@@ -31,39 +31,57 @@ module Streamly.Internal.Data.MutArray.Type
3131 -- ** Type
3232 -- $arrayNotes
3333 MutArray (..)
34− , MutByteArray
35− , MutableByteArray
34+
35+ -- ** Conversion
36+ -- *** Pinned and Unpinned
3637 , pin
3738 , unpin
3839 , isPinned
3940
40− -- -- * Constructing and Writing
41+ -- ** Casting
42+ , cast
43+ , castUnsafe -- XXX unsafeCast
44+ , asBytes
45+ , unsafePinnedAsPtr
46+ , unsafeAsPtr
47+
4148 -- ** Construction
42− , 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
4363
44− -- *** Uninitialized Arrays
45− , pinnedNew
46− , pinnedNewBytes
47− , pinnedNewAligned
48− , new
49− , 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
5070
51− -- *** From streams
71+ -- *** Stream Folds
5272 , ArrayUnsafe (..)
53− , writeNWithUnsafe
54− , writeNWith
55− , writeNUnsafe
56− , pinnedWriteNUnsafe
57− , writeN
58− , pinnedWriteN
59− , pinnedWriteNAligned
60−
61− , writeWith
62− , write
63− , pinnedWrite
64−
65− , writeRevN
66− -- , writeRev
73+ , unsafeCreateOfWith
74+ , unsafeCreateOf
75+ , unsafePinnedCreateOf
76+ , pinnedCreateOf
77+ , createOfWith
78+ , createOf
79+ , revCreateOf
80+
81+ , pinnedCreate
82+ , createWith
83+ , create
84+ -- , revCreate
6785
6886 -- *** From containers
6987 , fromListN
@@ -79,11 +97,11 @@ module Streamly.Internal.Data.MutArray.Type
7997 , fromByteStr#
8098 , fromPtrN
8199 , fromChunksK
82− , fromChunksRealloced
100+ , fromChunksRealloced -- fromSmallChunks
83101
84102 -- ** Random writes
85103 , putIndex
86− , putIndexUnsafe
104+ , putIndexUnsafe -- XXX unsafePutIndex
87105 , putIndices
88106 -- , putFromThenTo
89107 -- , putFrom -- start writing at the given position
@@ -91,77 +109,70 @@ module Streamly.Internal.Data.MutArray.Type
91109 -- , putFromTo
92110 -- , putFromRev
93111 -- , putUptoRev
94− , modifyIndexUnsafe
112+ , modifyIndexUnsafe -- XXX unsafeModifyIndex
95113 , modifyIndex
96114 , modifyIndices
97115 , modify
98116 , swapIndices
99117 , unsafeSwapIndices
100118
101− -- ** Growing and Shrinking
102− -- | Arrays grow only at the end, though it is possible to grow on both sides
103− -- and therefore have a cons as well as snoc. But that will require two
104− -- bounds in the array representation.
105−
106− -- *** Appending elements
107− , snocWith
108− , snoc
109− , snocLinear
110− , snocMay
111− , snocUnsafe
112−
113− -- *** Appending streams
114− , writeAppendNUnsafe
115− , writeAppendN
116− , writeAppendWith
117− , writeAppend
118−
119− -- ** Eliminating and Reading
119+ -- ** Reading
120120
121− -- *** To streams
122− , reader
123− , readerRevWith
124− , readerRev
121+ -- *** Indexing
122+ , getIndex
123+ , getIndexUnsafe -- XXX unsafeGetIndex
124+ -- , getFromThenTo
125+ , getIndexRev -- getRevIndex?
126+ , indexReader
127+ , indexReaderWith
125128
126− -- *** To containers
127− , toStreamDWith
128− , toStreamDRevWith
129− , toStreamKWith
130− , toStreamKRevWith
129+ -- *** To Streams
131130 , read
132131 , readRev
132+ , toStreamWith
133+ , toStreamRevWith
133134 , toStreamK
135+ , toStreamKWith
134136 , toStreamKRev
137+ , toStreamKRevWith
138+
139+ -- *** To Containers
135140 , toList
136141
142+ -- *** Unfolds
137143 -- experimental
138144 , producerWith
139145 , producer
140146
141− -- *** Random reads
142− , getIndex
143− , getIndexUnsafe
144− , getIndices
145− , getIndicesD
146− -- , getFromThenTo
147− , getIndexRev
148−
149− -- ** Memory Management
150− , blockSize
151− , arrayChunkBytes
152− , allocBytesToElemCount
153− , realloc
154− , resize
155− , resizeExp
156− , rightSize
147+ , reader
148+ , readerRevWith
149+ , readerRev
157150
158− -- ** Size
151+ -- ** Size and Capacity
152+ -- *** Size
159153 , length
160154 , byteLength
155+
156+ -- *** Capacity
161157 -- , capacity
162158 , byteCapacity
163159 , bytesFree
164160
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+
165176 -- ** In-place Mutation Algorithms
166177 , strip
167178 , reverse
@@ -172,58 +183,82 @@ module Streamly.Internal.Data.MutArray.Type
172183 , mergeBy
173184 , bubble
174185
175− -- ** Casting
176− , cast
177− , castUnsafe
178− , asBytes
179− , asPtrUnsafe
180− , 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.
181190
182− -- ** Folding
183− , foldl'
184− , foldr
185− , cmp
186− , byteCmp
187− , 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
188213
189− -- ** Arrays of arrays
214+ -- ** Serialization using Unbox
215+ , pokeAppend
216+ , pokeAppendMay
217+ , pokeSkipUnsafe -- unsafePokeSkip
218+
219+ -- ** Deserialization using Unbox
220+ , peekUncons
221+ , peekUnconsUnsafe -- unsafePeekUncons
222+ , peekSkipUnsafe -- unsafePeekSkip
223+
224+ -- Arrays of arrays
190225 -- We can add dimensionality parameter to the array type to get
191226 -- multidimensional arrays. Multidimensional arrays would just be a
192227 -- convenience wrapper on top of single dimensional arrays.
193228
194− -- | Operations dealing with multiple arrays, streams of arrays or
195− -- multidimensional array representations.
196−
197− -- *** Construct from streams
229+ -- ** Streams of Arrays
230+ -- *** Chunk
231+ -- | Group a stream into arrays.
198232 , chunksOf
199233 , pinnedChunksOf
200234 , buildChunks
201235
202− -- *** Eliminate to streams
203− , concatChunks
204− , concatChunksRev
236+ -- *** Split
237+ -- | Split an array into slices.
205238
206− -- *** Construct from arrays
207− -- get chunks without copying
208− , getSliceUnsafe
209− , getSlice
210239 -- , getSlicesFromLenN
211− , splitAt -- XXX should be able to express using getSlice
212− , breakOn
213−
214− -- ** Cloning arrays
215− , clone
216− , pinnedClone
217−
218− -- ** Appending arrays
219− , spliceCopy
220− , spliceWith
221− , splice
222− , spliceExp
223− , spliceUnsafe
224− -- , putSlice
225− -- , appendSlice
226− -- , appendSliceFrom
240+ , splitOn
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
227262
228263 -- ** Utilities
229264 , roundUpToPower2
@@ -232,12 +267,37 @@ module Streamly.Internal.Data.MutArray.Type
232267 , c_memchr
233268
234269 -- * Deprecated
270+ , asPtrUnsafe
235271 , writeChunks
236272 , flattenArrays
237273 , flattenArraysRev
238274 , fromArrayStreamK
239275 , fromStreamDN
240276 , 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
241301 )
242302where
243303
@@ -248,6 +308,7 @@ where
248308
249309import Control.Monad (when, void)
250310import Control.Monad.IO.Class (MonadIO(..))
