinnedNewBytes #-}
548+{-# DEPRECATED pinnedNewBytes "Please use pinnedEmptyOf with appropriate calculation" #-}
467549pinnedNewBytes :: MonadIO m =>
468550#ifdef DEVBUILD
469551 Unbox a =>
@@ -488,21 +570,31 @@ newAs ps =
488570 (error "new: alignment is not used in unpinned arrays.")
489571
490572-- XXX can unaligned allocation be more efficient when alignment is not needed?
491−--
492−-- | Allocates an empty pinned array that can hold 'count' items. The memory of
493−-- the array is uninitialized and the allocation is aligned as per the 'Unboxed'
494−-- instance of the type.
495−--
573+
574+-- | Allocates a pinned array of zero length but growable to the specified
575+-- capacity without reallocation.
576+{-# INLINE pinnedEmptyOf #-}
577+pinnedEmptyOf :: forall m a. (MonadIO m, Unbox a) => Int -> m (MutArray a)
578+pinnedEmptyOf = newAs Pinned
579+
580+-- XXX Deprecate in major
581+-- {-# DEPRECATED pinnedNew "Please use pinnedEmptyOf instead." #-}
496582{-# INLINE pinnedNew #-}
497583pinnedNew :: forall m a. (MonadIO m, Unbox a) => Int -> m (MutArray a)
498−pinnedNew = newAs Pinned
584+pinnedNew = pinnedEmptyOf
499585
500−-- | Allocates an empty unpinned array that can hold 'count' items. The memory
501−-- of the array is uninitialized.
586+-- | Allocates an unpinned array of zero length but growable to the specified
587+-- capacity without reallocation.
502588--
589+{-# INLINE emptyOf #-}
590+emptyOf :: (MonadIO m, Unbox a) => Int -> m (MutArray a)
591+emptyOf = newAs Unpinned
592+
593+-- XXX Deprecate in major
594+-- {-# DEPRECATED new "Please use emptyOf instead." #-}
503595{-# INLINE new #-}
504596new :: (MonadIO m, Unbox a) => Int -> m (MutArray a)
505−new = newAs Unpinned
597+new = emptyOf
506598
507599-------------------------------------------------------------------------------
508600-- Random writes
@@ -550,7 +642,7 @@ putIndices arr = FL.foldlM' step (return ())
550642
551643 where
552644
553− step () (i, x) = liftIO (putIndex i arr x)
645+ step () (i, x) = putIndex i arr x
554646
555647-- | Modify a given index of an array using a modifier function.
556648--
@@ -746,22 +838,19 @@ arrayChunkBytes = 1024
746838roundDownTo :: Int -> Int -> Int
747839roundDownTo elemSize size = size - (size `mod` elemSize)
748840
749−-- XXX See if resizing can be implemented by reading the old array as a stream
750−-- and then using writeN to the new array.
751−--
752841-- NOTE: we are passing elemSize explicitly to avoid an Unboxed constraint.
753842-- Since this is not inlined Unboxed consrraint leads to dictionary passing
754843-- which complicates some inspection tests.
755844--
756−{-# NOINLINE reallocExplicit #-}
757−reallocExplicit :: Int -> Int -> MutArray a -> IO (MutArray a)
758−reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
845+{-# NOINLINE reallocExplicitAs #-}
846+reallocExplicitAs :: PinnedState -> Int -> Int -> MutArray a -> IO (MutArray a)
847+reallocExplicitAs ps elemSize newCapacityInBytes MutArray{..} = do
759848 assertM(arrEnd <= arrBound)
760849
761850 -- Allocate new array
762851 let newCapMaxInBytes = roundUpLargeArray newCapacityInBytes
763852 contents <-
764− if Unboxed.isPinned arrContents
853+ if ps == Pinned
765854 then Unboxed.pinnedNew newCapMaxInBytes
766855 else Unboxed.new newCapMaxInBytes
767856 let !(MutByteArray mbarrFrom#) = arrContents
@@ -786,6 +875,10 @@ reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
786875 , arrBound = newCapInBytes
787876 }
788877
878+-- XXX Should these be called reallocBytes etc?
879+-- XXX We may also need reallocAs to allocate as pinned/unpinned explicitly. In
880+-- fact clone/pinnedClone can be implemented using reallocAs.
881+
789882-- | @realloc newCapacity array@ reallocates the array to the specified
790883-- capacity in bytes.
791884--
@@ -797,7 +890,12 @@ reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
797890-- If the original array is pinned, the newly allocated array is also pinned.
798891{-# INLINABLE realloc #-}
799892realloc :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m (MutArray a)
800−realloc bytes arr = liftIO $ reallocExplicit (SIZE_OF(a)) bytes arr
893+realloc bytes arr =
894+ let ps =
895+ if isPinned arr
896+ then Pinned
897+ else Unpinned
898+ in liftIO $ reallocExplicitAs ps (SIZE_OF(a)) bytes arr
801899
802900-- | @reallocWith label capSizer minIncrBytes array@. The label is used
803901-- in error messages and the capSizer is used to determine the capacity of the
@@ -820,14 +918,14 @@ reallocWith label capSizer minIncrBytes arr = do
820918 where
821919
822920 badSize newSize =
823− concat
921+ Prelude.concat
824922 [ label
825923 , ": new array size (in bytes) is less than required size "
826924 , show newSize
827925 , ". Please check the sizing function passed."
828926 ]
829927
830−-- | @resize newCapacity array@ changes the total capacity of the array so that
928+-- | @grow newCapacity array@ changes the total capacity of the array so that
831929-- it is enough to hold the specified number of elements. Nothing is done if
832930-- the specified capacity is less than the length of the array.
833931--
@@ -835,34 +933,46 @@ reallocWith label capSizer minIncrBytes arr = do
835933-- the block size (4K).
836934--
837935-- /Pre-release/
838−{-# INLINE resize #-}
839−resize :: forall m a. (MonadIO m, Unbox a) =>
936+{-# INLINE grow #-}
937+grow :: forall m a. (MonadIO m, Unbox a) =>
840938 Int -> MutArray a -> m (MutArray a)
841−resize nElems arr@MutArray{..} = do
939+grow nElems arr@MutArray{..} = do
842940 let req = SIZE_OF(a) * nElems
843− len = arrEnd - arrStart
844− if req < len
941+ cap = arrBound - arrStart
942+ if req < cap
845943 then return arr
846944 else realloc req arr
847945
848−-- | Like 'resize' but if the byte capacity is more than 'largeObjectThreshold'
849−-- then it is rounded up to the closest power of 2.
946+{-# DEPRECATED resize "Please use grow instead." #-}
947+{-# INLINE resize #-}
948+resize :: forall m a. (MonadIO m, Unbox a) =>
949+ Int -> MutArray a -> m (MutArray a)
950+resize = grow
951+
952+-- | Like 'grow' but if the requested byte capacity is more than
953+-- 'largeObjectThreshold' then it is rounded up to the closest power of 2.
850954--
851955-- /Pre-release/
852−{-# INLINE resizeExp #-}
853−resizeExp :: forall m a. (MonadIO m, Unbox a) =>
956+{-# INLINE growExp #-}
957+growExp :: forall m a. (MonadIO m, Unbox a) =>
854958 Int -> MutArray a -> m (MutArray a)
855−resizeExp nElems arr@MutArray{..} = do
959+growExp nElems arr@MutArray{..} = do
856960 let req = roundUpLargeArray (SIZE_OF(a) * nElems)
857961 req1 =
858962 if req > largeObjectThreshold
859963 then roundUpToPower2 req
860964 else req
861− len = arrEnd - arrStart
862− if req1 < len
965+ cap = arrBound - arrStart
966+ if req1 < cap
863967 then return arr
864968 else realloc req1 arr
865969
970+{-# DEPRECATED resizeExp "Please use growExp instead." #-}
971+{-# INLINE resizeExp #-}
972+resizeExp :: forall m a. (MonadIO m, Unbox a) =>
973+ Int -> MutArray a -> m (MutArray a)
974+resizeExp = growExp
975+
866976-- | Resize the allocated memory to drop any reserved free space at the end of
867977-- the array and reallocate it to reduce wastage.
868978--
@@ -927,7 +1037,7 @@ snocUnsafe arr@MutArray{..} = snocNewEnd (INDEX_NEXT(arrEnd,a)) arr
9271037{-# INLINE snocMay #-}
9281038snocMay :: forall m a. (MonadIO m, Unbox a) =>
9291039 MutArray a -> a -> m (Maybe (MutArray a))
930−snocMay arr@MutArray{..} x = liftIO $ do
1040+snocMay arr@MutArray{..} x = do
9311041 let newEnd = INDEX_NEXT(arrEnd,a)
9321042 if newEnd <= arrBound
9331043 then Just <$> snocNewEnd newEnd arr x
@@ -941,7 +1051,7 @@ snocWithRealloc :: forall m a. (MonadIO m, Unbox a) =>
9411051 -> a
9421052 -> m (MutArray a)
9431053snocWithRealloc sizer arr x = do
944− arr1 <- liftIO $ reallocWith "snocWith" sizer (SIZE_OF(a)) arr
1054+ arr1 <- reallocWith "snocWith" sizer (SIZE_OF(a)) arr
9451055 snocUnsafe arr1 x
9461056
9471057-- | @snocWith sizer arr elem@ mutates @arr@ to append @elem@. The length of
@@ -963,7 +1073,7 @@ snocWith :: forall m a. (MonadIO m, Unbox a) =>
9631073 -> MutArray a
9641074 -> a
9651075 -> m (MutArray a)
966−snocWith allocSize arr x = liftIO $ do
1076+snocWith allocSize arr x = do
9671077 let newEnd = INDEX_NEXT(arrEnd arr,a)
9681078 if newEnd <= arrBound arr
9691079 then snocNewEnd newEnd arr x
@@ -1007,6 +1117,126 @@ snoc = snocWith f
10071117 then oldSize * 2
10081118 else roundUpToPower2 oldSize * 2
10091119
1120+-------------------------------------------------------------------------------
1121+-- Serialization/Deserialization
1122+-------------------------------------------------------------------------------
1123+
1124+{-# INLINE pokeNewEnd #-}
1125+pokeNewEnd :: (MonadIO m, Unbox a) =>
1126+ Int -> MutArray Word8 -> a -> m (MutArray Word8)
1127+pokeNewEnd newEnd arr@MutArray{..} x = liftIO $ do
1128+ assert (newEnd <= arrBound) (return ())
1129+ liftIO $ pokeAt arrEnd arrContents x
1130+ return $ arr {arrEnd = newEnd}
1131+
1132+-- | Really really unsafe, unboxes a Haskell type and appends the resulting
1133+-- bytes to the byte array, may cause silent data corruption or if you are
1134+-- lucky a segfault if the array does not have enough space to append the
1135+-- element.
