@@ -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
@@ -72,13 +90,18 @@ module Streamly.Internal.Data.MutArray.Type
7290 , pinnedFromList
7391 , fromListRevN
7492 , fromListRev
75− , fromStreamDN
76− , fromStreamD
93+ , fromStreamN
94+ , fromStream
95+ , fromPureStreamN
7796 , fromPureStream
97+ , fromByteStr#
98+ , fromPtrN
99+ , fromChunksK
100+ , fromChunksRealloced -- fromSmallChunks
78101
79102 -- ** Random writes
80103 , putIndex
81− , putIndexUnsafe
104+ , putIndexUnsafe -- XXX unsafePutIndex
82105 , putIndices
83106 -- , putFromThenTo
84107 -- , putFrom -- start writing at the given position
@@ -86,77 +109,70 @@ module Streamly.Internal.Data.MutArray.Type
86109 -- , putFromTo
87110 -- , putFromRev
88111 -- , putUptoRev
89− , modifyIndexUnsafe
112+ , modifyIndexUnsafe -- XXX unsafeModifyIndex
90113 , modifyIndex
91114 , modifyIndices
92115 , modify
93116 , swapIndices
94117 , unsafeSwapIndices
95118
96− -- ** Growing and Shrinking
97− -- | Arrays grow only at the end, though it is possible to grow on both sides
98− -- and therefore have a cons as well as snoc. But that will require two
99− -- bounds in the array representation.
119+ -- ** Reading
100120
101− -- *** Appending elements
102− , snocWith
103− , snoc
104− , snocLinear
105− , snocMay
106− , snocUnsafe
107−
108− -- *** Appending streams
109− , writeAppendNUnsafe
110− , writeAppendN
111− , writeAppendWith
112− , writeAppend
113−
114− -- ** Eliminating and Reading
115−
116− -- *** To streams
117− , reader
118− , readerRevWith
119− , readerRev
121+ -- *** Indexing
122+ , getIndex
123+ , getIndexUnsafe -- XXX unsafeGetIndex
124+ -- , getFromThenTo
125+ , getIndexRev -- getRevIndex?
126+ , indexReader
127+ , indexReaderWith
120128
121− -- *** To containers
122− , toStreamDWith
123− , toStreamDRevWith
124− , toStreamKWith
125− , toStreamKRevWith
129+ -- *** To Streams
126130 , read
127131 , readRev
132+ , toStreamWith
133+ , toStreamRevWith
128134 , toStreamK
135+ , toStreamKWith
129136 , toStreamKRev
137+ , toStreamKRevWith
138+
139+ -- *** To Containers
130140 , toList
131141
142+ -- *** Unfolds
132143 -- experimental
133144 , producerWith
134145 , producer
135146
136− -- *** Random reads
137− , getIndex
138− , getIndexUnsafe
139− , getIndices
140− , getIndicesD
141− -- , getFromThenTo
142− , getIndexRev
143−
144− -- ** Memory Management
145− , blockSize
146− , arrayChunkBytes
147− , allocBytesToElemCount
148− , realloc
149− , resize
150− , resizeExp
151− , rightSize
147+ , reader
148+ , readerRevWith
149+ , readerRev
152150
153− -- ** Size
151+ -- ** Size and Capacity
152+ -- *** Size
154153 , length
155154 , byteLength
155+
156+ -- *** Capacity
156157 -- , capacity
157158 , byteCapacity
158159 , bytesFree
159160
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+
160176 -- ** In-place Mutation Algorithms
161177 , strip
162178 , reverse
@@ -167,62 +183,121 @@ module Streamly.Internal.Data.MutArray.Type
167183 , mergeBy
168184 , bubble
169185
170− -- ** Casting
171− , cast
172− , castUnsafe
173− , asBytes
174− , asPtrUnsafe
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.
175190
176− -- ** Folding
177− , foldl'
178− , foldr
179− , cmp
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
180218
181− -- ** Arrays of arrays
219+ -- ** Deserialization using Unbox
220+ , peekUncons
221+ , peekUnconsUnsafe -- unsafePeekUncons
222+ , peekSkipUnsafe -- unsafePeekSkip
223+
224+ -- Arrays of arrays
182225 -- We can add dimensionality parameter to the array type to get
183226 -- multidimensional arrays. Multidimensional arrays would just be a
184227 -- convenience wrapper on top of single dimensional arrays.
185228
186− -- | Operations dealing with multiple arrays, streams of arrays or
187− -- multidimensional array representations.
188−
189− -- *** Construct from streams
229+ -- ** Streams of Arrays
230+ -- *** Chunk
231+ -- | Group a stream into arrays.