311+import Data.Bifunctor (first)
251312import Data.Bits (shiftR, (.|.), (.&.))
252313import Data.Functor.Identity (Identity(..))
253314import Data.Proxy (Proxy(..))
@@ -256,7 +317,6 @@ import Foreign.C.Types (CSize(..), CInt(..))
256317import Foreign.Ptr (plusPtr, minusPtr, nullPtr)
257318import Streamly.Internal.Data.MutByteArray.Type
258319 ( MutByteArray(..)
259− , MutableByteArray
260320 , PinnedState(..)
261321 , getMutableByteArray#
262322 , putSliceUnsafe
@@ -275,13 +335,17 @@ import GHC.Ptr (Ptr(..))
275335import Streamly.Internal.Data.Fold.Type (Fold(..))
276336import Streamly.Internal.Data.Producer.Type (Producer (..))
277337import Streamly.Internal.Data.Stream.Type (Stream)
338+import Streamly.Internal.Data.Parser.Type (Parser (..))
278339import Streamly.Internal.Data.StreamK.Type (StreamK)
279340import Streamly.Internal.Data.SVar.Type (adaptState, defState)
341+import Streamly.Internal.Data.Tuple.Strict (Tuple'(..))
280342import Streamly.Internal.Data.Unfold.Type (Unfold(..))
281343import Streamly.Internal.System.IO (arrayPayloadSize, defaultChunkSize)
282344
283345import qualified Streamly.Internal.Data.Fold.Type as FL
284346import 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
285349import qualified Streamly.Internal.Data.Producer as Producer
286350import qualified Streamly.Internal.Data.Stream.Type as D
287351import qualified Streamly.Internal.Data.Stream.Lift as D
@@ -289,7 +353,7 @@ import qualified Streamly.Internal.Data.StreamK.Type as K
289353import qualified Prelude
290354
291355import Prelude hiding
292− (Foldable(..), read, unlines, splitAt, reverse, truncate)
356+ (Foldable(..), concat, read, unlines, splitAt, reverse, truncate)
293357
294358#include "DocTestDataMutArray.hs"
295359
@@ -415,6 +479,8 @@ isPinned MutArray{..} = Unboxed.isPinned arrContents
415479-- we need more than that. See stg_pinnedNewAlignedByteArrayzh and
416480-- allocatePinned in GHC source.
417481
482+-- XXX Rename to emptyAlignedWith, alignSize should be first arg.
483+
418484-- | @newArrayWith allocator alignment count@ allocates a new array of zero
419485-- length and with a capacity to hold @count@ elements, using @allocator
420486-- size alignment@ as the memory allocator function.
@@ -438,12 +504,25 @@ newArrayWith alloc alignSize count = do
438504 , arrBound = size
439505 }
440506
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." #-}
441520nil ::
442521#ifdef DEVBUILD
443522 Unbox a =>
444523#endif
445524 MutArray a
446−nil = MutArray Unboxed.nil 0 0 0
525+nil = empty
447526
448527{-# INLINE newBytesAs #-}
449528newBytesAs :: MonadIO m =>
@@ -460,12 +539,13 @@ newBytesAs ps bytes = do
460539 , arrBound = bytes
461540 }
462541
463−-- | Allocates a pinned empty array that can hold 'count' items. The memory of
464−-- the array is uninitialized and the allocation is aligned as per the
465−-- '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.
466545--
467546-- /Pre-release/
468547{-# INLINE pinnedNewBytes #-}
548+{-# DEPRECATED pinnedNewBytes "Please use pinnedEmptyOf with appropriate calculation" #-}
469549pinnedNewBytes :: MonadIO m =>
470550#ifdef DEVBUILD
471551 Unbox a =>
@@ -490,21 +570,31 @@ newAs ps =
490570 (error "new: alignment is not used in unpinned arrays.")
491571
492572-- XXX can unaligned allocation be more efficient when alignment is not needed?
493−--
494−-- | Allocates an empty pinned array that can hold 'count' items. The memory of
495−-- the array is uninitialized and the allocation is aligned as per the 'Unboxed'
496−-- instance of the type.
497−--
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." #-}
498582{-# INLINE pinnedNew #-}
499583pinnedNew :: forall m a. (MonadIO m, Unbox a) => Int -> m (MutArray a)
500−pinnedNew = newAs Pinned
584+pinnedNew = pinnedEmptyOf
501585
502−-- | Allocates an empty unpinned array that can hold 'count' items. The memory
503−-- of the array is uninitialized.
586+-- | Allocates an unpinned array of zero length but growable to the specified
587+-- capacity without reallocation.
504588--
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." #-}
505595{-# INLINE new #-}
506596new :: (MonadIO m, Unbox a) => Int -> m (MutArray a)
507−new = newAs Unpinned
597+new = emptyOf
508598
509599-------------------------------------------------------------------------------
510600-- Random writes
@@ -552,7 +642,7 @@ putIndices arr = FL.foldlM' step (return ())
552642
553643 where
554644
555− step () (i, x) = liftIO (putIndex i arr x)
645+ step () (i, x) = putIndex i arr x
556646
557647-- | Modify a given index of an array using a modifier function.
558648--
@@ -748,22 +838,19 @@ arrayChunkBytes = 1024
748838roundDownTo :: Int -> Int -> Int
749839roundDownTo elemSize size = size - (size `mod` elemSize)
750840
751−-- XXX See if resizing can be implemented by reading the old array as a stream
752−-- and then using writeN to the new array.
753−--
754841-- NOTE: we are passing elemSize explicitly to avoid an Unboxed constraint.
755842-- Since this is not inlined Unboxed consrraint leads to dictionary passing
756843-- which complicates some inspection tests.
757844--
758−{-# NOINLINE reallocExplicit #-}
759−reallocExplicit :: Int -> Int -> MutArray a -> IO (MutArray a)
760−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
761848 assertM(arrEnd <= arrBound)
762849
763850 -- Allocate new array
764851 let newCapMaxInBytes = roundUpLargeArray newCapacityInBytes
765852 contents <-
766− if Unboxed.isPinned arrContents
853+ if ps == Pinned
767854 then Unboxed.pinnedNew newCapMaxInBytes
768855 else Unboxed.new newCapMaxInBytes
769856 let !(MutByteArray mbarrFrom#) = arrContents
@@ -788,6 +875,10 @@ reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
788875 , arrBound = newCapInBytes
789876 }
790877
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+
791882-- | @realloc newCapacity array@ reallocates the array to the specified
792883-- capacity in bytes.
793884--
@@ -799,7 +890,12 @@ reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
799890-- If the original array is pinned, the newly allocated array is also pinned.
800891{-# INLINABLE realloc #-}
801892realloc :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m (MutArray a)
802−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
803899
804900-- | @reallocWith label capSizer minIncrBytes array@. The label is used
805901-- in error messages and the capSizer is used to determine the capacity of the
@@ -822,14 +918,14 @@ reallocWith label capSizer minIncrBytes arr = do
822918 where
823919
824920 badSize newSize =
825− concat
921+ Prelude.concat
826922 [ label
827923 , ": new array size (in bytes) is less than required size "
828924 , show newSize
829925 , ". Please check the sizing function passed."
