@@ -31,48 +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
39− -- ** Construction
40− , nil
41−
42− -- *** Uninitialized Arrays
43− , pinnedNew
44− , pinnedNewBytes
45− , pinnedNewAligned
46− , new
47− , newArrayWith
41+ -- ** Casting
42+ , cast
43+ , castUnsafe -- XXX unsafeCast
44+ , asBytes
45+ , unsafePinnedAsPtr
46+ , unsafeAsPtr
4847
49− -- ** Cloning arrays
48+ -- ** Construction
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
5061 , clone
5162 , pinnedClone
5263
53− -- ** Slicing
64+ -- *** Slicing
5465 -- | Get a subarray without copying
55− , getSliceUnsafe
66+ , getSliceUnsafe -- XXX unsafeGetSlice
5667 , getSlice
5768 , splitAt -- XXX should be able to express using getSlice
5869 , breakOn
5970
60− -- *** From streams
71+ -- *** Stream Folds
6172 , ArrayUnsafe (..)
62− , writeNWithUnsafe
63− , writeNWith
64− , writeNUnsafe
65− , pinnedWriteNUnsafe
66− , writeN
67− , pinnedWriteN
68− , pinnedWriteNAligned
69−
70− , writeWith
71− , write
72− , pinnedWrite
73−
74− , writeRevN
75− -- , writeRev
73+ , unsafeCreateOfWith
74+ , unsafeCreateOf
75+ , unsafePinnedCreateOf
76+ , pinnedCreateOf
77+ , createOfWith
78+ , createOf
79+ , revCreateOf
80+
81+ , pinnedCreate
82+ , createWith
83+ , create
84+ -- , revCreate
7685
7786 -- *** From containers
7887 , fromListN
@@ -88,11 +97,11 @@ module Streamly.Internal.Data.MutArray.Type
8897 , fromByteStr#
8998 , fromPtrN
9099 , fromChunksK
91− , fromChunksRealloced
100+ , fromChunksRealloced -- fromSmallChunks
92101
93102 -- ** Random writes
94103 , putIndex
95− , putIndexUnsafe
104+ , putIndexUnsafe -- XXX unsafePutIndex
96105 , putIndices
97106 -- , putFromThenTo
98107 -- , putFrom -- start writing at the given position
@@ -100,87 +109,70 @@ module Streamly.Internal.Data.MutArray.Type
100109 -- , putFromTo
101110 -- , putFromRev
102111 -- , putUptoRev
103− , modifyIndexUnsafe
112+ , modifyIndexUnsafe -- XXX unsafeModifyIndex
104113 , modifyIndex
105114 , modifyIndices
106115 , modify
107116 , swapIndices
108117 , unsafeSwapIndices
109118
110− -- ** Growing and Shrinking
111− -- | Arrays grow only at the end, though it is possible to grow on both sides
112− -- and therefore have a cons as well as snoc. But that will require two
113− -- bounds in the array representation.
114−
115− -- *** Appending elements
116− , snocWith
117− , snoc
118− , snocLinear
119− , snocMay
120− , snocUnsafe
121−
122− -- *** Appending streams
123− , writeAppendNUnsafe
124− , writeAppendN
125− , writeAppendWith
126− , writeAppend
127−
128− -- *** Appending arrays
129− , spliceCopy
130− , spliceWith
131− , splice
132− , spliceExp
133− , spliceUnsafe
134− -- , putSlice
135− -- , appendSlice
136− -- , appendSliceFrom
137−
138− -- ** Eliminating and Reading
119+ -- ** Reading
139120
140− -- *** To streams
141− , reader
142− , readerRevWith
143− , readerRev
121+ -- *** Indexing
122+ , getIndex
123+ , getIndexUnsafe -- XXX unsafeGetIndex
124+ -- , getFromThenTo
125+ , getIndexRev -- getRevIndex?
126+ , indexReader
127+ , indexReaderWith
144128
145− -- *** To containers
146− , toStreamWith
147− , toStreamRevWith
148− , toStreamKWith
149− , toStreamKRevWith
129+ -- *** To Streams
150130 , read
151131 , readRev
132+ , toStreamWith
133+ , toStreamRevWith
152134 , toStreamK
135+ , toStreamKWith
153136 , toStreamKRev
137+ , toStreamKRevWith
138+
139+ -- *** To Containers
154140 , toList
155141
142+ -- *** Unfolds
156143 -- experimental
157144 , producerWith
158145 , producer
159146
160− -- *** Random reads
161− , getIndex
162− , getIndexUnsafe
163− , getIndices
164− , getIndicesWith
165− -- , getFromThenTo
166− , getIndexRev
167−
168− -- ** Memory Management
169− , blockSize
170− , arrayChunkBytes
171− , allocBytesToElemCount
172− , realloc
173− , resize
174− , resizeExp
175− , rightSize
147+ , reader
148+ , readerRevWith
149+ , readerRev
176150
177− -- ** Size
151+ -- ** Size and Capacity
152+ -- *** Size
178153 , length
179154 , byteLength
155+
156+ -- *** Capacity
180157 -- , capacity
181158 , byteCapacity
182159 , bytesFree
183160
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+
184176 -- ** In-place Mutation Algorithms
185177 , strip
186178 , reverse
@@ -191,19 +183,43 @@ module Streamly.Internal.Data.MutArray.Type
191183 , mergeBy
192184 , bubble
193185
194− -- ** Casting
195− , cast
196− , castUnsafe
197− , asBytes
198− , asPtrUnsafe
199− , 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.
200190
201− -- ** Folding
202− , foldl'
203− , foldr
204− , cmp
205− , byteCmp
206− , 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
207223
208224 -- Arrays of arrays
209225 -- We can add dimensionality parameter to the array type to get
@@ -211,28 +227,38 @@ module Streamly.Internal.Data.MutArray.Type
211227 -- convenience wrapper on top of single dimensional arrays.
212228
213229 -- ** Streams of Arrays
214− -- *** Group a stream into arrays
230+ -- *** Chunk
231+ -- | Group a stream into arrays.
215232 , chunksOf
216233 , pinnedChunksOf
217234 , buildChunks
218235
219− -- *** Split an array into slices
236+ -- *** Split
237+ -- | Split an array into slices.
238+
220239 -- , getSlicesFromLenN
221240 , splitOn
222241 -- , slicesOf
223242
224− -- *** Flatten a stream of arrays
225− , concatChunks
226− , concatChunksRev
243+ -- *** Concat
244+ -- | Append the arrays in a stream to form a stream of elements.
245+ , concatWith
246+ , concatRevWith
247+ , concat
248+ , concatRev
227249
228− -- *** Compaction
250+ -- *** Compact
251+ -- | Append the arrays in a stream to form a stream of larger arrays.
229252 , SpliceState (..)
230− , pCompactChunksLE
231− , rCompactChunksLE
232− , fCompactChunksGE
233− , lCompactChunksGE
234− , compactChunksGE
235− , compactChunksEQ
253+ , pCompactLE
254+ , pPinnedCompactLE
255+ , compactLeAs
256+ , fCompactGE
257+ , fPinnedCompactGE
258+ , lCompactGE
259+ , lPinnedCompactGE
260+ , compactGE
261+ , compactEQ
236262
237263 -- ** Utilities
238264 , roundUpToPower2
@@ -241,12 +267,37 @@ module Streamly.Internal.Data.MutArray.Type
241267 , c_memchr
242268
243269 -- * Deprecated
270+ , asPtrUnsafe
244271 , writeChunks
245272 , flattenArrays
246273 , flattenArraysRev
247274 , fromArrayStreamK
248275 , fromStreamDN
249276 , 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
250301 )
251302where
252303
@@ -302,7 +353,7 @@ import qualified Streamly.Internal.Data.StreamK.Type as K
302353import qualified Prelude
303354
304355import Prelude hiding
305− (Foldable(..), read, unlines, splitAt, reverse, truncate)
356+ (Foldable(..), concat, read, unlines, splitAt, reverse, truncate)
306357
307358#include "DocTestDataMutArray.hs"
308359
@@ -428,6 +479,8 @@ isPinned MutArray{..} = Unboxed.isPinned arrContents
428479-- we need more than that. See stg_pinnedNewAlignedByteArrayzh and
429480-- allocatePinned in GHC source.
430481
482+-- XXX Rename to emptyAlignedWith, alignSize should be first arg.
483+
431484-- | @newArrayWith allocator alignment count@ allocates a new array of zero
432485-- length and with a capacity to hold @count@ elements, using @allocator
433486-- size alignment@ as the memory allocator function.