190232 , chunksOf
191233 , pinnedChunksOf
192− , writeChunks
234+ , buildChunks
193235
194− -- *** Eliminate to streams
195− , flattenArrays
196− , flattenArraysRev
197− , fromArrayStreamK
236+ -- *** Split
237+ -- | Split an array into slices.
198238
199− -- *** Construct from arrays
200− -- get chunks without copying
201− , getSliceUnsafe
202− , getSlice
203239 -- , getSlicesFromLenN
204− , splitAt -- XXX should be able to express using getSlice
205− , breakOn
206−
207− -- ** Cloning arrays
208− , clone
209− , pinnedClone
210−
211− -- ** Appending arrays
212− , spliceCopy
213− , spliceWith
214− , splice
215− , spliceExp
216− , spliceUnsafe
217− -- , putSlice
218− -- , appendSlice
219− -- , 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
220262
221263 -- ** Utilities
222264 , roundUpToPower2
223265 , memcpy
224266 , memcmp
225267 , c_memchr
268+
269+ -- * Deprecated
270+ , asPtrUnsafe
271+ , writeChunks
272+ , flattenArrays
273+ , flattenArraysRev
274+ , fromArrayStreamK
275+ , fromStreamDN
276+ , 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
226301 )
227302where
228303
@@ -233,6 +308,7 @@ where
233308
234309import Control.Monad (when, void)
235310import Control.Monad.IO.Class (MonadIO(..))
311+import Data.Bifunctor (first)
236312import Data.Bits (shiftR, (.|.), (.&.))
237313import Data.Functor.Identity (Identity(..))
238314import Data.Proxy (Proxy(..))
@@ -241,7 +317,6 @@ import Foreign.C.Types (CSize(..), CInt(..))
241317import Foreign.Ptr (plusPtr, minusPtr, nullPtr)
242318import Streamly.Internal.Data.MutByteArray.Type
243319 ( MutByteArray(..)
244− , MutableByteArray
245320 , PinnedState(..)
246321 , getMutableByteArray#
247322 , putSliceUnsafe
@@ -254,19 +329,23 @@ import GHC.Base
254329 , copyMutableByteArray#
255330 )
256331import GHC.Base (noinline)
257−import GHC.Exts (unsafeCoerce#)
332+import GHC.Exts (unsafeCoerce#, Addr#)
258333import GHC.Ptr (Ptr(..))
259334
260335import Streamly.Internal.Data.Fold.Type (Fold(..))
261336import Streamly.Internal.Data.Producer.Type (Producer (..))
262337import Streamly.Internal.Data.Stream.Type (Stream)
338+import Streamly.Internal.Data.Parser.Type (Parser (..))
263339import Streamly.Internal.Data.StreamK.Type (StreamK)
264340import Streamly.Internal.Data.SVar.Type (adaptState, defState)
341+import Streamly.Internal.Data.Tuple.Strict (Tuple'(..))
265342import Streamly.Internal.Data.Unfold.Type (Unfold(..))
266343import Streamly.Internal.System.IO (arrayPayloadSize, defaultChunkSize)
267344
268345import qualified Streamly.Internal.Data.Fold.Type as FL
269346import 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
270349import qualified Streamly.Internal.Data.Producer as Producer
271350import qualified Streamly.Internal.Data.Stream.Type as D
272351import qualified Streamly.Internal.Data.Stream.Lift as D
@@ -274,7 +353,7 @@ import qualified Streamly.Internal.Data.StreamK.Type as K
274353import qualified Prelude
275354
276355import Prelude hiding
277− (Foldable(..), read, unlines, splitAt, reverse, truncate)
356+ (Foldable(..), concat, read, unlines, splitAt, reverse, truncate)
278357
279358#include "DocTestDataMutArray.hs"
280359
@@ -282,6 +361,7 @@ import Prelude hiding
282361-- Foreign helpers
283362-------------------------------------------------------------------------------
284363
364+-- NOTE: Have to be "ccall unsafe" so that we can pass unpinned memory to these
285365foreign import ccall unsafe "string.h memcpy" c_memcpy
286366 :: Ptr Word8 -> Ptr Word8 -> CSize -> IO (Ptr Word8)
287367
@@ -291,6 +371,9 @@ foreign import ccall unsafe "string.h memchr" c_memchr
291371foreign import ccall unsafe "string.h memcmp" c_memcmp
292372 :: Ptr Word8 -> Ptr Word8 -> CSize -> IO CInt
293373
374+foreign import ccall unsafe "string.h strlen" c_strlen
375+ :: Ptr Word8 -> IO CSize
376+
294377-- | Given an 'Unboxed' type (unused first arg) and a number of bytes, return
295378-- how many elements of that type will completely fit in those bytes.