830926 ]
831927
832−-- | @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
833929-- it is enough to hold the specified number of elements. Nothing is done if
834930-- the specified capacity is less than the length of the array.
835931--
@@ -837,34 +933,46 @@ reallocWith label capSizer minIncrBytes arr = do
837933-- the block size (4K).
838934--
839935-- /Pre-release/
840−{-# INLINE resize #-}
841−resize :: forall m a. (MonadIO m, Unbox a) =>
936+{-# INLINE grow #-}
937+grow :: forall m a. (MonadIO m, Unbox a) =>
842938 Int -> MutArray a -> m (MutArray a)
843−resize nElems arr@MutArray{..} = do
939+grow nElems arr@MutArray{..} = do
844940 let req = SIZE_OF(a) * nElems
845− len = arrEnd - arrStart
846− if req < len
941+ cap = arrBound - arrStart
942+ if req < cap
847943 then return arr
848944 else realloc req arr
849945
850−-- | Like 'resize' but if the byte capacity is more than 'largeObjectThreshold'
851−-- 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.
852954--
853955-- /Pre-release/
854−{-# INLINE resizeExp #-}
855−resizeExp :: forall m a. (MonadIO m, Unbox a) =>
956+{-# INLINE growExp #-}
957+growExp :: forall m a. (MonadIO m, Unbox a) =>
856958 Int -> MutArray a -> m (MutArray a)
857−resizeExp nElems arr@MutArray{..} = do
959+growExp nElems arr@MutArray{..} = do
858960 let req = roundUpLargeArray (SIZE_OF(a) * nElems)
859961 req1 =
860962 if req > largeObjectThreshold
861963 then roundUpToPower2 req
862964 else req
863− len = arrEnd - arrStart
864− if req1 < len
965+ cap = arrBound - arrStart
966+ if req1 < cap
865967 then return arr
866968 else realloc req1 arr
867969
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+
868976-- | Resize the allocated memory to drop any reserved free space at the end of
869977-- the array and reallocate it to reduce wastage.
870978--
@@ -929,7 +1037,7 @@ snocUnsafe arr@MutArray{..} = snocNewEnd (INDEX_NEXT(arrEnd,a)) arr
9291037{-# INLINE snocMay #-}
9301038snocMay :: forall m a. (MonadIO m, Unbox a) =>
9311039 MutArray a -> a -> m (Maybe (MutArray a))
932−snocMay arr@MutArray{..} x = liftIO $ do
1040+snocMay arr@MutArray{..} x = do
9331041 let newEnd = INDEX_NEXT(arrEnd,a)
9341042 if newEnd <= arrBound
9351043 then Just <$> snocNewEnd newEnd arr x
@@ -943,7 +1051,7 @@ snocWithRealloc :: forall m a. (MonadIO m, Unbox a) =>
9431051 -> a
9441052 -> m (MutArray a)
9451053snocWithRealloc sizer arr x = do
946− arr1 <- liftIO $ reallocWith "snocWith" sizer (SIZE_OF(a)) arr
1054+ arr1 <- reallocWith "snocWith" sizer (SIZE_OF(a)) arr
9471055 snocUnsafe arr1 x
9481056
9491057-- | @snocWith sizer arr elem@ mutates @arr@ to append @elem@. The length of
@@ -965,7 +1073,7 @@ snocWith :: forall m a. (MonadIO m, Unbox a) =>
9651073 -> MutArray a
9661074 -> a
9671075 -> m (MutArray a)
968−snocWith allocSize arr x = liftIO $ do
1076+snocWith allocSize arr x = do
9691077 let newEnd = INDEX_NEXT(arrEnd arr,a)
9701078 if newEnd <= arrBound arr
9711079 then snocNewEnd newEnd arr x
@@ -1009,6 +1117,126 @@ snoc = snocWith f
10091117 then oldSize * 2
10101118 else roundUpToPower2 oldSize * 2
10111119
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+
10121240-------------------------------------------------------------------------------
10131241-- Random reads
10141242-------------------------------------------------------------------------------
@@ -1051,15 +1279,10 @@ getIndexRev i MutArray{..} = do
10511279data GetIndicesState contents start end st =
10521280 GetIndicesState contents start end st
10531281
1054−-- | Given an unfold that generates array indices, read the elements on those
1055−-- indices from the supplied MutArray. An error is thrown if an index is out of
1056−-- bounds.
1057−--
1058−-- /Pre-release/
1059−{-# INLINE getIndicesD #-}
1060−getIndicesD :: (Monad m, Unbox a) =>
1282+{-# INLINE indexReaderWith #-}
1283+indexReaderWith :: (Monad m, Unbox a) =>
10611284 (forall b. IO b -> m b) -> D.Stream m Int -> Unfold m (MutArray a) a
1062−getIndicesD liftio (D.Stream stepi sti) = Unfold step inject
1285+indexReaderWith liftio (D.Stream stepi sti) = Unfold step inject
10631286
10641287 where
10651288
@@ -1078,9 +1301,26 @@ getIndicesD liftio (D.Stream stepi sti) = Unfold step inject
10781301 D.Skip s -> return $ D.Skip (GetIndicesState contents start end s)
10791302 D.Stop -> return D.Stop
10801303
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+
1310+-- | Given an unfold that generates array indices, read the elements on those
1311+-- indices from the supplied MutArray. An error is thrown if an index is out of
1312+-- bounds.
1313+--
1314+-- /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." #-}
10811321{-# INLINE getIndices #-}
10821322getIndices :: (MonadIO m, Unbox a) => Stream m Int -> Unfold m (MutArray a) a
1083−getIndices = getIndicesD liftIO
1323+getIndices = indexReader
10841324
10851325-------------------------------------------------------------------------------
10861326-- Subarrays
@@ -1375,7 +1615,7 @@ chunksOfAs ps n (D.Stream step state) =
13751615--
13761616-- Same as the following but may be more efficient:
13771617--
1378−-- >>> chunksOf n = Stream.foldMany (MutArray.writeN n)
1618+-- >>> chunksOf n = Stream.foldMany (MutArray.createOf n)
13791619--
13801620-- /Pre-release/
13811621{-# INLINE_NORMAL chunksOf #-}
@@ -1384,14 +1624,14 @@ chunksOf :: forall m a. (MonadIO m, Unbox a)
13841624-- XXX the idiomatic implementation leads to large regression in the D.reverse'
13851625-- benchmark. It seems it has difficulty producing optimized code when
13861626-- converting to StreamK. Investigate GHC optimizations.
1387−-- chunksOf n = D.foldMany (writeN n)
1627+-- chunksOf n = D.foldMany (createOf n)
13881628chunksOf = chunksOfAs Unpinned
13891629
13901630-- | Like 'chunksOf' but creates pinned arrays.
13911631{-# INLINE_NORMAL pinnedChunksOf #-}
13921632pinnedChunksOf :: forall m a. (MonadIO m, Unbox a)
13931633 => Int -> D.Stream m a -> D.Stream m (MutArray a)
1394−-- pinnedChunksOf n = D.foldMany (pinnedWriteN n)
1634+-- pinnedChunksOf n = D.foldMany (pinnedCreateOf n)
13951635pinnedChunksOf = chunksOfAs Pinned
13961636
13971637-- | When we are buffering a stream of unknown size into an array we do not
@@ -1424,16 +1664,10 @@ data FlattenState s contents a =
14241664 OuterLoop s
14251665 | InnerLoop s contents !Int !Int
14261666
1427−-- | Use the "reader" unfold instead.