@@ -451,12 +504,25 @@ newArrayWith alloc alignSize count = do
451504 , arrBound = size
452505 }
453506
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." #-}
454520nil ::
455521#ifdef DEVBUILD
456522 Unbox a =>
457523#endif
458524 MutArray a
459−nil = MutArray Unboxed.nil 0 0 0
525+nil = empty
460526
461527{-# INLINE newBytesAs #-}
462528newBytesAs :: MonadIO m =>
@@ -473,12 +539,13 @@ newBytesAs ps bytes = do
473539 , arrBound = bytes
474540 }
475541
476−-- | Allocates a pinned empty array that can hold 'count' items. The memory of
477−-- the array is uninitialized and the allocation is aligned as per the
478−-- '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.
479545--
480546-- /Pre-release/
481547{-# INLINE pinnedNewBytes #-}
548+{-# DEPRECATED pinnedNewBytes "Please use pinnedEmptyOf with appropriate calculation" #-}
482549pinnedNewBytes :: MonadIO m =>
483550#ifdef DEVBUILD
484551 Unbox a =>
@@ -503,21 +570,31 @@ newAs ps =
503570 (error "new: alignment is not used in unpinned arrays.")
504571
505572-- XXX can unaligned allocation be more efficient when alignment is not needed?
506−--
507−-- | Allocates an empty pinned array that can hold 'count' items. The memory of
508−-- the array is uninitialized and the allocation is aligned as per the 'Unboxed'
509−-- instance of the type.
510−--
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." #-}
511582{-# INLINE pinnedNew #-}
512583pinnedNew :: forall m a. (MonadIO m, Unbox a) => Int -> m (MutArray a)
513−pinnedNew = newAs Pinned
584+pinnedNew = pinnedEmptyOf
514585
515−-- | Allocates an empty unpinned array that can hold 'count' items. The memory
516−-- of the array is uninitialized.
586+-- | Allocates an unpinned array of zero length but growable to the specified
587+-- capacity without reallocation.
517588--
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." #-}
518595{-# INLINE new #-}
519596new :: (MonadIO m, Unbox a) => Int -> m (MutArray a)
520−new = newAs Unpinned
597+new = emptyOf
521598
522599-------------------------------------------------------------------------------
523600-- Random writes
@@ -565,7 +642,7 @@ putIndices arr = FL.foldlM' step (return ())
565642
566643 where
567644
568− step () (i, x) = liftIO (putIndex i arr x)
645+ step () (i, x) = putIndex i arr x
569646
570647-- | Modify a given index of an array using a modifier function.
571648--
@@ -761,22 +838,19 @@ arrayChunkBytes = 1024
761838roundDownTo :: Int -> Int -> Int
762839roundDownTo elemSize size = size - (size `mod` elemSize)
763840
764−-- XXX See if resizing can be implemented by reading the old array as a stream
765−-- and then using writeN to the new array.
766−--
767841-- NOTE: we are passing elemSize explicitly to avoid an Unboxed constraint.
768842-- Since this is not inlined Unboxed consrraint leads to dictionary passing
769843-- which complicates some inspection tests.
770844--
771−{-# NOINLINE reallocExplicit #-}
772−reallocExplicit :: Int -> Int -> MutArray a -> IO (MutArray a)
773−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
774848 assertM(arrEnd <= arrBound)
775849
776850 -- Allocate new array
777851 let newCapMaxInBytes = roundUpLargeArray newCapacityInBytes
778852 contents <-
779− if Unboxed.isPinned arrContents
853+ if ps == Pinned
780854 then Unboxed.pinnedNew newCapMaxInBytes
781855 else Unboxed.new newCapMaxInBytes
782856 let !(MutByteArray mbarrFrom#) = arrContents
@@ -801,6 +875,10 @@ reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
801875 , arrBound = newCapInBytes
802876 }
803877
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+
804882-- | @realloc newCapacity array@ reallocates the array to the specified
805883-- capacity in bytes.
806884--
@@ -812,7 +890,12 @@ reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
812890-- If the original array is pinned, the newly allocated array is also pinned.
813891{-# INLINABLE realloc #-}
814892realloc :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m (MutArray a)
815−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
816899
817900-- | @reallocWith label capSizer minIncrBytes array@. The label is used
818901-- in error messages and the capSizer is used to determine the capacity of the
@@ -835,14 +918,14 @@ reallocWith label capSizer minIncrBytes arr = do
835918 where
836919
837920 badSize newSize =
838− concat
921+ Prelude.concat
839922 [ label
840923 , ": new array size (in bytes) is less than required size "
841924 , show newSize
842925 , ". Please check the sizing function passed."
843926 ]
844927
845−-- | @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
846929-- it is enough to hold the specified number of elements. Nothing is done if
847930-- the specified capacity is less than the length of the array.
848931--
@@ -850,34 +933,46 @@ reallocWith label capSizer minIncrBytes arr = do
850933-- the block size (4K).
851934--
852935-- /Pre-release/
853−{-# INLINE resize #-}
854−resize :: forall m a. (MonadIO m, Unbox a) =>
936+{-# INLINE grow #-}
937+grow :: forall m a. (MonadIO m, Unbox a) =>
855938 Int -> MutArray a -> m (MutArray a)
856−resize nElems arr@MutArray{..} = do
939+grow nElems arr@MutArray{..} = do
857940 let req = SIZE_OF(a) * nElems
858− len = arrEnd - arrStart
859− if req < len
941+ cap = arrBound - arrStart
942+ if req < cap
860943 then return arr
861944 else realloc req arr
862945
863−-- | Like 'resize' but if the byte capacity is more than 'largeObjectThreshold'
864−-- 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.
865954--
866955-- /Pre-release/
867−{-# INLINE resizeExp #-}
868−resizeExp :: forall m a. (MonadIO m, Unbox a) =>
956+{-# INLINE growExp #-}
957+growExp :: forall m a. (MonadIO m, Unbox a) =>
869958 Int -> MutArray a -> m (MutArray a)
870−resizeExp nElems arr@MutArray{..} = do
959+growExp nElems arr@MutArray{..} = do
871960 let req = roundUpLargeArray (SIZE_OF(a) * nElems)
872961 req1 =
873962 if req > largeObjectThreshold
874963 then roundUpToPower2 req
875964 else req
876− len = arrEnd - arrStart
877− if req1 < len
965+ cap = arrBound - arrStart
966+ if req1 < cap
878967 then return arr
879968 else realloc req1 arr
880969
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+
881976-- | Resize the allocated memory to drop any reserved free space at the end of
882977-- the array and reallocate it to reduce wastage.
883978--
@@ -942,7 +1037,7 @@ snocUnsafe arr@MutArray{..} = snocNewEnd (INDEX_NEXT(arrEnd,a)) arr
9421037{-# INLINE snocMay #-}
9431038snocMay :: forall m a. (MonadIO m, Unbox a) =>
9441039 MutArray a -> a -> m (Maybe (MutArray a))
945−snocMay arr@MutArray{..} x = liftIO $ do
1040+snocMay arr@MutArray{..} x = do
9461041 let newEnd = INDEX_NEXT(arrEnd,a)
9471042 if newEnd <= arrBound
9481043 then Just <$> snocNewEnd newEnd arr x
@@ -956,7 +1051,7 @@ snocWithRealloc :: forall m a. (MonadIO m, Unbox a) =>
9561051 -> a
9571052 -> m (MutArray a)
9581053snocWithRealloc sizer arr x = do
959− arr1 <- liftIO $ reallocWith "snocWith" sizer (SIZE_OF(a)) arr
1054+ arr1 <- reallocWith "snocWith" sizer (SIZE_OF(a)) arr
9601055 snocUnsafe arr1 x
9611056
9621057-- | @snocWith sizer arr elem@ mutates @arr@ to append @elem@. The length of
@@ -978,7 +1073,7 @@ snocWith :: forall m a. (MonadIO m, Unbox a) =>
9781073 -> MutArray a
9791074 -> a
9801075 -> m (MutArray a)
981−snocWith allocSize arr x = liftIO $ do
1076+snocWith allocSize arr x = do
9821077 let newEnd = INDEX_NEXT(arrEnd arr,a)
9831078 if newEnd <= arrBound arr
9841079 then snocNewEnd newEnd arr x
@@ -1022,6 +1117,126 @@ snoc = snocWith f
10221117 then oldSize * 2
10231118 else roundUpToPower2 oldSize * 2
10241119
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+
10251240-------------------------------------------------------------------------------
10261241-- Random reads
10271242-------------------------------------------------------------------------------
@@ -1064,10 +1279,10 @@ getIndexRev i MutArray{..} = do
10641279data GetIndicesState contents start end st =
10651280 GetIndicesState contents start end st
10661281
1067−{-# INLINE getIndicesWith #-}
1068−getIndicesWith :: (Monad m, Unbox a) =>
1282+{-# INLINE indexReaderWith #-}
1283+indexReaderWith :: (Monad m, Unbox a) =>
10691284 (forall b. IO b -> m b) -> D.Stream m Int -> Unfold m (MutArray a) a
1070−getIndicesWith liftio (D.Stream stepi sti) = Unfold step inject
1285+indexReaderWith liftio (D.Stream stepi sti) = Unfold step inject
10711286
10721287 where
10731288
@@ -1086,14 +1301,26 @@ getIndicesWith liftio (D.Stream stepi sti) = Unfold step inject
10861301 D.Skip s -> return $ D.Skip (GetIndicesState contents start end s)
10871302 D.Stop -> return D.Stop
10881303
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+
10891310-- | Given an unfold that generates array indices, read the elements on those
10901311-- indices from the supplied MutArray. An error is thrown if an index is out of
10911312-- bounds.