296379--
@@ -396,6 +479,8 @@ isPinned MutArray{..} = Unboxed.isPinned arrContents
396479-- we need more than that. See stg_pinnedNewAlignedByteArrayzh and
397480-- allocatePinned in GHC source.
398481
482+-- XXX Rename to emptyAlignedWith, alignSize should be first arg.
483+
399484-- | @newArrayWith allocator alignment count@ allocates a new array of zero
400485-- length and with a capacity to hold @count@ elements, using @allocator
401486-- size alignment@ as the memory allocator function.
@@ -419,12 +504,25 @@ newArrayWith alloc alignSize count = do
419504 , arrBound = size
420505 }
421506
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." #-}
422520nil ::
423521#ifdef DEVBUILD
424522 Unbox a =>
425523#endif
426524 MutArray a
427−nil = MutArray Unboxed.nil 0 0 0
525+nil = empty
428526
429527{-# INLINE newBytesAs #-}
430528newBytesAs :: MonadIO m =>
@@ -441,12 +539,13 @@ newBytesAs ps bytes = do
441539 , arrBound = bytes
442540 }
443541
444−-- | Allocates a pinned empty array that can hold 'count' items. The memory of
445−-- the array is uninitialized and the allocation is aligned as per the
446−-- '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.
447545--
448546-- /Pre-release/
449547{-# INLINE pinnedNewBytes #-}
548+{-# DEPRECATED pinnedNewBytes "Please use pinnedEmptyOf with appropriate calculation" #-}
450549pinnedNewBytes :: MonadIO m =>
451550#ifdef DEVBUILD
452551 Unbox a =>
@@ -471,21 +570,31 @@ newAs ps =
471570 (error "new: alignment is not used in unpinned arrays.")
472571
473572-- XXX can unaligned allocation be more efficient when alignment is not needed?
474−--
475−-- | Allocates an empty pinned array that can hold 'count' items. The memory of
476−-- the array is uninitialized and the allocation is aligned as per the 'Unboxed'
477−-- instance of the type.
478−--
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." #-}
479582{-# INLINE pinnedNew #-}
480583pinnedNew :: forall m a. (MonadIO m, Unbox a) => Int -> m (MutArray a)
481−pinnedNew = newAs Pinned
584+pinnedNew = pinnedEmptyOf
482585
483−-- | Allocates an empty unpinned array that can hold 'count' items. The memory
484−-- of the array is uninitialized.
586+-- | Allocates an unpinned array of zero length but growable to the specified
587+-- capacity without reallocation.
485588--
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." #-}
486595{-# INLINE new #-}
487596new :: (MonadIO m, Unbox a) => Int -> m (MutArray a)
488−new = newAs Unpinned
597+new = emptyOf
489598
490599-------------------------------------------------------------------------------
491600-- Random writes
@@ -533,7 +642,7 @@ putIndices arr = FL.foldlM' step (return ())
533642
534643 where
535644
536− step () (i, x) = liftIO (putIndex i arr x)
645+ step () (i, x) = putIndex i arr x
537646
538647-- | Modify a given index of an array using a modifier function.
539648--
@@ -729,22 +838,19 @@ arrayChunkBytes = 1024
729838roundDownTo :: Int -> Int -> Int
730839roundDownTo elemSize size = size - (size `mod` elemSize)
731840
732−-- XXX See if resizing can be implemented by reading the old array as a stream
733−-- and then using writeN to the new array.
734−--
735841-- NOTE: we are passing elemSize explicitly to avoid an Unboxed constraint.
736842-- Since this is not inlined Unboxed consrraint leads to dictionary passing
737843-- which complicates some inspection tests.
738844--
739−{-# NOINLINE reallocExplicit #-}
740−reallocExplicit :: Int -> Int -> MutArray a -> IO (MutArray a)
741−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
742848 assertM(arrEnd <= arrBound)
743849
744850 -- Allocate new array
745851 let newCapMaxInBytes = roundUpLargeArray newCapacityInBytes
746852 contents <-
747− if Unboxed.isPinned arrContents
853+ if ps == Pinned
748854 then Unboxed.pinnedNew newCapMaxInBytes
749855 else Unboxed.new newCapMaxInBytes
750856 let !(MutByteArray mbarrFrom#) = arrContents
@@ -769,6 +875,10 @@ reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
769875 , arrBound = newCapInBytes
770876 }
771877
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+
772882-- | @realloc newCapacity array@ reallocates the array to the specified
773883-- capacity in bytes.