1136+--
1137+-- /Internal/
1138+{-# INLINE pokeAppendUnsafe #-}
1139+pokeAppendUnsafe :: forall m a. (MonadIO m, Unbox a) =>
1140+ MutArray Word8 -> a -> m (MutArray Word8)
1141+pokeAppendUnsafe arr@MutArray{..} = pokeNewEnd (arrEnd + SIZE_OF(a)) arr
1142+
1143+-- | Skip the specified number of bytes in the array. The data in the skipped
1144+-- region remains uninitialzed.
1145+{-# INLINE pokeSkipUnsafe #-}
1146+pokeSkipUnsafe :: Int -> MutArray Word8 -> MutArray Word8
1147+pokeSkipUnsafe n arr@MutArray{..} = do
1148+ let newEnd = arrEnd + n
1149+ in assert (newEnd <= arrBound) (arr {arrEnd = newEnd})
1150+
1151+-- | Like 'pokeAppend' but does not grow the array when pre-allocated array
1152+-- capacity becomes full.
1153+--
1154+-- /Internal/
1155+{-# INLINE pokeAppendMay #-}
1156+pokeAppendMay :: forall m a. (MonadIO m, Unbox a) =>
1157+ MutArray Word8 -> a -> m (Maybe (MutArray Word8))
1158+pokeAppendMay arr@MutArray{..} x = liftIO $ do
1159+ let newEnd = arrEnd + SIZE_OF(a)
1160+ if newEnd <= arrBound
1161+ then Just <$> pokeNewEnd newEnd arr x
1162+ else return Nothing
1163+
1164+{-# NOINLINE pokeWithRealloc #-}
1165+pokeWithRealloc :: forall m a. (MonadIO m, Unbox a) =>
1166+ (Int -> Int)
1167+ -> MutArray Word8
1168+ -> a
1169+ -> m (MutArray Word8)
1170+pokeWithRealloc sizer arr x = do
1171+ arr1 <- liftIO $ reallocWith "pokeWithRealloc" sizer (SIZE_OF(a)) arr
1172+ pokeAppendUnsafe arr1 x
1173+
1174+{-# INLINE pokeAppendWith #-}
1175+pokeAppendWith :: forall m a. (MonadIO m, Unbox a) =>
1176+ (Int -> Int)
1177+ -> MutArray Word8
1178+ -> a
1179+ -> m (MutArray Word8)
1180+pokeAppendWith allocSize arr x = liftIO $ do
1181+ let newEnd = arrEnd arr + SIZE_OF(a)
1182+ if newEnd <= arrBound arr
1183+ then pokeNewEnd newEnd arr x
1184+ else pokeWithRealloc allocSize arr x
1185+
1186+-- | Unbox a Haskell type and append the resulting bytes to a mutable byte
1187+-- array. The array is grown exponentially when more space is needed.
1188+--
1189+-- Definition:
1190+--
1191+-- >>> pokeAppend arr x = MutArray.castUnsafe <$> MutArray.snoc (MutArray.castUnsafe arr) x
1192+--
1193+{-# INLINE pokeAppend #-}
1194+pokeAppend :: forall m a. (MonadIO m, Unbox a) =>
1195+ MutArray Word8 -> a -> m (MutArray Word8)
1196+pokeAppend = pokeAppendWith f
1197+
1198+ where
1199+
1200+ f oldSize =
1201+ if isPower2 oldSize
1202+ then oldSize * 2
1203+ else roundUpToPower2 oldSize * 2
1204+
1205+-- | Really really unsafe, create a Haskell value from an unboxed byte array,
1206+-- does not check if the array is big enough, may return garbage or if you are
1207+-- lucky may cause a segfault.
1208+--
1209+-- /Internal/
1210+{-# INLINE peekUnconsUnsafe #-}
1211+peekUnconsUnsafe :: forall m a. (MonadIO m, Unbox a) =>
1212+ MutArray Word8 -> m (a, MutArray Word8)
1213+peekUnconsUnsafe MutArray{..} = do
1214+ let start1 = arrStart + SIZE_OF(a)
1215+ assert (start1 <= arrEnd) (return ())
1216+ liftIO $ do
1217+ r <- peekAt arrStart arrContents
1218+ return (r, MutArray arrContents start1 arrEnd arrBound)
1219+
1220+-- | Discard the specified number of bytes in the array.
1221+{-# INLINE peekSkipUnsafe #-}
1222+peekSkipUnsafe :: Int -> MutArray Word8 -> MutArray Word8
1223+peekSkipUnsafe n MutArray{..} =
1224+ let start1 = arrStart + n
1225+ in assert (start1 <= arrEnd) (MutArray arrContents start1 arrEnd arrBound)
1226+
1227+-- | Create a Haskell value from its unboxed representation from the head of a
1228+-- byte array, return the value and the remaining array.
1229+{-# INLINE peekUncons #-}
1230+peekUncons :: forall m a. (MonadIO m, Unbox a) =>
1231+ MutArray Word8 -> m (Maybe a, MutArray Word8)
1232+peekUncons arr@MutArray{..} = do
1233+ let start1 = arrStart + SIZE_OF(a)
1234+ if start1 > arrEnd
1235+ then return (Nothing, arr)
1236+ else liftIO $ do
1237+ r <- peekAt arrStart arrContents
1238+ return (Just r, MutArray arrContents start1 arrEnd arrBound)
1239+
10101240-------------------------------------------------------------------------------
10111241-- Random reads
10121242-------------------------------------------------------------------------------
@@ -1049,10 +1279,10 @@ getIndexRev i MutArray{..} = do
10491279data GetIndicesState contents start end st =
10501280 GetIndicesState contents start end st
10511281
1052−{-# INLINE getIndicesWith #-}
1053−getIndicesWith :: (Monad m, Unbox a) =>
1282+{-# INLINE indexReaderWith #-}
1283+indexReaderWith :: (Monad m, Unbox a) =>
10541284 (forall b. IO b -> m b) -> D.Stream m Int -> Unfold m (MutArray a) a
1055−getIndicesWith liftio (D.Stream stepi sti) = Unfold step inject
1285+indexReaderWith liftio (D.Stream stepi sti) = Unfold step inject
10561286
10571287 where
10581288
@@ -1071,14 +1301,26 @@ getIndicesWith liftio (D.Stream stepi sti) = Unfold step inject
10711301 D.Skip s -> return $ D.Skip (GetIndicesState contents start end s)
10721302 D.Stop -> return D.Stop
10731303
1304+{-# DEPRECATED getIndicesWith "Please use indexReaderWith instead." #-}
1305+{-# INLINE getIndicesWith #-}
1306+getIndicesWith :: (Monad m, Unbox a) =>
1307+ (forall b. IO b -> m b) -> D.Stream m Int -> Unfold m (MutArray a) a
1308+getIndicesWith = indexReaderWith
1309+
10741310-- | Given an unfold that generates array indices, read the elements on those
10751311-- indices from the supplied MutArray. An error is thrown if an index is out of
10761312-- bounds.
10771313--
10781314-- /Pre-release/
1315+{-# INLINE indexReader #-}
1316+indexReader :: (MonadIO m, Unbox a) => Stream m Int -> Unfold m (MutArray a) a
1317+indexReader = indexReaderWith liftIO
1318+
1319+-- XXX DO NOT REMOVE, change the signature to use Stream instead of unfold
1320+{-# DEPRECATED getIndices "Please use indexReader instead." #-}
10791321{-# INLINE getIndices #-}
10801322getIndices :: (MonadIO m, Unbox a) => Stream m Int -> Unfold m (MutArray a) a
1081−getIndices = getIndicesWith liftIO
1323+getIndices = indexReader
10821324
10831325-------------------------------------------------------------------------------
10841326-- Subarrays
@@ -1373,7 +1615,7 @@ chunksOfAs ps n (D.Stream step state) =
13731615--
13741616-- Same as the following but may be more efficient:
13751617--
1376−-- >>> chunksOf n = Stream.foldMany (MutArray.writeN n)
1618+-- >>> chunksOf n = Stream.foldMany (MutArray.createOf n)
13771619--
13781620-- /Pre-release/
13791621{-# INLINE_NORMAL chunksOf #-}
@@ -1382,14 +1624,14 @@ chunksOf :: forall m a. (MonadIO m, Unbox a)
13821624-- XXX the idiomatic implementation leads to large regression in the D.reverse'
13831625-- benchmark. It seems it has difficulty producing optimized code when
13841626-- converting to StreamK. Investigate GHC optimizations.
1385−-- chunksOf n = D.foldMany (writeN n)
1627+-- chunksOf n = D.foldMany (createOf n)
13861628chunksOf = chunksOfAs Unpinned
13871629
13881630-- | Like 'chunksOf' but creates pinned arrays.