10921313--
10931314-- /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." #-}
10941321{-# INLINE getIndices #-}
10951322getIndices :: (MonadIO m, Unbox a) => Stream m Int -> Unfold m (MutArray a) a
1096−getIndices = getIndicesWith liftIO
1323+getIndices = indexReader
10971324
10981325-------------------------------------------------------------------------------
10991326-- Subarrays
@@ -1388,7 +1615,7 @@ chunksOfAs ps n (D.Stream step state) =
13881615--
13891616-- Same as the following but may be more efficient:
13901617--
1391−-- >>> chunksOf n = Stream.foldMany (MutArray.writeN n)
1618+-- >>> chunksOf n = Stream.foldMany (MutArray.createOf n)
13921619--
13931620-- /Pre-release/
13941621{-# INLINE_NORMAL chunksOf #-}
@@ -1397,14 +1624,14 @@ chunksOf :: forall m a. (MonadIO m, Unbox a)
13971624-- XXX the idiomatic implementation leads to large regression in the D.reverse'
13981625-- benchmark. It seems it has difficulty producing optimized code when
13991626-- converting to StreamK. Investigate GHC optimizations.
1400−-- chunksOf n = D.foldMany (writeN n)
1627+-- chunksOf n = D.foldMany (createOf n)
14011628chunksOf = chunksOfAs Unpinned
14021629
14031630-- | Like 'chunksOf' but creates pinned arrays.
14041631{-# INLINE_NORMAL pinnedChunksOf #-}
14051632pinnedChunksOf :: forall m a. (MonadIO m, Unbox a)
14061633 => Int -> D.Stream m a -> D.Stream m (MutArray a)
1407−-- pinnedChunksOf n = D.foldMany (pinnedWriteN n)
1634+-- pinnedChunksOf n = D.foldMany (pinnedCreateOf n)
14081635pinnedChunksOf = chunksOfAs Pinned
14091636
14101637-- | When we are buffering a stream of unknown size into an array we do not
@@ -1437,16 +1664,10 @@ data FlattenState s contents a =
14371664 OuterLoop s
14381665 | InnerLoop s contents !Int !Int
14391666
1440−-- | Use the "reader" unfold instead.
1441−--
1442−-- @concatChunks = unfoldMany reader@
1443−--
1444−-- We can try this if there are any fusion issues in the unfold.
1445−--
1446−{-# INLINE_NORMAL concatChunks #-}
1447−concatChunks :: forall m a. (MonadIO m, Unbox a)
1448− => D.Stream m (MutArray a) -> D.Stream m a
1449−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)
14501671
14511672 where
14521673
@@ -1463,24 +1684,30 @@ concatChunks (D.Stream step state) = D.Stream step' (OuterLoop state)
14631684 return $ D.Skip $ OuterLoop st
14641685
14651686 step' _ (InnerLoop st contents p end) = do
1466− x <- liftIO $ peekAt p contents
1687+ x <- liftio $ peekAt p contents
14671688 return $ D.Yield x (InnerLoop st contents (INDEX_NEXT(p,a)) end)
14681689
1469−{-# DEPRECATED flattenArrays "Please use \"unfoldMany reader\" instead." #-}
1470−flattenArrays :: forall m a. (MonadIO m, Unbox a)
1471− => D.Stream m (MutArray a) -> D.Stream m a
1472−flattenArrays = concatChunks
1473−
1474−-- | Use the "readerRev" unfold instead.
1690+-- | Use the "reader" unfold instead.
14751691--
1476−-- @concatChunks = unfoldMany readerRev@
1692+-- @concat = unfoldMany reader@
14771693--
14781694-- We can try this if there are any fusion issues in the unfold.
14791695--
1480−{-# INLINE_NORMAL concatChunksRev #-}
1481−concatChunksRev :: forall m a. (MonadIO m, Unbox a)
1696+{-# INLINE_NORMAL concat #-}
1697+concat :: forall m a. (MonadIO m, Unbox a)
14821698 => D.Stream m (MutArray a) -> D.Stream m a
1483−concatChunksRev (D.Stream step state) = D.Stream step' (OuterLoop state)
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)
1704+ => D.Stream m (MutArray a) -> D.Stream m a
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)
14841711
14851712 where
14861713
@@ -1498,14 +1725,26 @@ concatChunksRev (D.Stream step state) = D.Stream step' (OuterLoop state)
14981725 return $ D.Skip $ OuterLoop st
14991726
15001727 step' _ (InnerLoop st contents p start) = do
1501− x <- liftIO $ peekAt p contents
1728+ x <- liftio $ peekAt p contents
15021729 let cur = INDEX_PREV(p,a)
15031730 return $ D.Yield x (InnerLoop st contents cur start)
15041731
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+
15051743{-# DEPRECATED flattenArraysRev "Please use \"unfoldMany readerRev\" instead." #-}
1744+{-# INLINE flattenArraysRev #-}
15061745flattenArraysRev :: forall m a. (MonadIO m, Unbox a)
15071746 => D.Stream m (MutArray a) -> D.Stream m a
1508−flattenArraysRev = concatChunksRev
1747+flattenArraysRev = concatRev
15091748
15101749-------------------------------------------------------------------------------
15111750-- Unfolds
@@ -1734,7 +1973,7 @@ foldr f z arr = D.foldr f z $ read arr
17341973
17351974-- XXX Keep the bound intact to not lose any free space? Perf impact?
17361975
1737−-- | @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
17381977-- array.
17391978--
17401979-- Unsafe: Do not drive the fold beyond @n@ elements, it will lead to memory
@@ -1743,13 +1982,12 @@ foldr f z arr = D.foldr f z $ read arr
17431982-- Any free space left in the array after appending @n@ elements is lost.
17441983--
17451984-- /Internal/
1746−{-# INLINE_NORMAL writeAppendNUnsafe #-}
1747−writeAppendNUnsafe :: forall m a. (MonadIO m, Unbox a) =>
1985+{-# INLINE_NORMAL unsafeAppendN #-}
1986+unsafeAppendN :: forall m a. (MonadIO m, Unbox a) =>
17481987 Int
17491988 -> m (MutArray a)
17501989 -> Fold m a (MutArray a)
1751−writeAppendNUnsafe n action =
1752− fmap fromArrayUnsafe $ FL.foldlM' step initial
1990+unsafeAppendN n action = fmap fromArrayUnsafe $ FL.foldlM' step initial
17531991
17541992 where
17551993
@@ -1762,7 +2000,7 @@ writeAppendNUnsafe n action =
17622000 -- otherwise we lose that space.
17632001 arr1 <-
17642002 if free < needed
1765− then noinline reallocWith "writeAppendNUnsafeWith" (+ needed) needed arr
2003+ then noinline reallocWith "unsafeAppendN" (+ needed) needed arr
17662004 else return arr
17672005 return $ toArrayUnsafe arr1
17682006
@@ -1770,30 +2008,51 @@ writeAppendNUnsafe n action =
17702008 liftIO $ pokeAt end contents x
17712009 return $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
17722010
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+
17732019-- | Append @n@ elements to an existing array. Any free space left in the array
17742020-- after appending @n@ elements is lost.