774884--
@@ -780,7 +890,12 @@ reallocExplicit elemSize newCapacityInBytes MutArray{..} = do
780890-- If the original array is pinned, the newly allocated array is also pinned.
781891{-# INLINABLE realloc #-}
782892realloc :: forall m a. (MonadIO m, Unbox a) => Int -> MutArray a -> m (MutArray a)
783−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
784899
785900-- | @reallocWith label capSizer minIncrBytes array@. The label is used
786901-- in error messages and the capSizer is used to determine the capacity of the
@@ -803,14 +918,14 @@ reallocWith label capSizer minIncrBytes arr = do
803918 where
804919
805920 badSize newSize =
806− concat
921+ Prelude.concat
807922 [ label
808923 , ": new array size (in bytes) is less than required size "
809924 , show newSize
810925 , ". Please check the sizing function passed."
811926 ]
812927
813−-- | @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
814929-- it is enough to hold the specified number of elements. Nothing is done if
815930-- the specified capacity is less than the length of the array.
816931--
@@ -818,34 +933,46 @@ reallocWith label capSizer minIncrBytes arr = do
818933-- the block size (4K).
819934--
820935-- /Pre-release/
821−{-# INLINE resize #-}
822−resize :: forall m a. (MonadIO m, Unbox a) =>
936+{-# INLINE grow #-}
937+grow :: forall m a. (MonadIO m, Unbox a) =>
823938 Int -> MutArray a -> m (MutArray a)
824−resize nElems arr@MutArray{..} = do
939+grow nElems arr@MutArray{..} = do
825940 let req = SIZE_OF(a) * nElems
826− len = arrEnd - arrStart
827− if req < len
941+ cap = arrBound - arrStart
942+ if req < cap
828943 then return arr
829944 else realloc req arr
830945
831−-- | Like 'resize' but if the byte capacity is more than 'largeObjectThreshold'
832−-- 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.
833954--
834955-- /Pre-release/
835−{-# INLINE resizeExp #-}
836−resizeExp :: forall m a. (MonadIO m, Unbox a) =>
956+{-# INLINE growExp #-}
957+growExp :: forall m a. (MonadIO m, Unbox a) =>
837958 Int -> MutArray a -> m (MutArray a)
838−resizeExp nElems arr@MutArray{..} = do
959+growExp nElems arr@MutArray{..} = do
839960 let req = roundUpLargeArray (SIZE_OF(a) * nElems)
840961 req1 =
841962 if req > largeObjectThreshold
842963 then roundUpToPower2 req
843964 else req
844− len = arrEnd - arrStart
845− if req1 < len
965+ cap = arrBound - arrStart
966+ if req1 < cap
846967 then return arr
847968 else realloc req1 arr
848969
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+
849976-- | Resize the allocated memory to drop any reserved free space at the end of
850977-- the array and reallocate it to reduce wastage.
851978--
@@ -910,7 +1037,7 @@ snocUnsafe arr@MutArray{..} = snocNewEnd (INDEX_NEXT(arrEnd,a)) arr
9101037{-# INLINE snocMay #-}
9111038snocMay :: forall m a. (MonadIO m, Unbox a) =>
9121039 MutArray a -> a -> m (Maybe (MutArray a))
913−snocMay arr@MutArray{..} x = liftIO $ do
1040+snocMay arr@MutArray{..} x = do
9141041 let newEnd = INDEX_NEXT(arrEnd,a)
9151042 if newEnd <= arrBound
9161043 then Just <$> snocNewEnd newEnd arr x
@@ -924,7 +1051,7 @@ snocWithRealloc :: forall m a. (MonadIO m, Unbox a) =>
9241051 -> a
9251052 -> m (MutArray a)
9261053snocWithRealloc sizer arr x = do
927− arr1 <- liftIO $ reallocWith "snocWith" sizer (SIZE_OF(a)) arr
1054+ arr1 <- reallocWith "snocWith" sizer (SIZE_OF(a)) arr
9281055 snocUnsafe arr1 x
9291056
9301057-- | @snocWith sizer arr elem@ mutates @arr@ to append @elem@. The length of
@@ -946,7 +1073,7 @@ snocWith :: forall m a. (MonadIO m, Unbox a) =>
9461073 -> MutArray a
9471074 -> a
9481075 -> m (MutArray a)
949−snocWith allocSize arr x = liftIO $ do
1076+snocWith allocSize arr x = do
9501077 let newEnd = INDEX_NEXT(arrEnd arr,a)
9511078 if newEnd <= arrBound arr
9521079 then snocNewEnd newEnd arr x
@@ -990,6 +1117,126 @@ snoc = snocWith f
9901117 then oldSize * 2
9911118 else roundUpToPower2 oldSize * 2
9921119
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+
9931240-------------------------------------------------------------------------------
9941241-- Random reads
9951242-------------------------------------------------------------------------------
@@ -1032,15 +1279,10 @@ getIndexRev i MutArray{..} = do
10321279data GetIndicesState contents start end st =
10331280 GetIndicesState contents start end st
10341281
1035−-- | Given an unfold that generates array indices, read the elements on those
1036−-- indices from the supplied MutArray. An error is thrown if an index is out of
1037−-- bounds.