13891631{-# INLINE_NORMAL pinnedChunksOf #-}
13901632pinnedChunksOf :: forall m a. (MonadIO m, Unbox a)
13911633 => Int -> D.Stream m a -> D.Stream m (MutArray a)
1392−-- pinnedChunksOf n = D.foldMany (pinnedWriteN n)
1634+-- pinnedChunksOf n = D.foldMany (pinnedCreateOf n)
13931635pinnedChunksOf = chunksOfAs Pinned
13941636
13951637-- | When we are buffering a stream of unknown size into an array we do not
@@ -1422,16 +1664,10 @@ data FlattenState s contents a =
14221664 OuterLoop s
14231665 | InnerLoop s contents !Int !Int
14241666
1425−-- | Use the "reader" unfold instead.
1426−--
1427−-- @concatChunks = unfoldMany reader@
1428−--
1429−-- We can try this if there are any fusion issues in the unfold.
1430−--
1431−{-# INLINE_NORMAL concatChunks #-}
1432−concatChunks :: forall m a. (MonadIO m, Unbox a)
1433− => D.Stream m (MutArray a) -> D.Stream m a
1434−concatChunks (D.Stream step state) = D.Stream step' (OuterLoop state)
1667+{-# INLINE_NORMAL concatWith #-}
1668+concatWith :: forall m a. (Monad m, Unbox a)
1669+ => (forall b. IO b -> m b) -> D.Stream m (MutArray a) -> D.Stream m a
1670+concatWith liftio (D.Stream step state) = D.Stream step' (OuterLoop state)
14351671
14361672 where
14371673
@@ -1448,24 +1684,30 @@ concatChunks (D.Stream step state) = D.Stream step' (OuterLoop state)
14481684 return $ D.Skip $ OuterLoop st
14491685
14501686 step' _ (InnerLoop st contents p end) = do
1451− x <- liftIO $ peekAt p contents
1687+ x <- liftio $ peekAt p contents
14521688 return $ D.Yield x (InnerLoop st contents (INDEX_NEXT(p,a)) end)
14531689
1454−{-# DEPRECATED flattenArrays "Please use \"unfoldMany reader\" instead." #-}
1455−flattenArrays :: forall m a. (MonadIO m, Unbox a)
1456− => D.Stream m (MutArray a) -> D.Stream m a
1457−flattenArrays = concatChunks
1458−
1459−-- | Use the "readerRev" unfold instead.
1690+-- | Use the "reader" unfold instead.
14601691--
1461−-- @concatChunks = unfoldMany readerRev@
1692+-- @concat = unfoldMany reader@
14621693--
14631694-- We can try this if there are any fusion issues in the unfold.
14641695--
1465−{-# INLINE_NORMAL concatChunksRev #-}
1466−concatChunksRev :: forall m a. (MonadIO m, Unbox a)
1696+{-# INLINE_NORMAL concat #-}
1697+concat :: forall m a. (MonadIO m, Unbox a)
1698+ => D.Stream m (MutArray a) -> D.Stream m a
1699+concat = concatWith liftIO
1700+
1701+{-# DEPRECATED flattenArrays "Please use \"unfoldMany reader\" instead." #-}
1702+{-# INLINE flattenArrays #-}
1703+flattenArrays :: forall m a. (MonadIO m, Unbox a)
14671704 => D.Stream m (MutArray a) -> D.Stream m a
1468−concatChunksRev (D.Stream step state) = D.Stream step' (OuterLoop state)
1705+flattenArrays = concat
1706+
1707+{-# INLINE_NORMAL concatRevWith #-}
1708+concatRevWith :: forall m a. (Monad m, Unbox a)
1709+ => (forall b. IO b -> m b) -> D.Stream m (MutArray a) -> D.Stream m a
1710+concatRevWith liftio (D.Stream step state) = D.Stream step' (OuterLoop state)
14691711
14701712 where
14711713
@@ -1483,14 +1725,26 @@ concatChunksRev (D.Stream step state) = D.Stream step' (OuterLoop state)
14831725 return $ D.Skip $ OuterLoop st
14841726
14851727 step' _ (InnerLoop st contents p start) = do
1486− x <- liftIO $ peekAt p contents
1728+ x <- liftio $ peekAt p contents
14871729 let cur = INDEX_PREV(p,a)
14881730 return $ D.Yield x (InnerLoop st contents cur start)
14891731
1732+-- | Use the "readerRev" unfold instead.
1733+--
1734+-- @concat = unfoldMany readerRev@
1735+--
1736+-- We can try this if there are any fusion issues in the unfold.
1737+--
1738+{-# INLINE_NORMAL concatRev #-}
1739+concatRev :: forall m a. (MonadIO m, Unbox a)
1740+ => D.Stream m (MutArray a) -> D.Stream m a
1741+concatRev = concatRevWith liftIO
1742+
14901743{-# DEPRECATED flattenArraysRev "Please use \"unfoldMany readerRev\" instead." #-}
1744+{-# INLINE flattenArraysRev #-}
14911745flattenArraysRev :: forall m a. (MonadIO m, Unbox a)
14921746 => D.Stream m (MutArray a) -> D.Stream m a
1493−flattenArraysRev = concatChunksRev
1747+flattenArraysRev = concatRev
14941748
14951749-------------------------------------------------------------------------------
14961750-- Unfolds
@@ -1719,7 +1973,7 @@ foldr f z arr = D.foldr f z $ read arr
17191973
17201974-- XXX Keep the bound intact to not lose any free space? Perf impact?
17211975
1722−-- | @writeAppendNUnsafe n arr@ appends up to @n@ input items to the supplied
1976+-- | @unsafeAppendN n arr@ appends up to @n@ input items to the supplied
17231977-- array.
17241978--
17251979-- Unsafe: Do not drive the fold beyond @n@ elements, it will lead to memory
@@ -1728,13 +1982,12 @@ foldr f z arr = D.foldr f z $ read arr
17281982-- Any free space left in the array after appending @n@ elements is lost.
17291983--
17301984-- /Internal/
1731−{-# INLINE_NORMAL writeAppendNUnsafe #-}
1732−writeAppendNUnsafe :: forall m a. (MonadIO m, Unbox a) =>
1985+{-# INLINE_NORMAL unsafeAppendN #-}
1986+unsafeAppendN :: forall m a. (MonadIO m, Unbox a) =>
17331987 Int
17341988 -> m (MutArray a)
17351989 -> Fold m a (MutArray a)
1736−writeAppendNUnsafe n action =
1737− fmap fromArrayUnsafe $ FL.foldlM' step initial
1990+unsafeAppendN n action = fmap fromArrayUnsafe $ FL.foldlM' step initial
17381991
17391992 where
17401993
@@ -1747,7 +2000,7 @@ writeAppendNUnsafe n action =
17472000 -- otherwise we lose that space.
17482001 arr1 <-
17492002 if free < needed
1750− then noinline reallocWith "writeAppendNUnsafeWith" (+ needed) needed arr
2003+ then noinline reallocWith "unsafeAppendN" (+ needed) needed arr
17512004 else return arr
17522005 return $ toArrayUnsafe arr1
17532006
@@ -1755,30 +2008,51 @@ writeAppendNUnsafe n action =
17552008 liftIO $ pokeAt end contents x
17562009 return $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
17572010
2011+{-# DEPRECATED writeAppendNUnsafe "Please use unsafeAppendN instead." #-}
2012+{-# INLINE writeAppendNUnsafe #-}
2013+writeAppendNUnsafe :: forall m a. (MonadIO m, Unbox a) =>
2014+ Int
2015+ -> m (MutArray a)
2016+ -> Fold m a (MutArray a)
2017+writeAppendNUnsafe = unsafeAppendN
2018+
17582019-- | Append @n@ elements to an existing array. Any free space left in the array
17592020-- after appending @n@ elements is lost.
17602021--
1761−-- >>> writeAppendN n initial = Fold.take n (MutArray.writeAppendNUnsafe n initial)
2022+-- >>> appendN n initial = Fold.take n (MutArray.unsafeAppendN n initial)
17622023--
1763−{-# INLINE_NORMAL writeAppendN #-}
2024+{-# INLINE_NORMAL appendN #-}
2025+appendN :: forall m a. (MonadIO m, Unbox a) =>
2026+ Int -> m (MutArray a) -> Fold m a (MutArray a)
2027+appendN n initial = FL.take n (unsafeAppendN n initial)
2028+
2029+-- XXX Deprecate in major
2030+-- {-# DEPRECATED writeAppendN "Please use appendN instead." #-}
2031+{-# INLINE writeAppendN #-}
17642032writeAppendN :: forall m a. (MonadIO m, Unbox a) =>
17652033 Int -> m (MutArray a) -> Fold m a (MutArray a)
1766−writeAppendN n initial = FL.take n (writeAppendNUnsafe n initial)
2034+writeAppendN = appendN
17672035
1768−-- | @writeAppendWith realloc action@ mutates the array generated by @action@ to
2036+-- | @appendWith realloc action@ mutates the array generated by @action@ to
17692037-- append the input stream. If there is no reserved space available in the
17702038-- array it is reallocated to a size in bytes determined by @realloc oldSize@,
17712039-- where @oldSize@ is the current size of the array in bytes.
17722040--
17732041-- Note that the returned array may be a mutated version of original array.
17742042--
1775−-- >>> writeAppendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
2043+-- >>> appendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
17762044--
17772045-- /Pre-release/
2046+{-# INLINE appendWith #-}
2047+appendWith :: forall m a. (MonadIO m, Unbox a) =>
2048+ (Int -> Int) -> m (MutArray a) -> Fold m a (MutArray a)
2049+appendWith sizer = FL.foldlM' (snocWith sizer)
2050+
2051+{-# DEPRECATED writeAppendWith "Please use appendWith instead." #-}
17782052{-# INLINE writeAppendWith #-}
17792053writeAppendWith :: forall m a. (MonadIO m, Unbox a) =>
17802054 (Int -> Int) -> m (MutArray a) -> Fold m a (MutArray a)
1781−writeAppendWith sizer = FL.foldlM' (snocWith sizer)
2055+writeAppendWith = appendWith
17822056
17832057-- | @append action@ mutates the array generated by @action@ to append the
17842058-- input stream. If there is no reserved space available in the array it is
@@ -1786,26 +2060,33 @@ writeAppendWith sizer = FL.foldlM' (snocWith sizer)
17862060--
17872061-- Note that the returned array may be a mutated version of original array.