1428−--
1429−-- @concatChunks = unfoldMany reader@
1430−--
1431−-- We can try this if there are any fusion issues in the unfold.
1432−--
1433−{-# INLINE_NORMAL concatChunks #-}
1434−concatChunks :: forall m a. (MonadIO m, Unbox a)
1435− => D.Stream m (MutArray a) -> D.Stream m a
1436−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)
14371671
14381672 where
14391673
@@ -1450,24 +1684,30 @@ concatChunks (D.Stream step state) = D.Stream step' (OuterLoop state)
14501684 return $ D.Skip $ OuterLoop st
14511685
14521686 step' _ (InnerLoop st contents p end) = do
1453− x <- liftIO $ peekAt p contents
1687+ x <- liftio $ peekAt p contents
14541688 return $ D.Yield x (InnerLoop st contents (INDEX_NEXT(p,a)) end)
14551689
1456−{-# DEPRECATED flattenArrays "Please use \"unfoldMany reader\" instead." #-}
1457−flattenArrays :: forall m a. (MonadIO m, Unbox a)
1458− => D.Stream m (MutArray a) -> D.Stream m a
1459−flattenArrays = concatChunks
1460−
1461−-- | Use the "readerRev" unfold instead.
1690+-- | Use the "reader" unfold instead.
14621691--
1463−-- @concatChunks = unfoldMany readerRev@
1692+-- @concat = unfoldMany reader@
14641693--
14651694-- We can try this if there are any fusion issues in the unfold.
14661695--
1467−{-# INLINE_NORMAL concatChunksRev #-}
1468−concatChunksRev :: forall m a. (MonadIO m, Unbox a)
1696+{-# INLINE_NORMAL concat #-}
1697+concat :: forall m a. (MonadIO m, Unbox a)
14691698 => D.Stream m (MutArray a) -> D.Stream m a
1470−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)
14711711
14721712 where
14731713
@@ -1485,14 +1725,26 @@ concatChunksRev (D.Stream step state) = D.Stream step' (OuterLoop state)
14851725 return $ D.Skip $ OuterLoop st
14861726
14871727 step' _ (InnerLoop st contents p start) = do
1488− x <- liftIO $ peekAt p contents
1728+ x <- liftio $ peekAt p contents
14891729 let cur = INDEX_PREV(p,a)
14901730 return $ D.Yield x (InnerLoop st contents cur start)
14911731
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+
14921743{-# DEPRECATED flattenArraysRev "Please use \"unfoldMany readerRev\" instead." #-}
1744+{-# INLINE flattenArraysRev #-}
14931745flattenArraysRev :: forall m a. (MonadIO m, Unbox a)
14941746 => D.Stream m (MutArray a) -> D.Stream m a
1495−flattenArraysRev = concatChunksRev
1747+flattenArraysRev = concatRev
14961748
14971749-------------------------------------------------------------------------------
14981750-- Unfolds
@@ -1611,11 +1863,11 @@ toList MutArray{..} = liftIO $ go arrStart
16111863 x <- peekAt p arrContents
16121864 (:) x <$> go (INDEX_NEXT(p,a))
16131865
1614−{-# INLINE_NORMAL toStreamDWith #-}
1615−toStreamDWith ::
1866+{-# INLINE_NORMAL toStreamWith #-}
1867+toStreamWith ::
16161868 forall m a. (Monad m, Unbox a)
16171869 => (forall b. IO b -> m b) -> MutArray a -> D.Stream m a
1618−toStreamDWith liftio MutArray{..} = D.Stream step arrStart
1870+toStreamWith liftio MutArray{..} = D.Stream step arrStart
16191871
16201872 where
16211873
@@ -1631,7 +1883,7 @@ toStreamDWith liftio MutArray{..} = D.Stream step arrStart
16311883--
16321884{-# INLINE_NORMAL read #-}
16331885read :: forall m a. (MonadIO m, Unbox a) => MutArray a -> D.Stream m a
1634−read = toStreamDWith liftIO
1886+read = toStreamWith liftIO
16351887
16361888{-# INLINE toStreamKWith #-}
16371889toStreamKWith ::
@@ -1650,11 +1902,11 @@ toStreamKWith liftio MutArray{..} = go arrStart
16501902toStreamK :: forall m a. (MonadIO m, Unbox a) => MutArray a -> StreamK m a
16511903toStreamK = toStreamKWith liftIO
16521904
1653−{-# INLINE_NORMAL toStreamDRevWith #-}
1654−toStreamDRevWith ::
1905+{-# INLINE_NORMAL toStreamRevWith #-}
1906+toStreamRevWith ::
16551907 forall m a. (Monad m, Unbox a)
16561908 => (forall b. IO b -> m b) -> MutArray a -> D.Stream m a
1657−toStreamDRevWith liftio MutArray{..} =
1909+toStreamRevWith liftio MutArray{..} =
16581910 let p = INDEX_PREV(arrEnd,a)
16591911 in D.Stream step p
16601912
@@ -1672,7 +1924,7 @@ toStreamDRevWith liftio MutArray{..} =
16721924--
16731925{-# INLINE_NORMAL readRev #-}
16741926readRev :: forall m a. (MonadIO m, Unbox a) => MutArray a -> D.Stream m a
1675−readRev = toStreamDRevWith liftIO
1927+readRev = toStreamRevWith liftIO
16761928
16771929{-# INLINE toStreamKRevWith #-}
16781930toStreamKRevWith ::
@@ -1721,7 +1973,7 @@ foldr f z arr = D.foldr f z $ read arr
17211973
17221974-- XXX Keep the bound intact to not lose any free space? Perf impact?
17231975
1724−-- | @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
17251977-- array.
17261978--
17271979-- Unsafe: Do not drive the fold beyond @n@ elements, it will lead to memory
@@ -1730,13 +1982,12 @@ foldr f z arr = D.foldr f z $ read arr
17301982-- Any free space left in the array after appending @n@ elements is lost.
17311983--
17321984-- /Internal/
1733−{-# INLINE_NORMAL writeAppendNUnsafe #-}
1734−writeAppendNUnsafe :: forall m a. (MonadIO m, Unbox a) =>
1985+{-# INLINE_NORMAL unsafeAppendN #-}
1986+unsafeAppendN :: forall m a. (MonadIO m, Unbox a) =>
17351987 Int
17361988 -> m (MutArray a)
17371989 -> Fold m a (MutArray a)
1738−writeAppendNUnsafe n action =
1739− fmap fromArrayUnsafe $ FL.foldlM' step initial
1990+unsafeAppendN n action = fmap fromArrayUnsafe $ FL.foldlM' step initial
17401991
17411992 where
17421993
@@ -1749,7 +2000,7 @@ writeAppendNUnsafe n action =
17492000 -- otherwise we lose that space.
17502001 arr1 <-
17512002 if free < needed
1752− then noinline reallocWith "writeAppendNUnsafeWith" (+ needed) needed arr
2003+ then noinline reallocWith "unsafeAppendN" (+ needed) needed arr
17532004 else return arr
17542005 return $ toArrayUnsafe arr1
17552006
@@ -1757,30 +2008,51 @@ writeAppendNUnsafe n action =
17572008 liftIO $ pokeAt end contents x
17582009 return $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
17592010
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+
17602019-- | Append @n@ elements to an existing array. Any free space left in the array
17612020-- after appending @n@ elements is lost.