1038−--
1039−-- /Pre-release/
1040−{-# INLINE getIndicesD #-}
1041−getIndicesD :: (Monad m, Unbox a) =>
1282+{-# INLINE indexReaderWith #-}
1283+indexReaderWith :: (Monad m, Unbox a) =>
10421284 (forall b. IO b -> m b) -> D.Stream m Int -> Unfold m (MutArray a) a
1043−getIndicesD liftio (D.Stream stepi sti) = Unfold step inject
1285+indexReaderWith liftio (D.Stream stepi sti) = Unfold step inject
10441286
10451287 where
10461288
@@ -1059,9 +1301,26 @@ getIndicesD liftio (D.Stream stepi sti) = Unfold step inject
10591301 D.Skip s -> return $ D.Skip (GetIndicesState contents start end s)
10601302 D.Stop -> return D.Stop
10611303
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." #-}
10621321{-# INLINE getIndices #-}
10631322getIndices :: (MonadIO m, Unbox a) => Stream m Int -> Unfold m (MutArray a) a
1064−getIndices = getIndicesD liftIO
1323+getIndices = indexReader
10651324
10661325-------------------------------------------------------------------------------
10671326-- Subarrays
@@ -1356,7 +1615,7 @@ chunksOfAs ps n (D.Stream step state) =
13561615--
13571616-- Same as the following but may be more efficient:
13581617--
1359−-- >>> chunksOf n = Stream.foldMany (MutArray.writeN n)
1618+-- >>> chunksOf n = Stream.foldMany (MutArray.createOf n)
13601619--
13611620-- /Pre-release/
13621621{-# INLINE_NORMAL chunksOf #-}
@@ -1365,17 +1624,29 @@ chunksOf :: forall m a. (MonadIO m, Unbox a)
13651624-- XXX the idiomatic implementation leads to large regression in the D.reverse'
13661625-- benchmark. It seems it has difficulty producing optimized code when
13671626-- converting to StreamK. Investigate GHC optimizations.
1368−-- chunksOf n = D.foldMany (writeN n)
1627+-- chunksOf n = D.foldMany (createOf n)
13691628chunksOf = chunksOfAs Unpinned
13701629
13711630-- | Like 'chunksOf' but creates pinned arrays.
13721631{-# INLINE_NORMAL pinnedChunksOf #-}
13731632pinnedChunksOf :: forall m a. (MonadIO m, Unbox a)
13741633 => Int -> D.Stream m a -> D.Stream m (MutArray a)
1375−-- pinnedChunksOf n = D.foldMany (pinnedWriteN n)
1634+-- pinnedChunksOf n = D.foldMany (pinnedCreateOf n)
13761635pinnedChunksOf = chunksOfAs Pinned
13771636
1378−-- XXX This should take a PinnedState
1637+-- | When we are buffering a stream of unknown size into an array we do not
1638+-- know how much space to pre-allocate. So we start with the min size and emit
1639+-- the array then keep on doubling the size every time. Thus we do not need to
1640+-- guess the optimum chunk size.
1641+--
1642+-- We can incorporate this in chunksOfAs if the additional size parameter does
1643+-- not impact perf.
1644+--
1645+{-# INLINE _chunksOfRange #-}
1646+_chunksOfRange :: -- (MonadIO m, Unbox a) =>
1647+ PinnedState -> Int -> Int -> D.Stream m a -> D.Stream m (MutArray a)
1648+_chunksOfRange _ps _low _hi = undefined
1649+
13791650-- XXX buffer to a list instead?
13801651-- | Buffer the stream into arrays in memory.
13811652{-# INLINE arrayStreamKFromStreamDAs #-}
@@ -1393,16 +1664,10 @@ data FlattenState s contents a =
13931664 OuterLoop s
13941665 | InnerLoop s contents !Int !Int
13951666
1396−-- | Use the "reader" unfold instead.
1397−--
1398−-- @flattenArrays = unfoldMany reader@
1399−--
1400−-- We can try this if there are any fusion issues in the unfold.