17752021--
1776−-- >>> writeAppendN n initial = Fold.take n (MutArray.writeAppendNUnsafe n initial)
2022+-- >>> appendN n initial = Fold.take n (MutArray.unsafeAppendN n initial)
17772023--
1778−{-# 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 #-}
17792032writeAppendN :: forall m a. (MonadIO m, Unbox a) =>
17802033 Int -> m (MutArray a) -> Fold m a (MutArray a)
1781−writeAppendN n initial = FL.take n (writeAppendNUnsafe n initial)
2034+writeAppendN = appendN
17822035
1783−-- | @writeAppendWith realloc action@ mutates the array generated by @action@ to
2036+-- | @appendWith realloc action@ mutates the array generated by @action@ to
17842037-- append the input stream. If there is no reserved space available in the
17852038-- array it is reallocated to a size in bytes determined by @realloc oldSize@,
17862039-- where @oldSize@ is the current size of the array in bytes.
17872040--
17882041-- Note that the returned array may be a mutated version of original array.
17892042--
1790−-- >>> writeAppendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
2043+-- >>> appendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
17912044--
17922045-- /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." #-}
17932052{-# INLINE writeAppendWith #-}
17942053writeAppendWith :: forall m a. (MonadIO m, Unbox a) =>
17952054 (Int -> Int) -> m (MutArray a) -> Fold m a (MutArray a)
1796−writeAppendWith sizer = FL.foldlM' (snocWith sizer)
2055+writeAppendWith = appendWith
17972056
17982057-- | @append action@ mutates the array generated by @action@ to append the
17992058-- input stream. If there is no reserved space available in the array it is
@@ -1801,26 +2060,33 @@ writeAppendWith sizer = FL.foldlM' (snocWith sizer)
18012060--
18022061-- Note that the returned array may be a mutated version of original array.
18032062--
1804−-- >>> writeAppend = MutArray.writeAppendWith (* 2)
2063+-- >>> append = MutArray.appendWith (* 2)
18052064--
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." #-}
18062072{-# INLINE writeAppend #-}
18072073writeAppend :: forall m a. (MonadIO m, Unbox a) =>
18082074 m (MutArray a) -> Fold m a (MutArray a)
1809−writeAppend = writeAppendWith (* 2)
2075+writeAppend = append
18102076
18112077-- XXX We can carry bound as well in the state to make sure we do not lose the
18122078-- remaining capacity. Need to check perf impact.
18132079--
1814−-- | Like 'writeNUnsafe' but takes a new array allocator @alloc size@ function
1815−-- as argument.
2080+-- | Like 'unsafeCreateOf' but takes a new array allocator @alloc size@
2081+-- function as argument.
18162082--
1817−-- >>> writeNWithUnsafe alloc n = MutArray.writeAppendNUnsafe (alloc n) n
2083+-- >>> unsafeCreateOfWith alloc n = MutArray.unsafeAppendN (alloc n) n
18182084--
18192085-- /Pre-release/
1820−{-# INLINE_NORMAL writeNWithUnsafe #-}
1821−writeNWithUnsafe :: forall m a. (MonadIO m, Unbox a)
2086+{-# INLINE_NORMAL unsafeCreateOfWith #-}
2087+unsafeCreateOfWith :: forall m a. (MonadIO m, Unbox a)
18222088 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
1823−writeNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
2089+unsafeCreateOfWith alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
18242090
18252091 where
18262092
@@ -1831,40 +2097,64 @@ writeNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
18312097 return
18322098 $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
18332099
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+
18342106{-# INLINE_NORMAL writeNUnsafeAs #-}
18352107writeNUnsafeAs :: forall m a. (MonadIO m, Unbox a)
18362108 => PinnedState -> Int -> Fold m a (MutArray a)
1837−writeNUnsafeAs ps = writeNWithUnsafe (newAs ps)
2109+writeNUnsafeAs ps = unsafeCreateOfWith (newAs ps)
18382110
1839−-- | 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
18402112-- driver must not call the step function more than 'n' times otherwise it will
18412113-- corrupt the memory and crash. This function exists mainly because any
18422114-- conditional in the step function blocks fusion causing 10x performance
18432115-- slowdown.
18442116--
1845−-- >>> writeNUnsafe = MutArray.writeNWithUnsafe MutArray.new
2117+-- >>> unsafeCreateOf = MutArray.unsafeCreateOfWith MutArray.emptyOf
18462118--
1847−{-# 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 #-}
18482126writeNUnsafe :: forall m a. (MonadIO m, Unbox a)
18492127 => Int -> Fold m a (MutArray a)
1850−writeNUnsafe = writeNUnsafeAs Unpinned
2128+writeNUnsafe = unsafeCreateOf
18512129
1852−-- | Like 'writeNUnsafe' but creates a pinned array.
1853−{-# INLINE_NORMAL pinnedWriteNUnsafe #-}
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
2135+
2136+{-# DEPRECATED pinnedWriteNUnsafe "Please use unsafePinnedCreateOf instead." #-}
2137+{-# INLINE pinnedWriteNUnsafe #-}
18542138pinnedWriteNUnsafe :: forall m a. (MonadIO m, Unbox a)
18552139 => Int -> Fold m a (MutArray a)
1856−pinnedWriteNUnsafe = writeNUnsafeAs Pinned
2140+pinnedWriteNUnsafe = unsafePinnedCreateOf
18572141
1858−-- | @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
18592143-- allocated using the @alloc@ function.
18602144--
1861−-- >>> writeNWith alloc n = Fold.take n (MutArray.writeNWithUnsafe alloc n)
1862−-- >>> 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
18632147--
1864−{-# 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 #-}
18652155writeNWith :: forall m a. (MonadIO m, Unbox a)
18662156 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
1867−writeNWith alloc n = FL.take n (writeNWithUnsafe alloc n)
2157+writeNWith = createOfWith
18682158
18692159{-# INLINE_NORMAL writeNAs #-}
18702160writeNAs ::
@@ -1872,28 +2162,42 @@ writeNAs ::
18722162 => PinnedState
18732163 -> Int
18742164 -> Fold m a (MutArray a)
1875−writeNAs ps = writeNWith (newAs ps)
2165+writeNAs ps = createOfWith (newAs ps)
18762166
1877−-- | @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
18782168-- 'MutArray'.
18792169--
1880−-- >>> writeN = MutArray.writeNWith MutArray.new
1881−-- >>> writeN n = Fold.take n (MutArray.writeNUnsafe n)
1882−-- >>> 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)
18832173--
1884−{-# 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 #-}
18852181writeN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
1886−writeN = writeNAs Unpinned
2182+writeN = createOf
18872183
1888−-- | Like 'writeN' but creates a pinned array.
1889−{-# INLINE_NORMAL pinnedWriteN #-}
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
2191+
2192+{-# DEPRECATED pinnedWriteN "Please use pinnedCreateOf instead." #-}
2193+{-# INLINE pinnedWriteN #-}
18902194pinnedWriteN ::
18912195 forall m a. (MonadIO m, Unbox a)
18922196 => Int
18932197 -> Fold m a (MutArray a)
1894−pinnedWriteN = writeNAs Pinned
2198+pinnedWriteN = pinnedCreateOf
18952199
1896−-- | Like writeNWithUnsafe but writes the array in reverse order.
2200+-- | Like unsafeCreateOfWith but writes the array in reverse order.
18972201--
18982202-- /Internal/
18992203{-# INLINE_NORMAL writeRevNWithUnsafe #-}
@@ -1914,7 +2218,7 @@ writeRevNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
19142218 return
19152219 $ ArrayUnsafe contents ptr end
19162220
1917−-- | Like writeNWith but writes the array in reverse order.
2221+-- | Like createOfWith but writes the array in reverse order.
19182222--
19192223-- /Internal/
19202224{-# INLINE_NORMAL writeRevNWith #-}
@@ -1922,31 +2226,36 @@ writeRevNWith :: forall m a. (MonadIO m, Unbox a)
19222226 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
19232227writeRevNWith alloc n = FL.take n (writeRevNWithUnsafe alloc n)
19242228
1925−-- | Like writeN but writes the array in reverse order.
2229+-- | Like 'createOf' but writes the array in reverse order.
19262230--
19272231-- /Pre-release/
1928−{-# 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 #-}
19292238writeRevN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
1930−writeRevN = writeRevNWith new
2239+writeRevN = revCreateOf
19312240
19322241-- | @pinnedWriteNAligned align n@ folds a maximum of @n@ elements from the
19332242-- input stream to a 'MutArray' aligned to the given size.