17622021--
1763−-- >>> writeAppendN n initial = Fold.take n (MutArray.writeAppendNUnsafe n initial)
2022+-- >>> appendN n initial = Fold.take n (MutArray.unsafeAppendN n initial)
17642023--
1765−{-# 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 #-}
17662032writeAppendN :: forall m a. (MonadIO m, Unbox a) =>
17672033 Int -> m (MutArray a) -> Fold m a (MutArray a)
1768−writeAppendN n initial = FL.take n (writeAppendNUnsafe n initial)
2034+writeAppendN = appendN
17692035
1770−-- | @writeAppendWith realloc action@ mutates the array generated by @action@ to
2036+-- | @appendWith realloc action@ mutates the array generated by @action@ to
17712037-- append the input stream. If there is no reserved space available in the
17722038-- array it is reallocated to a size in bytes determined by @realloc oldSize@,
17732039-- where @oldSize@ is the current size of the array in bytes.
17742040--
17752041-- Note that the returned array may be a mutated version of original array.
17762042--
1777−-- >>> writeAppendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
2043+-- >>> appendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
17782044--
17792045-- /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." #-}
17802052{-# INLINE writeAppendWith #-}
17812053writeAppendWith :: forall m a. (MonadIO m, Unbox a) =>
17822054 (Int -> Int) -> m (MutArray a) -> Fold m a (MutArray a)
1783−writeAppendWith sizer = FL.foldlM' (snocWith sizer)
2055+writeAppendWith = appendWith
17842056
17852057-- | @append action@ mutates the array generated by @action@ to append the
17862058-- input stream. If there is no reserved space available in the array it is
@@ -1788,26 +2060,33 @@ writeAppendWith sizer = FL.foldlM' (snocWith sizer)
17882060--
17892061-- Note that the returned array may be a mutated version of original array.
17902062--
1791−-- >>> writeAppend = MutArray.writeAppendWith (* 2)
2063+-- >>> append = MutArray.appendWith (* 2)
17922064--
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." #-}
17932072{-# INLINE writeAppend #-}
17942073writeAppend :: forall m a. (MonadIO m, Unbox a) =>
17952074 m (MutArray a) -> Fold m a (MutArray a)
1796−writeAppend = writeAppendWith (* 2)
2075+writeAppend = append
17972076
17982077-- XXX We can carry bound as well in the state to make sure we do not lose the
17992078-- remaining capacity. Need to check perf impact.
18002079--
1801−-- | Like 'writeNUnsafe' but takes a new array allocator @alloc size@ function
1802−-- as argument.
2080+-- | Like 'unsafeCreateOf' but takes a new array allocator @alloc size@
2081+-- function as argument.
18032082--
1804−-- >>> writeNWithUnsafe alloc n = MutArray.writeAppendNUnsafe (alloc n) n
2083+-- >>> unsafeCreateOfWith alloc n = MutArray.unsafeAppendN (alloc n) n
18052084--
18062085-- /Pre-release/
1807−{-# INLINE_NORMAL writeNWithUnsafe #-}
1808−writeNWithUnsafe :: forall m a. (MonadIO m, Unbox a)
2086+{-# INLINE_NORMAL unsafeCreateOfWith #-}
2087+unsafeCreateOfWith :: forall m a. (MonadIO m, Unbox a)
18092088 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
1810−writeNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
2089+unsafeCreateOfWith alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
18112090
18122091 where
18132092
@@ -1818,40 +2097,64 @@ writeNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
18182097 return
18192098 $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
18202099
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+
18212106{-# INLINE_NORMAL writeNUnsafeAs #-}
18222107writeNUnsafeAs :: forall m a. (MonadIO m, Unbox a)
18232108 => PinnedState -> Int -> Fold m a (MutArray a)
1824−writeNUnsafeAs ps = writeNWithUnsafe (newAs ps)
2109+writeNUnsafeAs ps = unsafeCreateOfWith (newAs ps)
18252110
1826−-- | 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
18272112-- driver must not call the step function more than 'n' times otherwise it will
18282113-- corrupt the memory and crash. This function exists mainly because any
18292114-- conditional in the step function blocks fusion causing 10x performance
18302115-- slowdown.
18312116--
1832−-- >>> writeNUnsafe = MutArray.writeNWithUnsafe MutArray.new
2117+-- >>> unsafeCreateOf = MutArray.unsafeCreateOfWith MutArray.emptyOf
18332118--
1834−{-# 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 #-}
18352126writeNUnsafe :: forall m a. (MonadIO m, Unbox a)
18362127 => Int -> Fold m a (MutArray a)
1837−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
18382135
1839−-- | Like 'writeNUnsafe' but creates a pinned array.
1840−{-# INLINE_NORMAL pinnedWriteNUnsafe #-}
2136+{-# DEPRECATED pinnedWriteNUnsafe "Please use unsafePinnedCreateOf instead." #-}
2137+{-# INLINE pinnedWriteNUnsafe #-}
18412138pinnedWriteNUnsafe :: forall m a. (MonadIO m, Unbox a)
18422139 => Int -> Fold m a (MutArray a)
1843−pinnedWriteNUnsafe = writeNUnsafeAs Pinned
2140+pinnedWriteNUnsafe = unsafePinnedCreateOf
18442141
1845−-- | @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
18462143-- allocated using the @alloc@ function.
18472144--
1848−-- >>> writeNWith alloc n = Fold.take n (MutArray.writeNWithUnsafe alloc n)
1849−-- >>> 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
18502147--
1851−{-# 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 #-}
18522155writeNWith :: forall m a. (MonadIO m, Unbox a)
18532156 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
1854−writeNWith alloc n = FL.take n (writeNWithUnsafe alloc n)
2157+writeNWith = createOfWith
18552158
18562159{-# INLINE_NORMAL writeNAs #-}
18572160writeNAs ::
@@ -1859,28 +2162,42 @@ writeNAs ::
18592162 => PinnedState
18602163 -> Int
18612164 -> Fold m a (MutArray a)
1862−writeNAs ps = writeNWith (newAs ps)
2165+writeNAs ps = createOfWith (newAs ps)
18632166
1864−-- | @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
18652168-- 'MutArray'.
18662169--
1867−-- >>> writeN = MutArray.writeNWith MutArray.new
1868−-- >>> writeN n = Fold.take n (MutArray.writeNUnsafe n)
1869−-- >>> 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)
18702173--
1871−{-# 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 #-}
18722181writeN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
1873−writeN = writeNAs Unpinned
2182+writeN = createOf
18742183
1875−-- | Like 'writeN' but creates a pinned array.
1876−{-# 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 #-}
18772194pinnedWriteN ::
18782195 forall m a. (MonadIO m, Unbox a)
18792196 => Int
18802197 -> Fold m a (MutArray a)
1881−pinnedWriteN = writeNAs Pinned
2198+pinnedWriteN = pinnedCreateOf
18822199
1883−-- | Like writeNWithUnsafe but writes the array in reverse order.
2200+-- | Like unsafeCreateOfWith but writes the array in reverse order.
18842201--
18852202-- /Internal/
18862203{-# INLINE_NORMAL writeRevNWithUnsafe #-}
@@ -1901,7 +2218,7 @@ writeRevNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
19012218 return
19022219 $ ArrayUnsafe contents ptr end
19032220
1904−-- | Like writeNWith but writes the array in reverse order.
2221+-- | Like createOfWith but writes the array in reverse order.