1401−--
1402−{-# INLINE_NORMAL flattenArrays #-}
1403−flattenArrays :: forall m a. (MonadIO m, Unbox a)
1404− => D.Stream m (MutArray a) -> D.Stream m a
1405−flattenArrays (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)
14061671
14071672 where
14081673
@@ -1419,19 +1684,30 @@ flattenArrays (D.Stream step state) = D.Stream step' (OuterLoop state)
14191684 return $ D.Skip $ OuterLoop st
14201685
14211686 step' _ (InnerLoop st contents p end) = do
1422− x <- liftIO $ peekAt p contents
1687+ x <- liftio $ peekAt p contents
14231688 return $ D.Yield x (InnerLoop st contents (INDEX_NEXT(p,a)) end)
14241689
1425−-- | Use the "readerRev" unfold instead.
1690+-- | Use the "reader" unfold instead.
14261691--
1427−-- @flattenArrays = unfoldMany readerRev@
1692+-- @concat = unfoldMany reader@
14281693--
14291694-- We can try this if there are any fusion issues in the unfold.
14301695--
1431−{-# INLINE_NORMAL flattenArraysRev #-}
1432−flattenArraysRev :: forall m a. (MonadIO m, Unbox a)
1696+{-# INLINE_NORMAL concat #-}
1697+concat :: forall m a. (MonadIO m, Unbox a)
1698+ => D.Stream m (MutArray a) -> D.Stream m a
1699+concat = concatWith liftIO
1700+
1701+{-# DEPRECATED flattenArrays "Please use \"unfoldMany reader\" instead." #-}
1702+{-# INLINE flattenArrays #-}
1703+flattenArrays :: forall m a. (MonadIO m, Unbox a)
14331704 => D.Stream m (MutArray a) -> D.Stream m a
1434−flattenArraysRev (D.Stream step state) = D.Stream step' (OuterLoop state)
1705+flattenArrays = concat
1706+
1707+{-# INLINE_NORMAL concatRevWith #-}
1708+concatRevWith :: forall m a. (Monad m, Unbox a)
1709+ => (forall b. IO b -> m b) -> D.Stream m (MutArray a) -> D.Stream m a
1710+concatRevWith liftio (D.Stream step state) = D.Stream step' (OuterLoop state)
14351711
14361712 where
14371713
@@ -1449,10 +1725,27 @@ flattenArraysRev (D.Stream step state) = D.Stream step' (OuterLoop state)
14491725 return $ D.Skip $ OuterLoop st
14501726
14511727 step' _ (InnerLoop st contents p start) = do
1452− x <- liftIO $ peekAt p contents
1728+ x <- liftio $ peekAt p contents
14531729 let cur = INDEX_PREV(p,a)
14541730 return $ D.Yield x (InnerLoop st contents cur start)
14551731
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+
1743+{-# DEPRECATED flattenArraysRev "Please use \"unfoldMany readerRev\" instead." #-}
1744+{-# INLINE flattenArraysRev #-}
1745+flattenArraysRev :: forall m a. (MonadIO m, Unbox a)
1746+ => D.Stream m (MutArray a) -> D.Stream m a
1747+flattenArraysRev = concatRev
1748+
14561749-------------------------------------------------------------------------------
14571750-- Unfolds
14581751-------------------------------------------------------------------------------
@@ -1570,11 +1863,11 @@ toList MutArray{..} = liftIO $ go arrStart
15701863 x <- peekAt p arrContents
15711864 (:) x <$> go (INDEX_NEXT(p,a))
15721865
1573−{-# INLINE_NORMAL toStreamDWith #-}
1574−toStreamDWith ::
1866+{-# INLINE_NORMAL toStreamWith #-}
1867+toStreamWith ::
15751868 forall m a. (Monad m, Unbox a)
15761869 => (forall b. IO b -> m b) -> MutArray a -> D.Stream m a
1577−toStreamDWith liftio MutArray{..} = D.Stream step arrStart
1870+toStreamWith liftio MutArray{..} = D.Stream step arrStart
15781871
15791872 where
15801873
@@ -1590,7 +1883,7 @@ toStreamDWith liftio MutArray{..} = D.Stream step arrStart
15901883--
15911884{-# INLINE_NORMAL read #-}
15921885read :: forall m a. (MonadIO m, Unbox a) => MutArray a -> D.Stream m a
1593−read = toStreamDWith liftIO
1886+read = toStreamWith liftIO
15941887
15951888{-# INLINE toStreamKWith #-}
15961889toStreamKWith ::
@@ -1609,11 +1902,11 @@ toStreamKWith liftio MutArray{..} = go arrStart
16091902toStreamK :: forall m a. (MonadIO m, Unbox a) => MutArray a -> StreamK m a
16101903toStreamK = toStreamKWith liftIO
16111904
1612−{-# INLINE_NORMAL toStreamDRevWith #-}
1613−toStreamDRevWith ::
1905+{-# INLINE_NORMAL toStreamRevWith #-}
1906+toStreamRevWith ::
16141907 forall m a. (Monad m, Unbox a)
16151908 => (forall b. IO b -> m b) -> MutArray a -> D.Stream m a
1616−toStreamDRevWith liftio MutArray{..} =
1909+toStreamRevWith liftio MutArray{..} =
16171910 let p = INDEX_PREV(arrEnd,a)
16181911 in D.Stream step p
16191912
@@ -1631,7 +1924,7 @@ toStreamDRevWith liftio MutArray{..} =
16311924--
16321925{-# INLINE_NORMAL readRev #-}
16331926readRev :: forall m a. (MonadIO m, Unbox a) => MutArray a -> D.Stream m a
1634−readRev = toStreamDRevWith liftIO
1927+readRev = toStreamRevWith liftIO
16351928
16361929{-# INLINE toStreamKRevWith #-}
16371930toStreamKRevWith ::
@@ -1680,7 +1973,7 @@ foldr f z arr = D.foldr f z $ read arr
16801973
16811974-- XXX Keep the bound intact to not lose any free space? Perf impact?