19342243--
1935−-- >>> pinnedWriteNAligned align = MutArray.writeNWith (MutArray.pinnedNewAligned align)
1936−-- >>> 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)
19372246--
19382247-- /Pre-release/
19392248--
19402249{-# INLINE_NORMAL pinnedWriteNAligned #-}
19412250pinnedWriteNAligned :: forall m a. (MonadIO m, Unbox a)
19422251 => Int -> Int -> Fold m a (MutArray a)
1943−pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
2252+pinnedWriteNAligned align = createOfWith (pinnedNewAligned align)
19442253
19452254-- XXX Buffer to a list instead?
19462255
19472256-- | Buffer a stream into a stream of arrays.
19482257--
1949−-- >>> buildChunks n = Fold.many (MutArray.writeN n) Fold.toStreamK
2258+-- >>> buildChunks n = Fold.many (MutArray.createOf n) Fold.toStreamK
19502259--
19512260-- Breaking an array into an array stream can be useful to consume a large
19522261-- array sequentially such that memory of the array is released incrementatlly.
@@ -1958,9 +2267,10 @@ pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
19582267{-# INLINE_NORMAL buildChunks #-}
19592268buildChunks :: (MonadIO m, Unbox a) =>
19602269 Int -> Fold m a (StreamK n (MutArray a))
1961−buildChunks n = FL.many (writeN n) FL.toStreamK
2270+buildChunks n = FL.many (createOf n) FL.toStreamK
19622271
19632272{-# DEPRECATED writeChunks "Please use buildChunks instead." #-}
2273+{-# INLINE writeChunks #-}
19642274writeChunks :: (MonadIO m, Unbox a) =>
19652275 Int -> Fold m a (StreamK n (MutArray a))
19662276writeChunks = buildChunks
@@ -1968,29 +2278,29 @@ writeChunks = buildChunks
19682278{-# INLINE_NORMAL writeWithAs #-}
19692279writeWithAs :: forall m a. (MonadIO m, Unbox a)
19702280 => PinnedState -> Int -> Fold m a (MutArray a)
1971−-- writeWithAs ps n = FL.rmapM rightSize $ writeAppendWith (* 2) (newAs ps n)
2281+-- writeWithAs ps n = FL.rmapM rightSize $ appendWith (* 2) (newAs ps n)
19722282writeWithAs ps elemCount =
19732283 FL.rmapM extract $ FL.foldlM' step initial
19742284
19752285 where
19762286
19772287 initial = do
1978− when (elemCount < 0) $ error "writeWith: elemCount is negative"
1979− liftIO $ newAs ps elemCount
2288+ when (elemCount < 0) $ error "createWith: elemCount is negative"
2289+ newAs ps elemCount
19802290
19812291 step arr@(MutArray _ start end bound) x
19822292 | INDEX_NEXT(end,a) > bound = do
19832293 let oldSize = end - start
19842294 newSize = max (oldSize * 2) 1
1985− arr1 <- liftIO $ reallocExplicit (SIZE_OF(a)) newSize arr
2295+ arr1 <- liftIO $ reallocExplicitAs ps (SIZE_OF(a)) newSize arr
19862296 snocUnsafe arr1 x
19872297 step arr x = snocUnsafe arr x
19882298
19892299 extract = liftIO . rightSize
19902300
1991−-- XXX Compare writeWith with fromStreamD which uses an array of streams
2301+-- XXX Compare createWith with fromStreamD which uses an array of streams
19922302-- implementation. We can write this using buildChunks above if that is faster.
1993−-- If writeWith is faster then we should use that to implement
2303+-- If createWith is faster then we should use that to implement
19942304-- fromStreamD.
19952305--
19962306-- XXX The realloc based implementation needs to make one extra copy if we use
@@ -2005,39 +2315,56 @@ writeWithAs ps elemCount =
20052315-- XXX check if GHC's memory allocator is efficient enough. We can try the C
20062316-- malloc to compare against.
20072317
2008−-- | @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
20092319-- starts at a size big enough to hold minCount elements, the size is doubled
20102320-- every time the array needs to be grown.
20112321--
20122322-- /Caution! Do not use this on infinite streams./
20132323--
2014−-- >>> f n = MutArray.writeAppendWith (* 2) (MutArray.new n)
2015−-- >>> writeWith n = Fold.rmapM MutArray.rightSize (f n)
2016−-- >>> 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)
20172327--
20182328-- /Pre-release/
2019−{-# 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 #-}
20202337writeWith :: forall m a. (MonadIO m, Unbox a)
20212338 => Int -> Fold m a (MutArray a)
2022−-- writeWith n = FL.rmapM rightSize $ writeAppendWith (* 2) (new n)
2023−writeWith = writeWithAs Unpinned
2339+writeWith = createWith
20242340
20252341-- | Fold the whole input to a single array.
20262342--
2027−-- Same as 'writeWith' using an initial array size of 'arrayChunkBytes' bytes
2343+-- Same as 'createWith' using an initial array size of 'arrayChunkBytes' bytes
20282344-- rounded up to the element size.
20292345--
20302346-- /Caution! Do not use this on infinite streams./
20312347--
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." #-}
20322354{-# INLINE write #-}
20332355write :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2034−write = writeWith (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2356+write = create
20352357
2036−-- | 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." #-}
20372365{-# INLINE pinnedWrite #-}
20382366pinnedWrite :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2039−pinnedWrite =
2040− writeWithAs Pinned (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2367+pinnedWrite = pinnedCreate
20412368
20422369-------------------------------------------------------------------------------
20432370-- construct from streams, known size
@@ -2047,7 +2374,7 @@ pinnedWrite =
20472374fromStreamDNAs :: forall m a. (MonadIO m, Unbox a)
20482375 => PinnedState -> Int -> D.Stream m a -> m (MutArray a)
20492376fromStreamDNAs ps limit str = do
2050− (arr :: MutArray a) <- liftIO $ newAs ps limit
2377+ (arr :: MutArray a) <- newAs ps limit
20512378 end <- D.foldlM' (fwrite (arrContents arr)) (return $ arrEnd arr) $ D.take limit str
20522379 return $ arr {arrEnd = end}
20532380
@@ -2057,17 +2384,18 @@ fromStreamDNAs ps limit str = do
20572384 liftIO $ pokeAt ptr arrContents x
20582385 return $ INDEX_NEXT(ptr,a)
20592386
2060−-- | Use the 'writeN' fold instead.
2387+-- | Use the 'createOf' fold instead.
20612388--
2062−-- >>> fromStreamN n = Stream.fold (MutArray.writeN n)
2389+-- >>> fromStreamN n = Stream.fold (MutArray.createOf n)
20632390--
20642391{-# INLINE_NORMAL fromStreamN #-}
20652392fromStreamN :: forall m a. (MonadIO m, Unbox a)
20662393 => Int -> D.Stream m a -> m (MutArray a)
2067−-- fromStreamDN n = D.fold (writeN n)
2394+-- fromStreamDN n = D.fold (createOf n)
20682395fromStreamN = fromStreamDNAs Unpinned
20692396
20702397{-# DEPRECATED fromStreamDN "Please use fromStreamN instead." #-}
2398+{-# INLINE fromStreamDN #-}
20712399fromStreamDN :: forall m a. (MonadIO m, Unbox a)
20722400 => Int -> D.Stream m a -> m (MutArray a)
20732401fromStreamDN = fromStreamN
@@ -2090,20 +2418,20 @@ pinnedFromListN n xs = fromStreamDNAs Pinned n $ D.fromList xs
20902418-- /Pre-release/
20912419{-# INLINE fromListRevN #-}
20922420fromListRevN :: (MonadIO m, Unbox a) => Int -> [a] -> m (MutArray a)
2093−fromListRevN n xs = D.fold (writeRevN n) $ D.fromList xs
2421+fromListRevN n xs = D.fold (revCreateOf n) $ D.fromList xs
20942422
20952423-- | Convert a pure stream in Identity monad to a mutable array.
20962424{-# INLINABLE fromPureStreamN #-}
20972425fromPureStreamN :: (MonadIO m, Unbox a) =>
20982426 Int -> Stream Identity a -> m (MutArray a)
20992427fromPureStreamN n xs =
2100− liftIO $ D.fold (writeN n) $ D.morphInner (return . runIdentity) xs
2428+ D.fold (createOf n) $ D.morphInner (return . runIdentity) xs
21012429
21022430-- | Convert a pure stream in Identity monad to a mutable array.