17882062--
1789−-- >>> writeAppend = MutArray.writeAppendWith (* 2)
2063+-- >>> append = MutArray.appendWith (* 2)
17902064--
2065+{-# INLINE append #-}
2066+append :: forall m a. (MonadIO m, Unbox a) =>
2067+ m (MutArray a) -> Fold m a (MutArray a)
2068+append = appendWith (* 2)
2069+
2070+-- XXX Deprecate in major
2071+-- {-# DEPRECATED writeAppend "Please use append instead." #-}
17912072{-# INLINE writeAppend #-}
17922073writeAppend :: forall m a. (MonadIO m, Unbox a) =>
17932074 m (MutArray a) -> Fold m a (MutArray a)
1794−writeAppend = writeAppendWith (* 2)
2075+writeAppend = append
17952076
17962077-- XXX We can carry bound as well in the state to make sure we do not lose the
17972078-- remaining capacity. Need to check perf impact.
17982079--
1799−-- | Like 'writeNUnsafe' but takes a new array allocator @alloc size@ function
1800−-- as argument.
2080+-- | Like 'unsafeCreateOf' but takes a new array allocator @alloc size@
2081+-- function as argument.
18012082--
1802−-- >>> writeNWithUnsafe alloc n = MutArray.writeAppendNUnsafe (alloc n) n
2083+-- >>> unsafeCreateOfWith alloc n = MutArray.unsafeAppendN (alloc n) n
18032084--
18042085-- /Pre-release/
1805−{-# INLINE_NORMAL writeNWithUnsafe #-}
1806−writeNWithUnsafe :: forall m a. (MonadIO m, Unbox a)
2086+{-# INLINE_NORMAL unsafeCreateOfWith #-}
2087+unsafeCreateOfWith :: forall m a. (MonadIO m, Unbox a)
18072088 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
1808−writeNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
2089+unsafeCreateOfWith alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
18092090
18102091 where
18112092
@@ -1816,40 +2097,64 @@ writeNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
18162097 return
18172098 $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
18182099
2100+{-# DEPRECATED writeNWithUnsafe "Please use unsafeCreateOfWith instead." #-}
2101+{-# INLINE writeNWithUnsafe #-}
2102+writeNWithUnsafe :: forall m a. (MonadIO m, Unbox a)
2103+ => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
2104+writeNWithUnsafe = unsafeCreateOfWith
2105+
18192106{-# INLINE_NORMAL writeNUnsafeAs #-}
18202107writeNUnsafeAs :: forall m a. (MonadIO m, Unbox a)
18212108 => PinnedState -> Int -> Fold m a (MutArray a)
1822−writeNUnsafeAs ps = writeNWithUnsafe (newAs ps)
2109+writeNUnsafeAs ps = unsafeCreateOfWith (newAs ps)
18232110
1824−-- | Like 'writeN' but does not check the array bounds when writing. The fold
2111+-- | Like 'createOf' but does not check the array bounds when writing. The fold
18252112-- driver must not call the step function more than 'n' times otherwise it will
18262113-- corrupt the memory and crash. This function exists mainly because any
18272114-- conditional in the step function blocks fusion causing 10x performance
18282115-- slowdown.
18292116--
1830−-- >>> writeNUnsafe = MutArray.writeNWithUnsafe MutArray.new
2117+-- >>> unsafeCreateOf = MutArray.unsafeCreateOfWith MutArray.emptyOf
18312118--
1832−{-# INLINE_NORMAL writeNUnsafe #-}
2119+{-# INLINE_NORMAL unsafeCreateOf #-}
2120+unsafeCreateOf :: forall m a. (MonadIO m, Unbox a)
2121+ => Int -> Fold m a (MutArray a)
2122+unsafeCreateOf = writeNUnsafeAs Unpinned
2123+
2124+{-# DEPRECATED writeNUnsafe "Please use unsafeCreateOf instead." #-}
2125+{-# INLINE writeNUnsafe #-}
18332126writeNUnsafe :: forall m a. (MonadIO m, Unbox a)
18342127 => Int -> Fold m a (MutArray a)
1835−writeNUnsafe = writeNUnsafeAs Unpinned
2128+writeNUnsafe = unsafeCreateOf
2129+
2130+-- | Like 'unsafeCreateOf' but creates a pinned array.
2131+{-# INLINE_NORMAL unsafePinnedCreateOf #-}
2132+unsafePinnedCreateOf :: forall m a. (MonadIO m, Unbox a)
2133+ => Int -> Fold m a (MutArray a)
2134+unsafePinnedCreateOf = writeNUnsafeAs Pinned
18362135
1837−-- | Like 'writeNUnsafe' but creates a pinned array.
1838−{-# INLINE_NORMAL pinnedWriteNUnsafe #-}
2136+{-# DEPRECATED pinnedWriteNUnsafe "Please use unsafePinnedCreateOf instead." #-}
2137+{-# INLINE pinnedWriteNUnsafe #-}
18392138pinnedWriteNUnsafe :: forall m a. (MonadIO m, Unbox a)
18402139 => Int -> Fold m a (MutArray a)
1841−pinnedWriteNUnsafe = writeNUnsafeAs Pinned
2140+pinnedWriteNUnsafe = unsafePinnedCreateOf
18422141
1843−-- | @writeNWith alloc n@ folds a maximum of @n@ elements into an array
2142+-- | @createOfWith alloc n@ folds a maximum of @n@ elements into an array
18442143-- allocated using the @alloc@ function.
18452144--
1846−-- >>> writeNWith alloc n = Fold.take n (MutArray.writeNWithUnsafe alloc n)
1847−-- >>> writeNWith alloc n = MutArray.writeAppendN (alloc n) n
2145+-- >>> createOfWith alloc n = Fold.take n (MutArray.unsafeCreateOfWith alloc n)
2146+-- >>> createOfWith alloc n = MutArray.appendN (alloc n) n
18482147--
1849−{-# INLINE_NORMAL writeNWith #-}
2148+{-# INLINE_NORMAL createOfWith #-}
2149+createOfWith :: forall m a. (MonadIO m, Unbox a)
2150+ => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
2151+createOfWith alloc n = FL.take n (unsafeCreateOfWith alloc n)
2152+
2153+-- {-# DEPRECATED writeNWith "Please use createOfWith instead." #-}
2154+{-# INLINE writeNWith #-}
18502155writeNWith :: forall m a. (MonadIO m, Unbox a)
18512156 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
1852−writeNWith alloc n = FL.take n (writeNWithUnsafe alloc n)
2157+writeNWith = createOfWith
18532158
18542159{-# INLINE_NORMAL writeNAs #-}
18552160writeNAs ::
@@ -1857,28 +2162,42 @@ writeNAs ::
18572162 => PinnedState
18582163 -> Int
18592164 -> Fold m a (MutArray a)
1860−writeNAs ps = writeNWith (newAs ps)
2165+writeNAs ps = createOfWith (newAs ps)
18612166
1862−-- | @writeN n@ folds a maximum of @n@ elements from the input stream to an
2167+-- | @createOf n@ folds a maximum of @n@ elements from the input stream to an
18632168-- 'MutArray'.
18642169--
1865−-- >>> writeN = MutArray.writeNWith MutArray.new
1866−-- >>> writeN n = Fold.take n (MutArray.writeNUnsafe n)
1867−-- >>> writeN n = MutArray.writeAppendN n (MutArray.new n)
2170+-- >>> createOf = MutArray.createOfWith MutArray.new
2171+-- >>> createOf n = Fold.take n (MutArray.unsafeCreateOf n)
2172+-- >>> createOf n = MutArray.appendN n (MutArray.emptyOf n)
18682173--
1869−{-# INLINE_NORMAL writeN #-}
2174+{-# INLINE_NORMAL createOf #-}
2175+createOf :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
2176+createOf = writeNAs Unpinned
2177+
2178+-- XXX Deprecate in major
2179+-- {-# DEPRECATED writeN "Please use createOf instead." #-}
2180+{-# INLINE writeN #-}
18702181writeN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
1871−writeN = writeNAs Unpinned
2182+writeN = createOf
2183+
2184+-- | Like 'createOf' but creates a pinned array.
2185+{-# INLINE_NORMAL pinnedCreateOf #-}
2186+pinnedCreateOf ::
2187+ forall m a. (MonadIO m, Unbox a)
2188+ => Int
2189+ -> Fold m a (MutArray a)
2190+pinnedCreateOf = writeNAs Pinned
18722191
1873−-- | Like 'writeN' but creates a pinned array.
1874−{-# INLINE_NORMAL pinnedWriteN #-}
2192+{-# DEPRECATED pinnedWriteN "Please use pinnedCreateOf instead." #-}
2193+{-# INLINE pinnedWriteN #-}
18752194pinnedWriteN ::
18762195 forall m a. (MonadIO m, Unbox a)
18772196 => Int
18782197 -> Fold m a (MutArray a)
1879−pinnedWriteN = writeNAs Pinned
2198+pinnedWriteN = pinnedCreateOf
18802199
1881−-- | Like writeNWithUnsafe but writes the array in reverse order.
2200+-- | Like unsafeCreateOfWith but writes the array in reverse order.
18822201--
18832202-- /Internal/
18842203{-# INLINE_NORMAL writeRevNWithUnsafe #-}
@@ -1899,7 +2218,7 @@ writeRevNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
18992218 return
19002219 $ ArrayUnsafe contents ptr end
19012220
1902−-- | Like writeNWith but writes the array in reverse order.
2221+-- | Like createOfWith but writes the array in reverse order.
19032222--
19042223-- /Internal/
19052224{-# INLINE_NORMAL writeRevNWith #-}
@@ -1907,31 +2226,36 @@ writeRevNWith :: forall m a. (MonadIO m, Unbox a)
19072226 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
19082227writeRevNWith alloc n = FL.take n (writeRevNWithUnsafe alloc n)
19092228
1910−-- | Like writeN but writes the array in reverse order.