16821975
1683−-- | @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
16841977-- array.
16851978--
16861979-- Unsafe: Do not drive the fold beyond @n@ elements, it will lead to memory
@@ -1689,13 +1982,12 @@ foldr f z arr = D.foldr f z $ read arr
16891982-- Any free space left in the array after appending @n@ elements is lost.
16901983--
16911984-- /Internal/
1692−{-# INLINE_NORMAL writeAppendNUnsafe #-}
1693−writeAppendNUnsafe :: forall m a. (MonadIO m, Unbox a) =>
1985+{-# INLINE_NORMAL unsafeAppendN #-}
1986+unsafeAppendN :: forall m a. (MonadIO m, Unbox a) =>
16941987 Int
16951988 -> m (MutArray a)
16961989 -> Fold m a (MutArray a)
1697−writeAppendNUnsafe n action =
1698− fmap fromArrayUnsafe $ FL.foldlM' step initial
1990+unsafeAppendN n action = fmap fromArrayUnsafe $ FL.foldlM' step initial
16991991
17001992 where
17011993
@@ -1708,7 +2000,7 @@ writeAppendNUnsafe n action =
17082000 -- otherwise we lose that space.
17092001 arr1 <-
17102002 if free < needed
1711− then noinline reallocWith "writeAppendNUnsafeWith" (+ needed) needed arr
2003+ then noinline reallocWith "unsafeAppendN" (+ needed) needed arr
17122004 else return arr
17132005 return $ toArrayUnsafe arr1
17142006
@@ -1716,30 +2008,51 @@ writeAppendNUnsafe n action =
17162008 liftIO $ pokeAt end contents x
17172009 return $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
17182010
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+
17192019-- | Append @n@ elements to an existing array. Any free space left in the array
17202020-- after appending @n@ elements is lost.
17212021--
1722−-- >>> writeAppendN n initial = Fold.take n (MutArray.writeAppendNUnsafe n initial)
2022+-- >>> appendN n initial = Fold.take n (MutArray.unsafeAppendN n initial)
17232023--
1724−{-# 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 #-}
17252032writeAppendN :: forall m a. (MonadIO m, Unbox a) =>
17262033 Int -> m (MutArray a) -> Fold m a (MutArray a)
1727−writeAppendN n initial = FL.take n (writeAppendNUnsafe n initial)
2034+writeAppendN = appendN
17282035
1729−-- | @writeAppendWith realloc action@ mutates the array generated by @action@ to
2036+-- | @appendWith realloc action@ mutates the array generated by @action@ to
17302037-- append the input stream. If there is no reserved space available in the
17312038-- array it is reallocated to a size in bytes determined by @realloc oldSize@,
17322039-- where @oldSize@ is the current size of the array in bytes.
17332040--
17342041-- Note that the returned array may be a mutated version of original array.
17352042--
1736−-- >>> writeAppendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
2043+-- >>> appendWith sizer = Fold.foldlM' (MutArray.snocWith sizer)
17372044--
17382045-- /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." #-}
17392052{-# INLINE writeAppendWith #-}
17402053writeAppendWith :: forall m a. (MonadIO m, Unbox a) =>
17412054 (Int -> Int) -> m (MutArray a) -> Fold m a (MutArray a)
1742−writeAppendWith sizer = FL.foldlM' (snocWith sizer)
2055+writeAppendWith = appendWith
17432056
17442057-- | @append action@ mutates the array generated by @action@ to append the
17452058-- input stream. If there is no reserved space available in the array it is
@@ -1747,26 +2060,33 @@ writeAppendWith sizer = FL.foldlM' (snocWith sizer)
17472060--
17482061-- Note that the returned array may be a mutated version of original array.