21032431{-# INLINABLE fromPureStream #-}
21042432fromPureStream :: (MonadIO m, Unbox a) => Stream Identity a -> m (MutArray a)
21052433fromPureStream xs =
2106− liftIO $ D.fold write $ D.morphInner (return . runIdentity) xs
2434+ D.fold create $ D.morphInner (return . runIdentity) xs
21072435
21082436{-# INLINABLE fromPtrN #-}
21092437fromPtrN :: MonadIO m => Int -> Ptr Word8 -> m (MutArray Word8)
@@ -2112,7 +2440,7 @@ fromPtrN len addr = do
21122440 -- XXX We can implement a stream copy in a similar way by streaming Word64
21132441 -- first and then remaining Word8.
21142442 arr <- new len
2115− _ <- asUnpinnedPtrUnsafe arr
2443+ _ <- unsafeAsPtr arr
21162444 (\ptr -> liftIO $ c_memcpy ptr addr (fromIntegral len))
21172445 return (arr {arrEnd = len})
21182446
@@ -2128,7 +2456,7 @@ fromByteStr# addr = do
21282456 len <- liftIO $ c_strlen (Ptr addr)
21292457 let lenInt = fromIntegral len
21302458 arr <- new lenInt
2131− _ <- asUnpinnedPtrUnsafe arr (\ptr -> liftIO $ c_memcpy ptr (Ptr addr) len)
2459+ _ <- unsafeAsPtr arr (\ptr -> liftIO $ c_memcpy ptr (Ptr addr) len)
21322460 return (arr {arrEnd = lenInt})
21332461
21342462-------------------------------------------------------------------------------
@@ -2149,7 +2477,7 @@ fromChunksRealloced s = do
21492477 arr <- D.foldlM' spliceExp (pure a) strm
21502478 -- Reallocation is exponential so there may be 50% empty space in
21512479 -- worst case. One more reallocation to reclaim the space.
2152− liftIO $ rightSize arr
2480+ rightSize arr
21532481 Nothing -> pure nil
21542482
21552483-------------------------------------------------------------------------------
@@ -2187,6 +2515,7 @@ fromChunksK :: (Unbox a, MonadIO m) =>
21872515fromChunksK = fromChunkskAs Unpinned
21882516
21892517{-# DEPRECATED fromArrayStreamK "Please use fromChunksK instead." #-}
2518+{-# INLINE fromArrayStreamK #-}
21902519fromArrayStreamK :: (Unbox a, MonadIO m) =>
21912520 StreamK m (MutArray a) -> m (MutArray a)
21922521fromArrayStreamK = fromChunksK
@@ -2198,7 +2527,7 @@ fromStreamDAs ps m =
21982527 arrayStreamKFromStreamDAs Unpinned m >>= fromChunkskAs ps
21992528
22002529-- | Create an 'Array' from a stream. This is useful when we want to create a
2201−-- 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
22022531-- as efficient when the size is already known.
22032532--
22042533-- Note that if the input stream is too large memory allocation for the array
@@ -2210,19 +2539,19 @@ fromStreamDAs ps m =
22102539fromStream :: (MonadIO m, Unbox a) => Stream m a -> m (MutArray a)
22112540fromStream = fromStreamDAs Unpinned
22122541
2213−-- fromStream (Stream m) = P.fold write m
2542+-- fromStream (Stream m) = P.fold create m
22142543-- CAUTION: a very large number (millions) of arrays can degrade performance
22152544-- due to GC overhead because we need to buffer the arrays before we flatten
22162545-- all the arrays.
22172546--
2218−-- XXX Compare if this is faster or "fold write".
2547+-- XXX Compare if this is faster or "fold create".
22192548--
22202549-- | We could take the approach of doubling the memory allocation on each
22212550-- overflow. This would result in more or less the same amount of copying as in
22222551-- the chunking approach. However, if we have to shrink in the end then it may
22232552-- result in an extra copy of the entire data.
22242553--
2225−-- >>> fromStreamD = StreamD.fold MutArray.write
2554+-- >>> fromStreamD = StreamD.fold MutArray.create
22262555--
22272556{-# INLINE fromStreamD #-}
22282557{-# DEPRECATED fromStreamD "Please use fromStream instead." #-}
@@ -2252,6 +2581,10 @@ fromListRev xs = fromListRevN (Prelude.length xs) xs
22522581-- Cloning
22532582-------------------------------------------------------------------------------
22542583
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+
22552588{-# INLINE cloneAs #-}
22562589cloneAs ::
22572590 ( MonadIO m
@@ -2261,7 +2594,7 @@ cloneAs ::
22612594 )
22622595 => PinnedState -> MutArray a -> m (MutArray a)
22632596cloneAs ps src =
2264− liftIO $ do
2597+ do
22652598 let startSrc = arrStart src
22662599 srcLen = arrEnd src - startSrc
22672600 newArrContents <-
@@ -2300,17 +2633,16 @@ spliceCopy :: forall m a. MonadIO m =>
23002633 Unbox a =>
23012634#endif
23022635 MutArray a -> MutArray a -> m (MutArray a)
2303−spliceCopy arr1 arr2 = liftIO $ do
2636+spliceCopy arr1 arr2 = do
23042637 let start1 = arrStart arr1
23052638 start2 = arrStart arr2
23062639 len1 = arrEnd arr1 - start1
23072640 len2 = arrEnd arr2 - start2
2308− let newLen = len1 + len2
2641+ let len = len1 + len2
23092642 newArrContents <-
23102643 if Unboxed.isPinned (arrContents arr1)
2311− then Unboxed.pinnedNew newLen
2312− else Unboxed.new newLen
2313− let len = len1 + len2
2644+ then liftIO $ Unboxed.pinnedNew len
2645+ else liftIO $ Unboxed.new len
23142646 putSliceUnsafe (arrContents arr1) start1 newArrContents 0 len1
23152647 putSliceUnsafe (arrContents arr2) start2 newArrContents len1 len2
23162648 return $ MutArray newArrContents 0 len len
@@ -2322,7 +2654,7 @@ spliceCopy arr1 arr2 = liftIO $ do
23222654spliceUnsafe :: MonadIO m =>
23232655 MutArray a -> MutArray a -> m (MutArray a)
23242656spliceUnsafe dst src =
2325− liftIO $ do
2657+ do
23262658 let startSrc = arrStart src
23272659 srcLen = arrEnd src - startSrc
23282660 endDst = arrEnd dst
@@ -2344,7 +2676,7 @@ spliceWith :: forall m a. (MonadIO m, Unbox a) =>
23442676 (Int -> Int -> Int) -> MutArray a -> MutArray a -> m (MutArray a)
23452677spliceWith sizer dst@(MutArray _ start end bound) src = do
23462678{-
2347− let f = writeAppendWith (`sizer` byteLength src) (return dst)
2679+ let f = appendWith (`sizer` byteLength src) (return dst)
23482680 in D.fold f (toStreamD src)
23492681-}
23502682 assert (end <= bound) (return ())
@@ -2360,7 +2692,7 @@ spliceWith sizer dst@(MutArray _ start end bound) src = do
23602692 $ "splice: newSize is less than the total size "
23612693 ++ "of arrays being appended. Please check the "
23622694 ++ "sizer function passed."
2363− liftIO $ realloc newSizeInBytes dst
2695+ realloc newSizeInBytes dst
23642696 else return dst
23652697 spliceUnsafe dst1 src
23662698
@@ -2412,7 +2744,7 @@ splitOn predicate arr =
24122744{-# INLINE breakOn #-}
24132745breakOn :: MonadIO m
24142746 => Word8 -> MutArray Word8 -> m (MutArray Word8, Maybe (MutArray Word8))
2415−breakOn sep arr@MutArray{..} = asUnpinnedPtrUnsafe arr $ \p -> liftIO $ do
2747+breakOn sep arr@MutArray{..} = unsafeAsPtr arr $ \p -> liftIO $ do
24162748 -- XXX We do not need memchr here, we can use a Haskell equivalent.
24172749 -- Need efficient stream based primitives that work on Word64.
24182750 loc <- c_memchr p sep (fromIntegral $ byteLength arr)
@@ -2435,33 +2767,41 @@ breakOn sep arr@MutArray{..} = asUnpinnedPtrUnsafe arr $ \p -> liftIO $ do
24352767 }
24362768 )
24372769
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+
24382792-- | Create two slices of an array without copying the original array. The
24392793-- specified index @i@ is the first index of the second slice.