19052222--
19062223-- /Internal/
19072224{-# INLINE_NORMAL writeRevNWith #-}
@@ -1909,31 +2226,36 @@ writeRevNWith :: forall m a. (MonadIO m, Unbox a)
19092226 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
19102227writeRevNWith alloc n = FL.take n (writeRevNWithUnsafe alloc n)
19112228
1912−-- | Like writeN but writes the array in reverse order.
2229+-- | Like 'createOf' but writes the array in reverse order.
19132230--
19142231-- /Pre-release/
1915−{-# 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 #-}
19162238writeRevN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
1917−writeRevN = writeRevNWith new
2239+writeRevN = revCreateOf
19182240
19192241-- | @pinnedWriteNAligned align n@ folds a maximum of @n@ elements from the
19202242-- input stream to a 'MutArray' aligned to the given size.
19212243--
1922−-- >>> pinnedWriteNAligned align = MutArray.writeNWith (MutArray.pinnedNewAligned align)
1923−-- >>> 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)
19242246--
19252247-- /Pre-release/
19262248--
19272249{-# INLINE_NORMAL pinnedWriteNAligned #-}
19282250pinnedWriteNAligned :: forall m a. (MonadIO m, Unbox a)
19292251 => Int -> Int -> Fold m a (MutArray a)
1930−pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
2252+pinnedWriteNAligned align = createOfWith (pinnedNewAligned align)
19312253
19322254-- XXX Buffer to a list instead?
19332255
19342256-- | Buffer a stream into a stream of arrays.
19352257--
1936−-- >>> buildChunks n = Fold.many (MutArray.writeN n) Fold.toStreamK
2258+-- >>> buildChunks n = Fold.many (MutArray.createOf n) Fold.toStreamK
19372259--
19382260-- Breaking an array into an array stream can be useful to consume a large
19392261-- array sequentially such that memory of the array is released incrementatlly.
@@ -1945,9 +2267,10 @@ pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
19452267{-# INLINE_NORMAL buildChunks #-}
19462268buildChunks :: (MonadIO m, Unbox a) =>
19472269 Int -> Fold m a (StreamK n (MutArray a))
1948−buildChunks n = FL.many (writeN n) FL.toStreamK
2270+buildChunks n = FL.many (createOf n) FL.toStreamK
19492271
19502272{-# DEPRECATED writeChunks "Please use buildChunks instead." #-}
2273+{-# INLINE writeChunks #-}
19512274writeChunks :: (MonadIO m, Unbox a) =>
19522275 Int -> Fold m a (StreamK n (MutArray a))
19532276writeChunks = buildChunks
@@ -1955,29 +2278,29 @@ writeChunks = buildChunks
19552278{-# INLINE_NORMAL writeWithAs #-}
19562279writeWithAs :: forall m a. (MonadIO m, Unbox a)
19572280 => PinnedState -> Int -> Fold m a (MutArray a)
1958−-- writeWithAs ps n = FL.rmapM rightSize $ writeAppendWith (* 2) (newAs ps n)
2281+-- writeWithAs ps n = FL.rmapM rightSize $ appendWith (* 2) (newAs ps n)
19592282writeWithAs ps elemCount =
19602283 FL.rmapM extract $ FL.foldlM' step initial
19612284
19622285 where
19632286
19642287 initial = do
1965− when (elemCount < 0) $ error "writeWith: elemCount is negative"
1966− liftIO $ newAs ps elemCount
2288+ when (elemCount < 0) $ error "createWith: elemCount is negative"
2289+ newAs ps elemCount
19672290
19682291 step arr@(MutArray _ start end bound) x
19692292 | INDEX_NEXT(end,a) > bound = do
19702293 let oldSize = end - start
19712294 newSize = max (oldSize * 2) 1
1972− arr1 <- liftIO $ reallocExplicit (SIZE_OF(a)) newSize arr
2295+ arr1 <- liftIO $ reallocExplicitAs ps (SIZE_OF(a)) newSize arr
19732296 snocUnsafe arr1 x
19742297 step arr x = snocUnsafe arr x
19752298
19762299 extract = liftIO . rightSize
19772300
1978−-- XXX Compare writeWith with fromStreamD which uses an array of streams
2301+-- XXX Compare createWith with fromStreamD which uses an array of streams
19792302-- implementation. We can write this using buildChunks above if that is faster.
1980−-- If writeWith is faster then we should use that to implement
2303+-- If createWith is faster then we should use that to implement
19812304-- fromStreamD.
19822305--
19832306-- XXX The realloc based implementation needs to make one extra copy if we use
@@ -1992,39 +2315,56 @@ writeWithAs ps elemCount =
19922315-- XXX check if GHC's memory allocator is efficient enough. We can try the C
19932316-- malloc to compare against.
19942317
1995−-- | @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
19962319-- starts at a size big enough to hold minCount elements, the size is doubled
19972320-- every time the array needs to be grown.
19982321--
19992322-- /Caution! Do not use this on infinite streams./
20002323--
2001−-- >>> f n = MutArray.writeAppendWith (* 2) (MutArray.new n)
2002−-- >>> writeWith n = Fold.rmapM MutArray.rightSize (f n)
2003−-- >>> 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)
20042327--
20052328-- /Pre-release/
2006−{-# 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 #-}
20072337writeWith :: forall m a. (MonadIO m, Unbox a)
20082338 => Int -> Fold m a (MutArray a)
2009−-- writeWith n = FL.rmapM rightSize $ writeAppendWith (* 2) (new n)
2010−writeWith = writeWithAs Unpinned
2339+writeWith = createWith
20112340
20122341-- | Fold the whole input to a single array.
20132342--
2014−-- Same as 'writeWith' using an initial array size of 'arrayChunkBytes' bytes
2343+-- Same as 'createWith' using an initial array size of 'arrayChunkBytes' bytes
20152344-- rounded up to the element size.
20162345--
20172346-- /Caution! Do not use this on infinite streams./
20182347--
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." #-}
20192354{-# INLINE write #-}
20202355write :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2021−write = writeWith (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2356+write = create
20222357
2023−-- | 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." #-}
20242365{-# INLINE pinnedWrite #-}
20252366pinnedWrite :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
2026−pinnedWrite =
2027− writeWithAs Pinned (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2367+pinnedWrite = pinnedCreate
20282368
20292369-------------------------------------------------------------------------------
20302370-- construct from streams, known size
@@ -2034,7 +2374,7 @@ pinnedWrite =
20342374fromStreamDNAs :: forall m a. (MonadIO m, Unbox a)
20352375 => PinnedState -> Int -> D.Stream m a -> m (MutArray a)
20362376fromStreamDNAs ps limit str = do
2037− (arr :: MutArray a) <- liftIO $ newAs ps limit
2377+ (arr :: MutArray a) <- newAs ps limit
20382378 end <- D.foldlM' (fwrite (arrContents arr)) (return $ arrEnd arr) $ D.take limit str
20392379 return $ arr {arrEnd = end}
20402380
@@ -2044,17 +2384,18 @@ fromStreamDNAs ps limit str = do
20442384 liftIO $ pokeAt ptr arrContents x
20452385 return $ INDEX_NEXT(ptr,a)
20462386
2047−-- | Use the 'writeN' fold instead.
2387+-- | Use the 'createOf' fold instead.