2229+-- | Like 'createOf' but writes the array in reverse order.
19112230--
19122231-- /Pre-release/
1913−{-# INLINE_NORMAL writeRevN #-}
2232+{-# INLINE_NORMAL revCreateOf #-}
2233+revCreateOf :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
2234+revCreateOf = writeRevNWith new
2235+
2236+{-# DEPRECATED writeRevN "Please use revCreateOf instead." #-}
2237+{-# INLINE writeRevN #-}
19142238writeRevN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
1915−writeRevN = writeRevNWith new
2239+writeRevN = revCreateOf
19162240
19172241-- | @pinnedWriteNAligned align n@ folds a maximum of @n@ elements from the
19182242-- input stream to a 'MutArray' aligned to the given size.
19192243--
1920−-- >>> pinnedWriteNAligned align = MutArray.writeNWith (MutArray.pinnedNewAligned align)
1921−-- >>> pinnedWriteNAligned align n = MutArray.writeAppendN n (MutArray.pinnedNewAligned align n)
2244+-- >>> pinnedWriteNAligned align = MutArray.createOfWith (MutArray.pinnedNewAligned align)
2245+-- >>> pinnedWriteNAligned align n = MutArray.appendN n (MutArray.pinnedNewAligned align n)
19222246--
19232247-- /Pre-release/
19242248--
19252249{-# INLINE_NORMAL pinnedWriteNAligned #-}
19262250pinnedWriteNAligned :: forall m a. (MonadIO m, Unbox a)
19272251 => Int -> Int -> Fold m a (MutArray a)
1928−pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
2252+pinnedWriteNAligned align = createOfWith (pinnedNewAligned align)
19292253
19302254-- XXX Buffer to a list instead?
19312255
19322256-- | Buffer a stream into a stream of arrays.
19332257--
1934−-- >>> buildChunks n = Fold.many (MutArray.writeN n) Fold.toStreamK
2258+-- >>> buildChunks n = Fold.many (MutArray.createOf n) Fold.toStreamK
19352259--
19362260-- Breaking an array into an array stream can be useful to consume a large
19372261-- array sequentially such that memory of the array is released incrementatlly.
@@ -1943,9 +2267,10 @@ pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
19432267{-# INLINE_NORMAL buildChunks #-}
19442268buildChunks :: (MonadIO m, Unbox a) =>
19452269 Int -> Fold m a (StreamK n (MutArray a))
1946−buildChunks n = FL.many (writeN n) FL.toStreamK
2270+buildChunks n = FL.many (createOf n) FL.toStreamK
19472271
19482272{-# DEPRECATED writeChunks "Please use buildChunks instead." #-}
2273+{-# INLINE writeChunks #-}
19492274writeChunks :: (MonadIO m, Unbox a) =>
19502275 Int -> Fold m a (StreamK n (MutArray a))
19512276writeChunks = buildChunks
@@ -1953,29 +2278,29 @@ writeChunks = buildChunks
19532278{-# INLINE_NORMAL writeWithAs #-}
19542279writeWithAs :: forall m a. (MonadIO m, Unbox a)
19552280 => PinnedState -> Int -> Fold m a (MutArray a)
1956−-- writeWithAs ps n = FL.rmapM rightSize $ writeAppendWith (* 2) (newAs ps n)
2281+-- writeWithAs ps n = FL.rmapM rightSize $ appendWith (* 2) (newAs ps n)
19572282writeWithAs ps elemCount =
19582283 FL.rmapM extract $ FL.foldlM' step initial
19592284
19602285 where
19612286
19622287 initial = do
1963− when (elemCount < 0) $ error "writeWith: elemCount is negative"
1964− liftIO $ newAs ps elemCount
2288+ when (elemCount < 0) $ error "createWith: elemCount is negative"
2289+ newAs ps elemCount
19652290
19662291 step arr@(MutArray _ start end bound) x
19672292 | INDEX_NEXT(end,a) > bound = do
19682293 let oldSize = end - start
19692294 newSize = max (oldSize * 2) 1
1970− arr1 <- liftIO $ reallocExplicit (SIZE_OF(a)) newSize arr
2295+ arr1 <- liftIO $ reallocExplicitAs ps (SIZE_OF(a)) newSize arr
19712296 snocUnsafe arr1 x
19722297 step arr x = snocUnsafe arr x
19732298
19742299 extract = liftIO . rightSize
19752300
1976−-- XXX Compare writeWith with fromStreamD which uses an array of streams
2301+-- XXX Compare createWith with fromStreamD which uses an array of streams
19772302-- implementation. We can write this using buildChunks above if that is faster.
1978−-- If writeWith is faster then we should use that to implement
2303+-- If createWith is faster then we should use that to implement
19792304-- fromStreamD.
19802305--
19812306-- XXX The realloc based implementation needs to make one extra copy if we use
@@ -1990,39 +2315,56 @@ writeWithAs ps elemCount =
19902315-- XXX check if GHC's memory allocator is efficient enough. We can try the C
19912316-- malloc to compare against.
19922317
1993−-- | @writeWith minCount@ folds the whole input to a single array. The array
2318+-- | @createWith minCount@ folds the whole input to a single array. The array
19942319-- starts at a size big enough to hold minCount elements, the size is doubled
19952320-- every time the array needs to be grown.
19962321--
19972322-- /Caution! Do not use this on infinite streams./
19982323--
1999−-- >>> f n = MutArray.writeAppendWith (* 2) (MutArray.new n)
2000−-- >>> writeWith n = Fold.rmapM MutArray.rightSize (f n)
2001−-- >>> writeWith n = Fold.rmapM MutArray.fromChunksK (MutArray.buildChunks n)
2324+-- >>> f n = MutArray.appendWith (* 2) (MutArray.emptyOf n)
2325+-- >>> createWith n = Fold.rmapM MutArray.rightSize (f n)
2326+-- >>> createWith n = Fold.rmapM MutArray.fromChunksK (MutArray.buildChunks n)
20022327--
20032328-- /Pre-release/
2004−{-# INLINE_NORMAL writeWith #-}
2329+{-# INLINE_NORMAL createWith #-}
2330+createWith :: forall m a. (MonadIO m, Unbox a)
2331+ => Int -> Fold m a (MutArray a)
2332+-- createWith n = FL.rmapM rightSize $ appendWith (* 2) (emptyOf n)
2333+createWith = writeWithAs Unpinned
2334+
2335+{-# DEPRECATED writeWith "Please use createWith instead." #-}
2336+{-# INLINE writeWith #-}
20052337writeWith :: forall m a. (MonadIO m, Unbox a)
20062338 => Int -> Fold m a (MutArray a)
2007−-- writeWith n = FL.rmapM rightSize $ writeAppendWith (* 2) (new n)
2008−writeWith = writeWithAs Unpinned
2339+writeWith = createWith
20092340
20102341-- | Fold the whole input to a single array.
20112342--
2012−-- Same as 'writeWith' using an initial array size of 'arrayChunkBytes' bytes
2343+-- Same as 'createWith' using an initial array size of 'arrayChunkBytes' bytes
20132344-- rounded up to the element size.
20142345--
20152346-- /Caution! Do not use this on infinite streams./
20162347--
2348+{-# INLINE create #-}
2349+create :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2350+create = createWith (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2351+
2352+-- XXX Deprecate in major
2353+-- {-# DEPRECATED write "Please use create instead." #-}
20172354{-# INLINE write #-}
20182355write :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2019−write = writeWith (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2356+write = create
20202357
2021−-- | Like 'write' but creates a pinned array.
2358+-- | Like 'create' but creates a pinned array.
2359+{-# INLINE pinnedCreate #-}
2360+pinnedCreate :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2361+pinnedCreate =
2362+ writeWithAs Pinned (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2363+
2364+{-# DEPRECATED pinnedWrite "Please use pinnedCreate instead." #-}
20222365{-# INLINE pinnedWrite #-}
20232366pinnedWrite :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2024−pinnedWrite =
2025− writeWithAs Pinned (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2367+pinnedWrite = pinnedCreate
20262368
20272369-------------------------------------------------------------------------------
20282370-- construct from streams, known size
@@ -2032,7 +2374,7 @@ pinnedWrite =
20322374fromStreamDNAs :: forall m a. (MonadIO m, Unbox a)
20332375 => PinnedState -> Int -> D.Stream m a -> m (MutArray a)
20342376fromStreamDNAs ps limit str = do
2035− (arr :: MutArray a) <- liftIO $ newAs ps limit
2377+ (arr :: MutArray a) <- newAs ps limit
20362378 end <- D.foldlM' (fwrite (arrContents arr)) (return $ arrEnd arr) $ D.take limit str
20372379 return $ arr {arrEnd = end}
20382380
@@ -2042,17 +2384,18 @@ fromStreamDNAs ps limit str = do
20422384 liftIO $ pokeAt ptr arrContents x
20432385 return $ INDEX_NEXT(ptr,a)
20442386
2045−-- | Use the 'writeN' fold instead.
2387+-- | Use the 'createOf' fold instead.
20462388--
2047−-- >>> fromStreamN n = Stream.fold (MutArray.writeN n)
2389+-- >>> fromStreamN n = Stream.fold (MutArray.createOf n)
20482390--
20492391{-# INLINE_NORMAL fromStreamN #-}
20502392fromStreamN :: forall m a. (MonadIO m, Unbox a)
20512393 => Int -> D.Stream m a -> m (MutArray a)
2052−-- fromStreamDN n = D.fold (writeN n)
2394+-- fromStreamDN n = D.fold (createOf n)
20532395fromStreamN = fromStreamDNAs Unpinned
20542396
20552397{-# DEPRECATED fromStreamDN "Please use fromStreamN instead." #-}
2398+{-# INLINE fromStreamDN #-}
20562399fromStreamDN :: forall m a. (MonadIO m, Unbox a)
20572400 => Int -> D.Stream m a -> m (MutArray a)
20582401fromStreamDN = fromStreamN
@@ -2075,20 +2418,20 @@ pinnedFromListN n xs = fromStreamDNAs Pinned n $ D.fromList xs
20752418-- /Pre-release/
20762419{-# INLINE fromListRevN #-}
20772420fromListRevN :: (MonadIO m, Unbox a) => Int -> [a] -> m (MutArray a)
2078−fromListRevN n xs = D.fold (writeRevN n) $ D.fromList xs
2421+fromListRevN n xs = D.fold (revCreateOf n) $ D.fromList xs
20792422
20802423-- | Convert a pure stream in Identity monad to a mutable array.