24402794--
24412795{-# INLINE splitAt #-}
24422796splitAt :: forall a. Unbox a => Int -> MutArray a -> (MutArray a, MutArray a)
2443−splitAt i arr@MutArray{..} =
2797+splitAt i arr =
24442798 let maxIndex = length arr - 1
24452799 in if i < 0
24462800 then error "sliceAt: negative array index"
24472801 else if i > maxIndex
24482802 then error $ "sliceAt: specified array index " ++ show i
24492803 ++ " is beyond the maximum index " ++ show maxIndex
2450− else let off = i * SIZE_OF(a)
2451− p = arrStart + off
2452− in ( MutArray
2453− { arrContents = arrContents
2454− , arrStart = arrStart
2455− , arrEnd = p
2456− , arrBound = p
2457− }
2458− , MutArray
2459− { arrContents = arrContents
2460− , arrStart = p
2461− , arrEnd = arrEnd
2462− , arrBound = arrBound
2463− }
2464− )
2804+ else unsafeSplitAt i arr
24652805
24662806-------------------------------------------------------------------------------
24672807-- Casting
@@ -2511,9 +2851,13 @@ cast arr =
25112851-- guarantees that unsafe calls will be performed in the calling thread. Making
25122852-- it safe to pass heap-allocated objects to unsafe functions.
25132853
2514−-- 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
25152855-- operations. And we are safe from Haskell perspective because we will be
25162856-- 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.
25172861
25182862-- | Use a @MutArray a@ as @Ptr a@. This is useful when we want to pass an
25192863-- array as a pointer to some operating system call or to a "safe" FFI call.
@@ -2531,16 +2875,21 @@ cast arr =
25312875--
25322876-- /Pre-release/
25332877--
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." #-}
25342885{-# INLINE asPtrUnsafe #-}
25352886asPtrUnsafe :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2536−asPtrUnsafe arr f =
2537− Unboxed.asPtrUnsafe
2538− (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
2887+asPtrUnsafe = unsafePinnedAsPtr
25392888
2540−{-# INLINE asUnpinnedPtrUnsafe #-}
2541−asUnpinnedPtrUnsafe :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2542−asUnpinnedPtrUnsafe arr f =
2543− Unboxed.asUnpinnedPtrUnsafe
2889+{-# INLINE unsafeAsPtr #-}
2890+unsafeAsPtr :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2891+unsafeAsPtr arr f =
2892+ Unboxed.unsafeAsPtr
25442893 (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
25452894
25462895-------------------------------------------------------------------------------
@@ -2558,28 +2907,26 @@ asUnpinnedPtrUnsafe arr f =
25582907-- /Pre-release/
25592908{-# INLINE byteCmp #-}
25602909byteCmp :: MonadIO m => MutArray a -> MutArray a -> m Ordering
2561−byteCmp arr1 arr2 =
2562− liftIO
2563− $ do
2564− let marr1 = getMutableByteArray# (arrContents arr1)
2565− marr2 = getMutableByteArray# (arrContents arr2)
2566− !(I# st1#) = arrStart arr1
2567− !(I# st2#) = arrStart arr2
2568− !(I# len#) = byteLength arr1
2569− case compare (byteLength arr1) (byteLength arr2) of
2570− EQ -> do
2571− r <- liftIO $ IO $ \s# ->
2572− let res =
2573− I#
2574− (compareByteArrays#
2575− (unsafeCoerce# marr1)
2576− st1#
2577− (unsafeCoerce# marr2)
2578− st2#
2579− len#)
2580− in (# s#, res #)
2581− return $ compare r 0
2582− 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
25832930
25842931{-# INLINE cmp #-}
25852932{-# DEPRECATED cmp "Please use byteCmp instead." #-}
@@ -2599,46 +2946,48 @@ byteEq arr1 arr2 = fmap (EQ ==) $ byteCmp arr1 arr2
25992946-- Compact
26002947-------------------------------------------------------------------------------
26012948
2602−-- | Parser @pCompactChunksLE n@ coalesces adjacent arrays in the input stream
2603−-- only if the combined size would be less than or equal to n.
2604−--
2605−-- /Internal/
2606−{-# INLINE_NORMAL pCompactChunksLE #-}
2607−pCompactChunksLE ::
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 ::
26082956 forall m a. (MonadIO m, Unbox a)
2609− => Int -> Parser (MutArray a) m (MutArray a)
2610−pCompactChunksLE n = Parser step initial extract
2957+ => PinnedState -> Int -> Parser (MutArray a) m (MutArray a)
2958+pCompactLeAs ps maxElems = Parser step initial extract
26112959
26122960 where
26132961
2614− nBytes = n * SIZE_OF(a)
2962+ maxBytes = maxElems * SIZE_OF(a)
26152963
2616− functionName = "Streamly.Internal.Data.MutArray.pCompactChunksLE"
2964+ functionName = "Streamly.Internal.Data.MutArray.pCompactLE"
26172965
26182966 initial =
26192967 return
2620− $ if n <= 0
2968+ $ if maxElems <= 0
26212969 then error
26222970 $ functionName
26232971 ++ ": the size of arrays ["
2624− ++ show n ++ "] must be a natural number"
2972+ ++ show maxElems ++ "] must be a natural number"
26252973 else Parser.IPartial Nothing
26262974
26272975 step Nothing arr =
26282976 return
26292977 $ let len = byteLength arr
2630− in if len >= nBytes
2978+ in if len >= maxBytes
26312979 then Parser.Done 0 arr
26322980 else Parser.Partial 0 (Just arr)
2981+ -- XXX Split the last array to use the space more compactly.
26332982 step (Just buf) arr =
26342983 let len = byteLength buf + byteLength arr
2635− in if len > nBytes
2984+ in if len > maxBytes
26362985 then return $ Parser.Done 1 buf
26372986 else do
2638− -- XXX check this in the Nothing case
26392987 buf1 <-
2640− if byteCapacity buf < nBytes
2641− then liftIO $ realloc nBytes buf
2988+ if byteCapacity buf < maxBytes
2989+ then liftIO $ reallocExplicitAs
2990+ ps (SIZE_OF(a)) maxBytes buf
26422991 else return buf
26432992 buf2 <- spliceUnsafe buf1 arr
26442993 return $ Parser.Partial 0 (Just buf2)
@@ -2646,6 +2995,30 @@ pCompactChunksLE n = Parser step initial extract
26462995 extract Nothing = return $ Parser.Done 0 nil
26472996 extract (Just buf) = return $ Parser.Done 0 buf
26482997
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+
26493022data SpliceState s arr
26503023 = SpliceInitial s
26513024 | SpliceBuffering s arr
@@ -2657,48 +3030,47 @@ data SpliceState s arr
26573030-- immutable array never has additional space so a new array is allocated
26583031-- instead of mutating it.
26593032
2660−-- | Scan @rCompactChunksLE n@ coalesces adjacent arrays in the input stream
2661−-- only if the combined size would be less than or equal to n.
2662−--
2663−-- /Internal/
2664−{-# INLINE_NORMAL rCompactChunksLE #-}
2665−rCompactChunksLE :: (MonadIO m, Unbox a)
2666− => Int -> D.Stream m (MutArray a) -> D.Stream m (MutArray a)
2667−rCompactChunksLE n (D.Stream step state) =
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) =
26683037 D.Stream step' (SpliceInitial state)
26693038
26703039 where
26713040
2672− functionName = "Streamly.Internal.Data.MutArray.rCompactChunksLE"
3041+ maxBytes = maxElems * SIZE_OF(a)
3042+
3043+ functionName = "Streamly.Internal.Data.MutArray.rCompactLE"
26733044
26743045 {-# INLINE_LATE step' #-}
26753046 step' gst (SpliceInitial st) = do
2676− when (n <= 0) $
3047+ when (maxElems <= 0) $
26773048 -- XXX we can pass the module string from the higher level API
2678− error $ functionName ++ ": the size of arrays [" ++ show n
3049+ error $ functionName ++ ": the size of arrays [" ++ show maxElems
26793050 ++ "] must be a natural number"
26803051 r <- step gst st
26813052 case r of
26823053 D.Yield arr s -> return $
26833054 let len = byteLength arr
2684− in if len >= n
3055+ in if len >= maxBytes
26853056 then D.Skip (SpliceYielding arr (SpliceInitial s))
26863057 else D.Skip (SpliceBuffering s arr)
26873058 D.Skip s -> return $ D.Skip (SpliceInitial s)
26883059 D.Stop -> return D.Stop
26893060
3061+ -- XXX Split the last array to use the space more compactly.