20482388--
2049−-- >>> fromStreamN n = Stream.fold (MutArray.writeN n)
2389+-- >>> fromStreamN n = Stream.fold (MutArray.createOf n)
20502390--
20512391{-# INLINE_NORMAL fromStreamN #-}
20522392fromStreamN :: forall m a. (MonadIO m, Unbox a)
20532393 => Int -> D.Stream m a -> m (MutArray a)
2054−-- fromStreamDN n = D.fold (writeN n)
2394+-- fromStreamDN n = D.fold (createOf n)
20552395fromStreamN = fromStreamDNAs Unpinned
20562396
20572397{-# DEPRECATED fromStreamDN "Please use fromStreamN instead." #-}
2398+{-# INLINE fromStreamDN #-}
20582399fromStreamDN :: forall m a. (MonadIO m, Unbox a)
20592400 => Int -> D.Stream m a -> m (MutArray a)
20602401fromStreamDN = fromStreamN
@@ -2077,20 +2418,20 @@ pinnedFromListN n xs = fromStreamDNAs Pinned n $ D.fromList xs
20772418-- /Pre-release/
20782419{-# INLINE fromListRevN #-}
20792420fromListRevN :: (MonadIO m, Unbox a) => Int -> [a] -> m (MutArray a)
2080−fromListRevN n xs = D.fold (writeRevN n) $ D.fromList xs
2421+fromListRevN n xs = D.fold (revCreateOf n) $ D.fromList xs
20812422
20822423-- | Convert a pure stream in Identity monad to a mutable array.
20832424{-# INLINABLE fromPureStreamN #-}
20842425fromPureStreamN :: (MonadIO m, Unbox a) =>
20852426 Int -> Stream Identity a -> m (MutArray a)
20862427fromPureStreamN n xs =
2087− liftIO $ D.fold (writeN n) $ D.morphInner (return . runIdentity) xs
2428+ D.fold (createOf n) $ D.morphInner (return . runIdentity) xs
20882429
20892430-- | Convert a pure stream in Identity monad to a mutable array.
20902431{-# INLINABLE fromPureStream #-}
20912432fromPureStream :: (MonadIO m, Unbox a) => Stream Identity a -> m (MutArray a)
20922433fromPureStream xs =
2093− liftIO $ D.fold write $ D.morphInner (return . runIdentity) xs
2434+ D.fold create $ D.morphInner (return . runIdentity) xs
20942435
20952436{-# INLINABLE fromPtrN #-}
20962437fromPtrN :: MonadIO m => Int -> Ptr Word8 -> m (MutArray Word8)
@@ -2099,7 +2440,7 @@ fromPtrN len addr = do
20992440 -- XXX We can implement a stream copy in a similar way by streaming Word64
21002441 -- first and then remaining Word8.
21012442 arr <- new len
2102− _ <- asUnpinnedPtrUnsafe arr
2443+ _ <- unsafeAsPtr arr
21032444 (\ptr -> liftIO $ c_memcpy ptr addr (fromIntegral len))
21042445 return (arr {arrEnd = len})
21052446
@@ -2115,7 +2456,7 @@ fromByteStr# addr = do
21152456 len <- liftIO $ c_strlen (Ptr addr)
21162457 let lenInt = fromIntegral len
21172458 arr <- new lenInt
2118− _ <- asUnpinnedPtrUnsafe arr (\ptr -> liftIO $ c_memcpy ptr (Ptr addr) len)
2459+ _ <- unsafeAsPtr arr (\ptr -> liftIO $ c_memcpy ptr (Ptr addr) len)
21192460 return (arr {arrEnd = lenInt})
21202461
21212462-------------------------------------------------------------------------------
@@ -2136,7 +2477,7 @@ fromChunksRealloced s = do
21362477 arr <- D.foldlM' spliceExp (pure a) strm
21372478 -- Reallocation is exponential so there may be 50% empty space in
21382479 -- worst case. One more reallocation to reclaim the space.
2139− liftIO $ rightSize arr
2480+ rightSize arr
21402481 Nothing -> pure nil
21412482
21422483-------------------------------------------------------------------------------
@@ -2174,6 +2515,7 @@ fromChunksK :: (Unbox a, MonadIO m) =>
21742515fromChunksK = fromChunkskAs Unpinned
21752516
21762517{-# DEPRECATED fromArrayStreamK "Please use fromChunksK instead." #-}
2518+{-# INLINE fromArrayStreamK #-}
21772519fromArrayStreamK :: (Unbox a, MonadIO m) =>
21782520 StreamK m (MutArray a) -> m (MutArray a)
21792521fromArrayStreamK = fromChunksK
@@ -2185,7 +2527,7 @@ fromStreamDAs ps m =
21852527 arrayStreamKFromStreamDAs Unpinned m >>= fromChunkskAs ps
21862528
21872529-- | Create an 'Array' from a stream. This is useful when we want to create a
2188−-- 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
21892531-- as efficient when the size is already known.
21902532--
21912533-- Note that if the input stream is too large memory allocation for the array
@@ -2197,19 +2539,19 @@ fromStreamDAs ps m =
21972539fromStream :: (MonadIO m, Unbox a) => Stream m a -> m (MutArray a)
21982540fromStream = fromStreamDAs Unpinned
21992541
2200−-- fromStream (Stream m) = P.fold write m
2542+-- fromStream (Stream m) = P.fold create m
22012543-- CAUTION: a very large number (millions) of arrays can degrade performance
22022544-- due to GC overhead because we need to buffer the arrays before we flatten
22032545-- all the arrays.
22042546--
2205−-- XXX Compare if this is faster or "fold write".
2547+-- XXX Compare if this is faster or "fold create".
22062548--
22072549-- | We could take the approach of doubling the memory allocation on each
22082550-- overflow. This would result in more or less the same amount of copying as in
22092551-- the chunking approach. However, if we have to shrink in the end then it may
22102552-- result in an extra copy of the entire data.
22112553--
2212−-- >>> fromStreamD = StreamD.fold MutArray.write
2554+-- >>> fromStreamD = StreamD.fold MutArray.create
22132555--
22142556{-# INLINE fromStreamD #-}
22152557{-# DEPRECATED fromStreamD "Please use fromStream instead." #-}
@@ -2239,6 +2581,10 @@ fromListRev xs = fromListRevN (Prelude.length xs) xs
22392581-- Cloning
22402582-------------------------------------------------------------------------------
22412583
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+
22422588{-# INLINE cloneAs #-}
22432589cloneAs ::
22442590 ( MonadIO m
@@ -2248,7 +2594,7 @@ cloneAs ::
22482594 )
22492595 => PinnedState -> MutArray a -> m (MutArray a)
22502596cloneAs ps src =
2251− liftIO $ do
2597+ do
22522598 let startSrc = arrStart src
22532599 srcLen = arrEnd src - startSrc
22542600 newArrContents <-
@@ -2287,17 +2633,16 @@ spliceCopy :: forall m a. MonadIO m =>
22872633 Unbox a =>
22882634#endif
22892635 MutArray a -> MutArray a -> m (MutArray a)
2290−spliceCopy arr1 arr2 = liftIO $ do
2636+spliceCopy arr1 arr2 = do
22912637 let start1 = arrStart arr1
22922638 start2 = arrStart arr2
22932639 len1 = arrEnd arr1 - start1
22942640 len2 = arrEnd arr2 - start2
2295− let newLen = len1 + len2
2641+ let len = len1 + len2
22962642 newArrContents <-
22972643 if Unboxed.isPinned (arrContents arr1)
2298− then Unboxed.pinnedNew newLen
2299− else Unboxed.new newLen
2300− let len = len1 + len2
2644+ then liftIO $ Unboxed.pinnedNew len
2645+ else liftIO $ Unboxed.new len
23012646 putSliceUnsafe (arrContents arr1) start1 newArrContents 0 len1
23022647 putSliceUnsafe (arrContents arr2) start2 newArrContents len1 len2
23032648 return $ MutArray newArrContents 0 len len
@@ -2309,7 +2654,7 @@ spliceCopy arr1 arr2 = liftIO $ do
23092654spliceUnsafe :: MonadIO m =>
23102655 MutArray a -> MutArray a -> m (MutArray a)
23112656spliceUnsafe dst src =
2312− liftIO $ do
2657+ do
23132658 let startSrc = arrStart src
23142659 srcLen = arrEnd src - startSrc
23152660 endDst = arrEnd dst
@@ -2331,7 +2676,7 @@ spliceWith :: forall m a. (MonadIO m, Unbox a) =>
23312676 (Int -> Int -> Int) -> MutArray a -> MutArray a -> m (MutArray a)
23322677spliceWith sizer dst@(MutArray _ start end bound) src = do
23332678{-
2334− let f = writeAppendWith (`sizer` byteLength src) (return dst)
2679+ let f = appendWith (`sizer` byteLength src) (return dst)
23352680 in D.fold f (toStreamD src)
23362681-}
23372682 assert (end <= bound) (return ())
@@ -2347,7 +2692,7 @@ spliceWith sizer dst@(MutArray _ start end bound) src = do
23472692 $ "splice: newSize is less than the total size "
23482693 ++ "of arrays being appended. Please check the "
23492694 ++ "sizer function passed."