20812424{-# INLINABLE fromPureStreamN #-}
20822425fromPureStreamN :: (MonadIO m, Unbox a) =>
20832426 Int -> Stream Identity a -> m (MutArray a)
20842427fromPureStreamN n xs =
2085− liftIO $ D.fold (writeN n) $ D.morphInner (return . runIdentity) xs
2428+ D.fold (createOf n) $ D.morphInner (return . runIdentity) xs
20862429
20872430-- | Convert a pure stream in Identity monad to a mutable array.
20882431{-# INLINABLE fromPureStream #-}
20892432fromPureStream :: (MonadIO m, Unbox a) => Stream Identity a -> m (MutArray a)
20902433fromPureStream xs =
2091− liftIO $ D.fold write $ D.morphInner (return . runIdentity) xs
2434+ D.fold create $ D.morphInner (return . runIdentity) xs
20922435
20932436{-# INLINABLE fromPtrN #-}
20942437fromPtrN :: MonadIO m => Int -> Ptr Word8 -> m (MutArray Word8)
@@ -2097,7 +2440,7 @@ fromPtrN len addr = do
20972440 -- XXX We can implement a stream copy in a similar way by streaming Word64
20982441 -- first and then remaining Word8.
20992442 arr <- new len
2100− _ <- asUnpinnedPtrUnsafe arr
2443+ _ <- unsafeAsPtr arr
21012444 (\ptr -> liftIO $ c_memcpy ptr addr (fromIntegral len))
21022445 return (arr {arrEnd = len})
21032446
@@ -2113,7 +2456,7 @@ fromByteStr# addr = do
21132456 len <- liftIO $ c_strlen (Ptr addr)
21142457 let lenInt = fromIntegral len
21152458 arr <- new lenInt
2116− _ <- asUnpinnedPtrUnsafe arr (\ptr -> liftIO $ c_memcpy ptr (Ptr addr) len)
2459+ _ <- unsafeAsPtr arr (\ptr -> liftIO $ c_memcpy ptr (Ptr addr) len)
21172460 return (arr {arrEnd = lenInt})
21182461
21192462-------------------------------------------------------------------------------
@@ -2134,7 +2477,7 @@ fromChunksRealloced s = do
21342477 arr <- D.foldlM' spliceExp (pure a) strm
21352478 -- Reallocation is exponential so there may be 50% empty space in
21362479 -- worst case. One more reallocation to reclaim the space.
2137− liftIO $ rightSize arr
2480+ rightSize arr
21382481 Nothing -> pure nil
21392482
21402483-------------------------------------------------------------------------------
@@ -2172,6 +2515,7 @@ fromChunksK :: (Unbox a, MonadIO m) =>
21722515fromChunksK = fromChunkskAs Unpinned
21732516
21742517{-# DEPRECATED fromArrayStreamK "Please use fromChunksK instead." #-}
2518+{-# INLINE fromArrayStreamK #-}
21752519fromArrayStreamK :: (Unbox a, MonadIO m) =>
21762520 StreamK m (MutArray a) -> m (MutArray a)
21772521fromArrayStreamK = fromChunksK
@@ -2183,7 +2527,7 @@ fromStreamDAs ps m =
21832527 arrayStreamKFromStreamDAs Unpinned m >>= fromChunkskAs ps
21842528
21852529-- | Create an 'Array' from a stream. This is useful when we want to create a
2186−-- single array from a stream of unknown size. 'writeN' is at least twice
2530+-- single array from a stream of unknown size. 'createOf' is at least twice
21872531-- as efficient when the size is already known.
21882532--
21892533-- Note that if the input stream is too large memory allocation for the array
@@ -2195,19 +2539,19 @@ fromStreamDAs ps m =
21952539fromStream :: (MonadIO m, Unbox a) => Stream m a -> m (MutArray a)
21962540fromStream = fromStreamDAs Unpinned
21972541
2198−-- fromStream (Stream m) = P.fold write m
2542+-- fromStream (Stream m) = P.fold create m
21992543-- CAUTION: a very large number (millions) of arrays can degrade performance
22002544-- due to GC overhead because we need to buffer the arrays before we flatten
22012545-- all the arrays.
22022546--
2203−-- XXX Compare if this is faster or "fold write".
2547+-- XXX Compare if this is faster or "fold create".
22042548--
22052549-- | We could take the approach of doubling the memory allocation on each
22062550-- overflow. This would result in more or less the same amount of copying as in
22072551-- the chunking approach. However, if we have to shrink in the end then it may
22082552-- result in an extra copy of the entire data.
22092553--
2210−-- >>> fromStreamD = StreamD.fold MutArray.write
2554+-- >>> fromStreamD = StreamD.fold MutArray.create
22112555--
22122556{-# INLINE fromStreamD #-}
22132557{-# DEPRECATED fromStreamD "Please use fromStream instead." #-}
@@ -2237,6 +2581,10 @@ fromListRev xs = fromListRevN (Prelude.length xs) xs
22372581-- Cloning
22382582-------------------------------------------------------------------------------
22392583
2584+-- Arrays are aligned on 64-bit boundaries. The fastest way to copy an array is
2585+-- to unsafeCast it to Word64, read it, write it to Word64 array and unsafeCast
2586+-- it again. We can use SIMD read/write as well.
2587+
22402588{-# INLINE cloneAs #-}
22412589cloneAs ::
22422590 ( MonadIO m
@@ -2246,7 +2594,7 @@ cloneAs ::
22462594 )
22472595 => PinnedState -> MutArray a -> m (MutArray a)
22482596cloneAs ps src =
2249− liftIO $ do
2597+ do
22502598 let startSrc = arrStart src
22512599 srcLen = arrEnd src - startSrc
22522600 newArrContents <-
@@ -2285,17 +2633,16 @@ spliceCopy :: forall m a. MonadIO m =>
22852633 Unbox a =>
22862634#endif
22872635 MutArray a -> MutArray a -> m (MutArray a)
2288−spliceCopy arr1 arr2 = liftIO $ do
2636+spliceCopy arr1 arr2 = do
22892637 let start1 = arrStart arr1
22902638 start2 = arrStart arr2
22912639 len1 = arrEnd arr1 - start1
22922640 len2 = arrEnd arr2 - start2
2293− let newLen = len1 + len2
2641+ let len = len1 + len2
22942642 newArrContents <-
22952643 if Unboxed.isPinned (arrContents arr1)
2296− then Unboxed.pinnedNew newLen
2297− else Unboxed.new newLen
2298− let len = len1 + len2
2644+ then liftIO $ Unboxed.pinnedNew len
2645+ else liftIO $ Unboxed.new len
22992646 putSliceUnsafe (arrContents arr1) start1 newArrContents 0 len1
23002647 putSliceUnsafe (arrContents arr2) start2 newArrContents len1 len2
23012648 return $ MutArray newArrContents 0 len len
@@ -2307,7 +2654,7 @@ spliceCopy arr1 arr2 = liftIO $ do
23072654spliceUnsafe :: MonadIO m =>
23082655 MutArray a -> MutArray a -> m (MutArray a)
23092656spliceUnsafe dst src =
2310− liftIO $ do
2657+ do
23112658 let startSrc = arrStart src
23122659 srcLen = arrEnd src - startSrc
23132660 endDst = arrEnd dst
@@ -2329,7 +2676,7 @@ spliceWith :: forall m a. (MonadIO m, Unbox a) =>
23292676 (Int -> Int -> Int) -> MutArray a -> MutArray a -> m (MutArray a)
23302677spliceWith sizer dst@(MutArray _ start end bound) src = do
23312678{-
2332− let f = writeAppendWith (`sizer` byteLength src) (return dst)
2679+ let f = appendWith (`sizer` byteLength src) (return dst)
23332680 in D.fold f (toStreamD src)
23342681-}
23352682 assert (end <= bound) (return ())
@@ -2345,7 +2692,7 @@ spliceWith sizer dst@(MutArray _ start end bound) src = do
23452692 $ "splice: newSize is less than the total size "
23462693 ++ "of arrays being appended. Please check the "
23472694 ++ "sizer function passed."
2348− liftIO $ realloc newSizeInBytes dst
2695+ realloc newSizeInBytes dst
23492696 else return dst
23502697 spliceUnsafe dst1 src
23512698
@@ -2397,7 +2744,7 @@ splitOn predicate arr =
23972744{-# INLINE breakOn #-}
23982745breakOn :: MonadIO m
23992746 => Word8 -> MutArray Word8 -> m (MutArray Word8, Maybe (MutArray Word8))
2400−breakOn sep arr@MutArray{..} = asUnpinnedPtrUnsafe arr $ \p -> liftIO $ do
2747+breakOn sep arr@MutArray{..} = unsafeAsPtr arr $ \p -> liftIO $ do
24012748 -- XXX We do not need memchr here, we can use a Haskell equivalent.
24022749 -- Need efficient stream based primitives that work on Word64.
24032750 loc <- c_memchr p sep (fromIntegral $ byteLength arr)
@@ -2420,33 +2767,41 @@ breakOn sep arr@MutArray{..} = asUnpinnedPtrUnsafe arr $ \p -> liftIO $ do
24202767 }
24212768 )
24222769
2770+-- | Like 'splitAt' but does not check whether the index is valid.