26903062 step' gst (SpliceBuffering st buf) = do
26913063 r <- step gst st
26923064 case r of
26933065 D.Yield arr s -> do
26943066 let len = byteLength buf + byteLength arr
2695− if len > n
3067+ if len > maxBytes
26963068 then return $
26973069 D.Skip (SpliceYielding buf (SpliceBuffering s arr))
26983070 else do
2699− -- XXX check this in the SpliceInitial case
2700− buf1 <- if byteCapacity buf < n
2701− then liftIO $ realloc n buf
3071+ buf1 <- if byteCapacity buf < maxBytes
3072+ then liftIO $ reallocExplicitAs
3073+ ps (SIZE_OF(a)) maxBytes buf
27023074 else return buf
27033075 buf2 <- spliceUnsafe buf1 arr
27043076 return $ D.Skip (SpliceBuffering s buf2)
@@ -2709,113 +3081,124 @@ rCompactChunksLE n (D.Stream step state) =
27093081
27103082 step' _ (SpliceYielding arr next) = return $ D.Yield arr next
27113083
2712−-- | Fold @fCompactChunksGE n@ coalesces adjacent arrays in the input stream
2713−-- until the size becomes greater than or equal to n.
2714−--
2715−{-# INLINE_NORMAL fCompactChunksGE #-}
2716−fCompactChunksGE ::
3084+
3085+{-# INLINE_NORMAL fCompactGeAs #-}
3086+fCompactGeAs ::
27173087 forall m a. (MonadIO m, Unbox a)
2718− => Int -> FL.Fold m (MutArray a) (MutArray a)
2719−fCompactChunksGE n = Fold step initial extract extract
3088+ => PinnedState -> Int -> FL.Fold m (MutArray a) (MutArray a)
3089+fCompactGeAs ps minElems = Fold step initial extract extract
27203090
27213091 where
27223092
2723− nBytes = n * SIZE_OF(a)
3093+ minBytes = minElems * SIZE_OF(a)
27243094
2725− functionName = "Streamly.Internal.Data.MutArray.fCompactChunksGE"
3095+ functionName = "Streamly.Internal.Data.MutArray.fCompactGE"
27263096
27273097 initial =
27283098 return
2729− $ if n < 0
3099+ $ if minElems < 0
27303100 then error
27313101 $ functionName
27323102 ++ ": the size of arrays ["
2733− ++ show n ++ "] must be a natural number"
3103+ ++ show minElems ++ "] must be a natural number"
27343104 else FL.Partial Nothing
27353105
27363106 step Nothing arr =
27373107 return
27383108 $ let len = byteLength arr
2739− in if len >= nBytes
3109+ in if len >= minBytes
27403110 then FL.Done arr
27413111 else FL.Partial (Just arr)
3112+ -- XXX Buffer arrays as a list to avoid copy and reallocations
27423113 step (Just buf) arr = do
27433114 let len = byteLength buf + byteLength arr
27443115 buf1 <-
27453116 if byteCapacity buf < len
2746− then liftIO $ realloc (max len nBytes) buf
3117+ then liftIO $ reallocExplicitAs
3118+ ps (SIZE_OF(a)) (max minBytes len) buf
27473119 else return buf
27483120 buf2 <- spliceUnsafe buf1 arr
2749− if len >= n
3121+ if len >= minBytes
27503122 then return $ FL.Done buf2
27513123 else return $ FL.Partial (Just buf2)
27523124
27533125 extract Nothing = return nil
27543126 extract (Just buf) = return buf
27553127
2756−-- | Like 'compactChunksGE' but for transforming folds instead of stream.
2757−--
2758−-- >>> lCompactChunksGE n = Fold.many (MutArray.fCompactChunksGE n)
3128+-- | Fold @fCompactGE minElems@ coalesces adjacent arrays in the input stream
3129+-- until the size becomes greater than or equal to @minElems@.
27593130--
2760−{-# INLINE_NORMAL lCompactChunksGE #-}
2761−lCompactChunksGE :: (MonadIO m, Unbox a)
2762− => Int -> Fold m (MutArray a) () -> Fold m (MutArray a) ()
2763−-- lCompactChunksGE n = Fold.many (fCompactChunksGE n)
2764−lCompactChunksGE n (Fold step1 initial1 _ final1) =
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) =
27653152 Fold step initial extract final
27663153
27673154 where
27683155
2769− functionName = "Streamly.Internal.Data.MutArray.lCompactChunksGE"
3156+ minBytes = minElems * SIZE_OF(a)
3157+
3158+ functionName = "Streamly.Internal.Data.MutArray.lCompactGE"
27703159
27713160 initial = do
2772− when (n <= 0) $
3161+ when (minElems <= 0) $
27733162 -- XXX we can pass the module string from the higher level API
27743163 error $ functionName ++ ": the size of arrays ["
2775− ++ show n ++ "] must be a natural number"
3164+ ++ show minElems ++ "] must be a natural number"
27763165
27773166 r <- initial1
27783167 return $ first (Tuple' Nothing) r
27793168
2780− step (Tuple' Nothing r1) arr =
2781− let len = byteLength arr
2782− in if len >= n
2783− then do
2784− r <- step1 r1 arr
2785− case r of
2786− FL.Done _ -> return $ FL.Done ()
2787− FL.Partial s -> do
2788− _ <- final1 s
2789− res <- initial1
2790− return $ first (Tuple' Nothing) res
2791− else return $ FL.Partial $ Tuple' (Just arr) r1
2792−
2793− step (Tuple' (Just buf) r1) arr = do
2794− let len = byteLength buf + byteLength arr
2795− buf1 <- if byteCapacity buf < len
2796− then liftIO $ realloc (max n len) buf
2797− else return buf
2798− buf2 <- spliceUnsafe buf1 arr
2799−
2800− -- XXX this is common in both the equations of step
2801− if len >= n
3169+ {-# INLINE runInner #-}
3170+ runInner len acc buf =
3171+ if len >= minBytes
28023172 then do
2803− r <- step1 r1 buf2
3173+ r <- step1 acc buf
28043174 case r of
28053175 FL.Done _ -> return $ FL.Done ()
28063176 FL.Partial s -> do
28073177 _ <- final1 s
28083178 res <- initial1
28093179 return $ first (Tuple' Nothing) res
2810− else return $ FL.Partial $ Tuple' (Just buf2) r1
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
28113194
28123195 -- XXX Several folds do extract >=> final, therefore, we need to make final
2813− -- return "m b" rather than using extract post it if we want extract to be
3196+ -- return "m b" rather than using extract post it if we want extract to be
28143197 -- partial.
28153198 --
28163199 -- extract forces the pending buffer to be sent to the fold which is not
28173200 -- what we want.
2818− extract _ = error "lCompactChunksGE: not designed for scanning"
3201+ extract _ = error "lCompactGE: not designed for scanning"
28193202
28203203 final (Tuple' Nothing r1) = final1 r1
28213204 final (Tuple' (Just buf) r1) = do
@@ -2824,25 +3207,42 @@ lCompactChunksGE n (Fold step1 initial1 _ final1) =
28243207 FL.Partial rr -> final1 rr
28253208 FL.Done _ -> return ()
28263209
2827−-- | @compactChunksGE n stream@ coalesces adjacent arrays in the @stream@ until
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
28283228-- the size becomes greater than or equal to @n@.
28293229--
2830−-- >>> compactChunksGE n = Stream.foldMany (MutArray.fCompactChunksGE n)
3230+-- >>> compactGE n = Stream.foldMany (MutArray.fCompactGE n)
28313231--
2832−{-# INLINE compactChunksGE #-}
2833−compactChunksGE ::
3232+{-# INLINE compactGE #-}
3233+compactGE ::
28343234 (MonadIO m, Unbox a)
28353235 => Int -> Stream m (MutArray a) -> Stream m (MutArray a)
2836−compactChunksGE n = D.foldMany (fCompactChunksGE n)
3236+compactGE n = D.foldMany (fCompactGE n)
28373237
2838−-- | 'compactChunksEQ n' coalesces adajacent arrays in the input stream to
3238+-- | 'compactEQ n' coalesces adajacent arrays in the input stream to
28393239-- arrays of exact size @n@.
28403240--
28413241-- /Unimplemented/
2842−{-# INLINE compactChunksEQ #-}
2843−compactChunksEQ :: -- (MonadIO m, Unbox a) =>
3242+{-# INLINE compactEQ #-}
3243+compactEQ :: -- (MonadIO m, Unbox a) =>
28443244 Int -> Stream m (MutArray a) -> Stream m (MutArray a)
2845−compactChunksEQ _n = undefined -- D.parseManyD (pCompactChunksEQ n)
3245+compactEQ _n = undefined -- D.parseManyD (pCompactEQ n)
28463246
28473247-------------------------------------------------------------------------------
28483248-- In-place mutation algorithms