2350− liftIO $ realloc newSizeInBytes dst
2695+ realloc newSizeInBytes dst
23512696 else return dst
23522697 spliceUnsafe dst1 src
23532698
@@ -2384,11 +2729,22 @@ spliceExp = spliceWith (\l1 l2 -> max (l1 * 2) (l1 + l2))
23842729-- Splitting
23852730-------------------------------------------------------------------------------
23862731
2732+-- | Generate a stream of array slices using a predicate. The array element
2733+-- matching the predicate is dropped.
2734+--
2735+-- /Pre-release/
2736+{-# INLINE splitOn #-}
2737+splitOn :: (MonadIO m, Unbox a) =>
2738+ (a -> Bool) -> MutArray a -> Stream m (MutArray a)
2739+splitOn predicate arr =
2740+ fmap (\(i, len) -> getSliceUnsafe i len arr)
2741+ $ D.indexOnSuffix predicate (read arr)
2742+
23872743-- | Drops the separator byte
23882744{-# INLINE breakOn #-}
23892745breakOn :: MonadIO m
23902746 => Word8 -> MutArray Word8 -> m (MutArray Word8, Maybe (MutArray Word8))
2391−breakOn sep arr@MutArray{..} = asUnpinnedPtrUnsafe arr $ \p -> liftIO $ do
2747+breakOn sep arr@MutArray{..} = unsafeAsPtr arr $ \p -> liftIO $ do
23922748 -- XXX We do not need memchr here, we can use a Haskell equivalent.
23932749 -- Need efficient stream based primitives that work on Word64.
23942750 loc <- c_memchr p sep (fromIntegral $ byteLength arr)
@@ -2411,32 +2767,41 @@ breakOn sep arr@MutArray{..} = asUnpinnedPtrUnsafe arr $ \p -> liftIO $ do
24112767 }
24122768 )
24132769
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+
24142792-- | Create two slices of an array without copying the original array. The
24152793-- specified index @i@ is the first index of the second slice.
24162794--
2795+{-# INLINE splitAt #-}
24172796splitAt :: forall a. Unbox a => Int -> MutArray a -> (MutArray a, MutArray a)
2418−splitAt i arr@MutArray{..} =
2797+splitAt i arr =
24192798 let maxIndex = length arr - 1
24202799 in if i < 0
24212800 then error "sliceAt: negative array index"
24222801 else if i > maxIndex
24232802 then error $ "sliceAt: specified array index " ++ show i
24242803 ++ " is beyond the maximum index " ++ show maxIndex
2425− else let off = i * SIZE_OF(a)
2426− p = arrStart + off
2427− in ( MutArray
2428− { arrContents = arrContents
2429− , arrStart = arrStart
2430− , arrEnd = p
2431− , arrBound = p
2432− }
2433− , MutArray
2434− { arrContents = arrContents
2435− , arrStart = p
2436− , arrEnd = arrEnd
2437− , arrBound = arrBound
2438− }
2439− )
2804+ else unsafeSplitAt i arr
24402805
24412806-------------------------------------------------------------------------------
24422807-- Casting
@@ -2486,9 +2851,13 @@ cast arr =
24862851-- guarantees that unsafe calls will be performed in the calling thread. Making
24872852-- it safe to pass heap-allocated objects to unsafe functions.
24882853
2489−-- 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
24902855-- operations. And we are safe from Haskell perspective because we will be
24912856-- 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.
24922861
24932862-- | Use a @MutArray a@ as @Ptr a@. This is useful when we want to pass an
24942863-- array as a pointer to some operating system call or to a "safe" FFI call.
@@ -2506,16 +2875,21 @@ cast arr =
25062875--
25072876-- /Pre-release/
25082877--
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." #-}
25092885{-# INLINE asPtrUnsafe #-}
25102886asPtrUnsafe :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2511−asPtrUnsafe arr f =
2512− Unboxed.asPtrUnsafe
2513− (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
2887+asPtrUnsafe = unsafePinnedAsPtr
25142888
2515−{-# INLINE asUnpinnedPtrUnsafe #-}
2516−asUnpinnedPtrUnsafe :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2517−asUnpinnedPtrUnsafe arr f =
2518− Unboxed.asUnpinnedPtrUnsafe
2889+{-# INLINE unsafeAsPtr #-}
2890+unsafeAsPtr :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2891+unsafeAsPtr arr f =
2892+ Unboxed.unsafeAsPtr
25192893 (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
25202894
25212895-------------------------------------------------------------------------------
@@ -2533,28 +2907,26 @@ asUnpinnedPtrUnsafe arr f =
25332907-- /Pre-release/
25342908{-# INLINE byteCmp #-}
25352909byteCmp :: MonadIO m => MutArray a -> MutArray a -> m Ordering
2536−byteCmp arr1 arr2 =
2537− liftIO
2538− $ do
2539− let marr1 = getMutableByteArray# (arrContents arr1)
2540− marr2 = getMutableByteArray# (arrContents arr2)
2541− !(I# st1#) = arrStart arr1
2542− !(I# st2#) = arrStart arr2
2543− !(I# len#) = byteLength arr1
2544− case compare (byteLength arr1) (byteLength arr2) of
2545− EQ -> do
2546− r <- liftIO $ IO $ \s# ->
2547− let res =
2548− I#
2549− (compareByteArrays#
2550− (unsafeCoerce# marr1)
2551− st1#
2552− (unsafeCoerce# marr2)
2553− st2#
2554− len#)
2555− in (# s#, res #)
2556− return $ compare r 0
2557− 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
25582930
25592931{-# INLINE cmp #-}
25602932{-# DEPRECATED cmp "Please use byteCmp instead." #-}
@@ -2571,7 +2943,309 @@ byteEq :: MonadIO m => MutArray a -> MutArray a -> m Bool
25712943byteEq arr1 arr2 = fmap (EQ ==) $ byteCmp arr1 arr2
25722944
25732945-------------------------------------------------------------------------------
2574−-- 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
25753249-------------------------------------------------------------------------------
25763250
25773251-- | Strip elements which match with predicate from both ends.