2771+--
2772+{-# INLINE unsafeSplitAt #-}
2773+unsafeSplitAt :: forall a. Unbox a =>
2774+ Int -> MutArray a -> (MutArray a, MutArray a)
2775+unsafeSplitAt i MutArray{..} =
2776+ let off = i * SIZE_OF(a)
2777+ p = arrStart + off
2778+ in ( MutArray
2779+ { arrContents = arrContents
2780+ , arrStart = arrStart
2781+ , arrEnd = p
2782+ , arrBound = p
2783+ }
2784+ , MutArray
2785+ { arrContents = arrContents
2786+ , arrStart = p
2787+ , arrEnd = arrEnd
2788+ , arrBound = arrBound
2789+ }
2790+ )
2791+
24232792-- | Create two slices of an array without copying the original array. The
24242793-- specified index @i@ is the first index of the second slice.
24252794--
24262795{-# INLINE splitAt #-}
24272796splitAt :: forall a. Unbox a => Int -> MutArray a -> (MutArray a, MutArray a)
2428−splitAt i arr@MutArray{..} =
2797+splitAt i arr =
24292798 let maxIndex = length arr - 1
24302799 in if i < 0
24312800 then error "sliceAt: negative array index"
24322801 else if i > maxIndex
24332802 then error $ "sliceAt: specified array index " ++ show i
24342803 ++ " is beyond the maximum index " ++ show maxIndex
2435− else let off = i * SIZE_OF(a)
2436− p = arrStart + off
2437− in ( MutArray
2438− { arrContents = arrContents
2439− , arrStart = arrStart
2440− , arrEnd = p
2441− , arrBound = p
2442− }
2443− , MutArray
2444− { arrContents = arrContents
2445− , arrStart = p
2446− , arrEnd = arrEnd
2447− , arrBound = arrBound
2448− }
2449− )
2804+ else unsafeSplitAt i arr
24502805
24512806-------------------------------------------------------------------------------
24522807-- Casting
@@ -2496,9 +2851,13 @@ cast arr =
24962851-- guarantees that unsafe calls will be performed in the calling thread. Making
24972852-- it safe to pass heap-allocated objects to unsafe functions.
24982853
2499−-- Should we just name it asPtr, the unsafety is implicit for any pointer
2854+-- XXX Should we just name it asPtr, the unsafety is implicit for any pointer
25002855-- operations. And we are safe from Haskell perspective because we will be
25012856-- pinning the memory.
2857+--
2858+-- XXX we cannot pass the length of the ptr here as in some cases it may not be
2859+-- available e.g. a null terminated C string. However, we can create another
2860+-- flavor of the API e.g. asPtrN.
25022861
25032862-- | Use a @MutArray a@ as @Ptr a@. This is useful when we want to pass an
25042863-- array as a pointer to some operating system call or to a "safe" FFI call.
@@ -2516,16 +2875,21 @@ cast arr =
25162875--
25172876-- /Pre-release/
25182877--
2878+{-# INLINE unsafePinnedAsPtr #-}
2879+unsafePinnedAsPtr :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2880+unsafePinnedAsPtr arr f =
2881+ Unboxed.unsafePinnedAsPtr
2882+ (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
2883+
2884+{-# DEPRECATED asPtrUnsafe "Please use unsafePinnedAsPtr instead." #-}
25192885{-# INLINE asPtrUnsafe #-}
25202886asPtrUnsafe :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2521−asPtrUnsafe arr f =
2522− Unboxed.asPtrUnsafe
2523− (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
2887+asPtrUnsafe = unsafePinnedAsPtr
25242888
2525−{-# INLINE asUnpinnedPtrUnsafe #-}
2526−asUnpinnedPtrUnsafe :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2527−asUnpinnedPtrUnsafe arr f =
2528− Unboxed.asUnpinnedPtrUnsafe
2889+{-# INLINE unsafeAsPtr #-}
2890+unsafeAsPtr :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2891+unsafeAsPtr arr f =
2892+ Unboxed.unsafeAsPtr
25292893 (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
25302894
25312895-------------------------------------------------------------------------------
@@ -2543,28 +2907,26 @@ asUnpinnedPtrUnsafe arr f =
25432907-- /Pre-release/
25442908{-# INLINE byteCmp #-}
25452909byteCmp :: MonadIO m => MutArray a -> MutArray a -> m Ordering
2546−byteCmp arr1 arr2 =
2547− liftIO
2548− $ do
2549− let marr1 = getMutableByteArray# (arrContents arr1)
2550− marr2 = getMutableByteArray# (arrContents arr2)
2551− !(I# st1#) = arrStart arr1
2552− !(I# st2#) = arrStart arr2
2553− !(I# len#) = byteLength arr1
2554− case compare (byteLength arr1) (byteLength arr2) of
2555− EQ -> do
2556− r <- liftIO $ IO $ \s# ->
2557− let res =
2558− I#
2559− (compareByteArrays#
2560− (unsafeCoerce# marr1)
2561− st1#
2562− (unsafeCoerce# marr2)
2563− st2#
2564− len#)
2565− in (# s#, res #)
2566− return $ compare r 0
2567− x -> return x
2910+byteCmp arr1 arr2 = do
2911+ let marr1 = getMutableByteArray# (arrContents arr1)
2912+ marr2 = getMutableByteArray# (arrContents arr2)
2913+ !(I# st1#) = arrStart arr1
2914+ !(I# st2#) = arrStart arr2
2915+ !(I# len#) = byteLength arr1
2916+ case compare (byteLength arr1) (byteLength arr2) of
2917+ EQ -> do
2918+ r <- liftIO $ IO $ \s# ->
2919+ let res =
2920+ I#
2921+ (compareByteArrays#
2922+ (unsafeCoerce# marr1)
2923+ st1#
2924+ (unsafeCoerce# marr2)
2925+ st2#
2926+ len#)
2927+ in (# s#, res #)
2928+ return $ compare r 0
2929+ x -> return x
25682930
25692931{-# INLINE cmp #-}
25702932{-# DEPRECATED cmp "Please use byteCmp instead." #-}
@@ -2581,7 +2943,309 @@ byteEq :: MonadIO m => MutArray a -> MutArray a -> m Bool
25812943byteEq arr1 arr2 = fmap (EQ ==) $ byteCmp arr1 arr2
25822944
25832945-------------------------------------------------------------------------------
2584−-- NFData
2946+-- Compact
2947+-------------------------------------------------------------------------------
2948+
2949+-- Note: LE versions avoid an extra copy compared to GE. LE parser trades
2950+-- backtracking one array in lieu of avoiding a copy. However, LE and GE both
2951+-- can leave some memory unused. They can split the last array to fit it
2952+-- exactly in the space.
2953+
2954+{-# INLINE_NORMAL pCompactLeAs #-}
2955+pCompactLeAs ::
2956+ forall m a. (MonadIO m, Unbox a)
2957+ => PinnedState -> Int -> Parser (MutArray a) m (MutArray a)
2958+pCompactLeAs ps maxElems = Parser step initial extract
2959+
2960+ where
2961+
2962+ maxBytes = maxElems * SIZE_OF(a)
2963+
2964+ functionName = "Streamly.Internal.Data.MutArray.pCompactLE"
2965+
2966+ initial =
2967+ return
2968+ $ if maxElems <= 0
2969+ then error
2970+ $ functionName
2971+ ++ ": the size of arrays ["
2972+ ++ show maxElems ++ "] must be a natural number"
2973+ else Parser.IPartial Nothing
2974+
2975+ step Nothing arr =
2976+ return
2977+ $ let len = byteLength arr
2978+ in if len >= maxBytes
2979+ then Parser.Done 0 arr
2980+ else Parser.Partial 0 (Just arr)
2981+ -- XXX Split the last array to use the space more compactly.
2982+ step (Just buf) arr =
2983+ let len = byteLength buf + byteLength arr
2984+ in if len > maxBytes
2985+ then return $ Parser.Done 1 buf
2986+ else do
2987+ buf1 <-
2988+ if byteCapacity buf < maxBytes
2989+ then liftIO $ reallocExplicitAs
2990+ ps (SIZE_OF(a)) maxBytes buf
2991+ else return buf
2992+ buf2 <- spliceUnsafe buf1 arr
2993+ return $ Parser.Partial 0 (Just buf2)
2994+
2995+ extract Nothing = return $ Parser.Done 0 nil
2996+ extract (Just buf) = return $ Parser.Done 0 buf
2997+
2998+-- | Parser @pCompactLE maxElems@ coalesces adjacent arrays in the input stream
2999+-- only if the combined size would be less than or equal to @maxElems@
3000+-- elements. Note that it won't split an array if the original array is already
3001+-- larger than maxElems.
3002+--
3003+-- @maxElems@ must be greater than 0.
3004+--
3005+-- Generates unpinned arrays irrespective of the pinning status of input
3006+-- arrays.
3007+--
3008+-- /Internal/
3009+{-# INLINE pCompactLE #-}
3010+pCompactLE ::
3011+ forall m a. (MonadIO m, Unbox a)
3012+ => Int -> Parser (MutArray a) m (MutArray a)
3013+pCompactLE = pCompactLeAs Unpinned
3014+
3015+-- | Pinned version of 'pCompactLE'.
3016+{-# INLINE pPinnedCompactLE #-}
3017+pPinnedCompactLE ::
3018+ forall m a. (MonadIO m, Unbox a)
3019+ => Int -> Parser (MutArray a) m (MutArray a)
3020+pPinnedCompactLE = pCompactLeAs Pinned
3021+
3022+data SpliceState s arr
3023+ = SpliceInitial s
3024+ | SpliceBuffering s arr
3025+ | SpliceYielding arr (SpliceState s arr)
3026+ | SpliceFinish
3027+
3028+-- This mutates the first array (if it has space) to append values from the
3029+-- second one. This would work for immutable arrays as well because an
3030+-- immutable array never has additional space so a new array is allocated
3031+-- instead of mutating it.
3032+
3033+{-# INLINE_NORMAL compactLeAs #-}
3034+compactLeAs :: forall m a. (MonadIO m, Unbox a)
3035+ => PinnedState -> Int -> D.Stream m (MutArray a) -> D.Stream m (MutArray a)
3036+compactLeAs ps maxElems (D.Stream step state) =
3037+ D.Stream step' (SpliceInitial state)
3038+
3039+ where
3040+
3041+ maxBytes = maxElems * SIZE_OF(a)
3042+
3043+ functionName = "Streamly.Internal.Data.MutArray.rCompactLE"
3044+
3045+ {-# INLINE_LATE step' #-}
3046+ step' gst (SpliceInitial st) = do
3047+ when (maxElems <= 0) $
3048+ -- XXX we can pass the module string from the higher level API
3049+ error $ functionName ++ ": the size of arrays [" ++ show maxElems
3050+ ++ "] must be a natural number"
3051+ r <- step gst st
3052+ case r of
3053+ D.Yield arr s -> return $
3054+ let len = byteLength arr
3055+ in if len >= maxBytes
3056+ then D.Skip (SpliceYielding arr (SpliceInitial s))
3057+ else D.Skip (SpliceBuffering s arr)
3058+ D.Skip s -> return $ D.Skip (SpliceInitial s)
3059+ D.Stop -> return D.Stop
3060+
3061+ -- XXX Split the last array to use the space more compactly.
3062+ step' gst (SpliceBuffering st buf) = do
3063+ r <- step gst st
3064+ case r of
3065+ D.Yield arr s -> do
3066+ let len = byteLength buf + byteLength arr
3067+ if len > maxBytes
3068+ then return $
3069+ D.Skip (SpliceYielding buf (SpliceBuffering s arr))
3070+ else do
3071+ buf1 <- if byteCapacity buf < maxBytes
3072+ then liftIO $ reallocExplicitAs
3073+ ps (SIZE_OF(a)) maxBytes buf
3074+ else return buf
3075+ buf2 <- spliceUnsafe buf1 arr
3076+ return $ D.Skip (SpliceBuffering s buf2)
3077+ D.Skip s -> return $ D.Skip (SpliceBuffering s buf)
3078+ D.Stop -> return $ D.Skip (SpliceYielding buf SpliceFinish)
3079+
3080+ step' _ SpliceFinish = return D.Stop
3081+
3082+ step' _ (SpliceYielding arr next) = return $ D.Yield arr next
3083+
3084+
3085+{-# INLINE_NORMAL fCompactGeAs #-}
3086+fCompactGeAs ::
3087+ forall m a. (MonadIO m, Unbox a)
3088+ => PinnedState -> Int -> FL.Fold m (MutArray a) (MutArray a)
3089+fCompactGeAs ps minElems = Fold step initial extract extract
3090+
3091+ where
3092+
3093+ minBytes = minElems * SIZE_OF(a)
3094+
3095+ functionName = "Streamly.Internal.Data.MutArray.fCompactGE"
3096+
3097+ initial =
3098+ return
3099+ $ if minElems < 0
3100+ then error
3101+ $ functionName
3102+ ++ ": the size of arrays ["
3103+ ++ show minElems ++ "] must be a natural number"
3104+ else FL.Partial Nothing
3105+
3106+ step Nothing arr =
3107+ return
3108+ $ let len = byteLength arr
3109+ in if len >= minBytes
3110+ then FL.Done arr
3111+ else FL.Partial (Just arr)
3112+ -- XXX Buffer arrays as a list to avoid copy and reallocations
3113+ step (Just buf) arr = do
3114+ let len = byteLength buf + byteLength arr
3115+ buf1 <-
3116+ if byteCapacity buf < len
3117+ then liftIO $ reallocExplicitAs
3118+ ps (SIZE_OF(a)) (max minBytes len) buf
3119+ else return buf
3120+ buf2 <- spliceUnsafe buf1 arr
3121+ if len >= minBytes
3122+ then return $ FL.Done buf2
3123+ else return $ FL.Partial (Just buf2)
3124+
3125+ extract Nothing = return nil
3126+ extract (Just buf) = return buf
3127+
3128+-- | Fold @fCompactGE minElems@ coalesces adjacent arrays in the input stream
3129+-- until the size becomes greater than or equal to @minElems@.
3130+--
3131+-- Generates unpinned arrays irrespective of the pinning status of input
3132+-- arrays.
3133+{-# INLINE fCompactGE #-}
3134+fCompactGE ::
3135+ forall m a. (MonadIO m, Unbox a)
3136+ => Int -> FL.Fold m (MutArray a) (MutArray a)
3137+fCompactGE = fCompactGeAs Unpinned
3138+
3139+-- | Pinned version of 'fCompactGE'.
3140+{-# INLINE fPinnedCompactGE #-}
3141+fPinnedCompactGE ::
3142+ forall m a. (MonadIO m, Unbox a)
3143+ => Int -> FL.Fold m (MutArray a) (MutArray a)
3144+fPinnedCompactGE = fCompactGeAs Pinned
3145+
3146+{-# INLINE_NORMAL lCompactGeAs #-}
3147+lCompactGeAs :: forall m a. (MonadIO m, Unbox a)
3148+ => PinnedState -> Int -> Fold m (MutArray a) () -> Fold m (MutArray a) ()
3149+-- The fold version turns out to be a little bit slower.
3150+-- lCompactGeAs ps n = FL.many (fCompactGeAs ps n)
3151+lCompactGeAs ps minElems (Fold step1 initial1 _ final1) =
3152+ Fold step initial extract final
3153+
3154+ where
3155+
3156+ minBytes = minElems * SIZE_OF(a)
3157+
3158+ functionName = "Streamly.Internal.Data.MutArray.lCompactGE"
3159+
3160+ initial = do
3161+ when (minElems <= 0) $
3162+ -- XXX we can pass the module string from the higher level API
3163+ error $ functionName ++ ": the size of arrays ["
3164+ ++ show minElems ++ "] must be a natural number"
3165+
3166+ r <- initial1
3167+ return $ first (Tuple' Nothing) r
3168+
3169+ {-# INLINE runInner #-}
3170+ runInner len acc buf =
3171+ if len >= minBytes
3172+ then do
3173+ r <- step1 acc buf
3174+ case r of
3175+ FL.Done _ -> return $ FL.Done ()
3176+ FL.Partial s -> do
3177+ _ <- final1 s
3178+ res <- initial1
3179+ return $ first (Tuple' Nothing) res
3180+ else return $ FL.Partial $ Tuple' (Just buf) acc
3181+
3182+ step (Tuple' Nothing r1) arr =
3183+ runInner (byteLength arr) r1 arr
3184+
3185+ -- XXX Buffer arrays as a list to avoid copy and reallocations
3186+ step (Tuple' (Just buf) r1) arr = do
3187+ let len = byteLength buf + byteLength arr
3188+ buf1 <- if byteCapacity buf < len
3189+ then liftIO $ reallocExplicitAs
3190+ ps (SIZE_OF(a)) (max minBytes len) buf
3191+ else return buf
3192+ buf2 <- spliceUnsafe buf1 arr
3193+ runInner len r1 buf2
3194+
3195+ -- XXX Several folds do extract >=> final, therefore, we need to make final
3196+ -- return "m b" rather than using extract post it if we want extract to be
3197+ -- partial.
3198+ --
3199+ -- extract forces the pending buffer to be sent to the fold which is not
3200+ -- what we want.
3201+ extract _ = error "lCompactGE: not designed for scanning"
3202+
3203+ final (Tuple' Nothing r1) = final1 r1
3204+ final (Tuple' (Just buf) r1) = do
3205+ r <- step1 r1 buf
3206+ case r of
3207+ FL.Partial rr -> final1 rr
3208+ FL.Done _ -> return ()
3209+
3210+-- | Like 'compactGE' but for transforming folds instead of stream.
3211+--
3212+-- >>> lCompactGE n = Fold.many (MutArray.fCompactGE n)
3213+--
3214+-- Generates unpinned arrays irrespective of the pinning status of input
3215+-- arrays.
3216+{-# INLINE lCompactGE #-}
3217+lCompactGE :: forall m a. (MonadIO m, Unbox a)
3218+ => Int -> Fold m (MutArray a) () -> Fold m (MutArray a) ()
3219+lCompactGE = lCompactGeAs Unpinned
3220+
3221+-- | Pinned version of 'lCompactGE'.
3222+{-# INLINE lPinnedCompactGE #-}
3223+lPinnedCompactGE :: forall m a. (MonadIO m, Unbox a)
3224+ => Int -> Fold m (MutArray a) () -> Fold m (MutArray a) ()
3225+lPinnedCompactGE = lCompactGeAs Pinned
3226+
3227+-- | @compactGE n stream@ coalesces adjacent arrays in the @stream@ until
3228+-- the size becomes greater than or equal to @n@.
3229+--
3230+-- >>> compactGE n = Stream.foldMany (MutArray.fCompactGE n)
3231+--
3232+{-# INLINE compactGE #-}
3233+compactGE ::
3234+ (MonadIO m, Unbox a)
3235+ => Int -> Stream m (MutArray a) -> Stream m (MutArray a)
3236+compactGE n = D.foldMany (fCompactGE n)
3237+
3238+-- | 'compactEQ n' coalesces adajacent arrays in the input stream to
3239+-- arrays of exact size @n@.
3240+--
3241+-- /Unimplemented/
3242+{-# INLINE compactEQ #-}
3243+compactEQ :: -- (MonadIO m, Unbox a) =>
3244+ Int -> Stream m (MutArray a) -> Stream m (MutArray a)
3245+compactEQ _n = undefined -- D.parseManyD (pCompactEQ n)
3246+
3247+-------------------------------------------------------------------------------
3248+-- In-place mutation algorithms
25853249-------------------------------------------------------------------------------
25863250
25873251-- | Strip elements which match with predicate from both ends.