17492062--
1750−-- >>> writeAppend = MutArray.writeAppendWith (* 2)
2063+-- >>> append = MutArray.appendWith (* 2)
17512064--
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." #-}
17522072{-# INLINE writeAppend #-}
17532073writeAppend :: forall m a. (MonadIO m, Unbox a) =>
17542074 m (MutArray a) -> Fold m a (MutArray a)
1755−writeAppend = writeAppendWith (* 2)
2075+writeAppend = append
17562076
17572077-- XXX We can carry bound as well in the state to make sure we do not lose the
17582078-- remaining capacity. Need to check perf impact.
17592079--
1760−-- | Like 'writeNUnsafe' but takes a new array allocator @alloc size@ function
1761−-- as argument.
2080+-- | Like 'unsafeCreateOf' but takes a new array allocator @alloc size@
2081+-- function as argument.
17622082--
1763−-- >>> writeNWithUnsafe alloc n = MutArray.writeAppendNUnsafe (alloc n) n
2083+-- >>> unsafeCreateOfWith alloc n = MutArray.unsafeAppendN (alloc n) n
17642084--
17652085-- /Pre-release/
1766−{-# INLINE_NORMAL writeNWithUnsafe #-}
1767−writeNWithUnsafe :: forall m a. (MonadIO m, Unbox a)
2086+{-# INLINE_NORMAL unsafeCreateOfWith #-}
2087+unsafeCreateOfWith :: forall m a. (MonadIO m, Unbox a)
17682088 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
1769−writeNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
2089+unsafeCreateOfWith alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
17702090
17712091 where
17722092
@@ -1777,40 +2097,64 @@ writeNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
17772097 return
17782098 $ ArrayUnsafe contents start (INDEX_NEXT(end,a))
17792099
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+
17802106{-# INLINE_NORMAL writeNUnsafeAs #-}
17812107writeNUnsafeAs :: forall m a. (MonadIO m, Unbox a)
17822108 => PinnedState -> Int -> Fold m a (MutArray a)
1783−writeNUnsafeAs ps = writeNWithUnsafe (newAs ps)
2109+writeNUnsafeAs ps = unsafeCreateOfWith (newAs ps)
17842110
1785−-- | 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
17862112-- driver must not call the step function more than 'n' times otherwise it will
17872113-- corrupt the memory and crash. This function exists mainly because any
17882114-- conditional in the step function blocks fusion causing 10x performance
17892115-- slowdown.
17902116--
1791−-- >>> writeNUnsafe = MutArray.writeNWithUnsafe MutArray.new
2117+-- >>> unsafeCreateOf = MutArray.unsafeCreateOfWith MutArray.emptyOf
17922118--
1793−{-# 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 #-}
17942126writeNUnsafe :: forall m a. (MonadIO m, Unbox a)
17952127 => Int -> Fold m a (MutArray a)
1796−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
17972135
1798−-- | Like 'writeNUnsafe' but creates a pinned array.
1799−{-# INLINE_NORMAL pinnedWriteNUnsafe #-}
2136+{-# DEPRECATED pinnedWriteNUnsafe "Please use unsafePinnedCreateOf instead." #-}
2137+{-# INLINE pinnedWriteNUnsafe #-}
18002138pinnedWriteNUnsafe :: forall m a. (MonadIO m, Unbox a)
18012139 => Int -> Fold m a (MutArray a)
1802−pinnedWriteNUnsafe = writeNUnsafeAs Pinned
2140+pinnedWriteNUnsafe = unsafePinnedCreateOf
18032141
1804−-- | @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
18052143-- allocated using the @alloc@ function.
18062144--
1807−-- >>> writeNWith alloc n = Fold.take n (MutArray.writeNWithUnsafe alloc n)
1808−-- >>> 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
18092147--
1810−{-# 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 #-}
18112155writeNWith :: forall m a. (MonadIO m, Unbox a)
18122156 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
1813−writeNWith alloc n = FL.take n (writeNWithUnsafe alloc n)
2157+writeNWith = createOfWith
18142158
18152159{-# INLINE_NORMAL writeNAs #-}
18162160writeNAs ::
@@ -1818,28 +2162,42 @@ writeNAs ::
18182162 => PinnedState
18192163 -> Int
18202164 -> Fold m a (MutArray a)
1821−writeNAs ps = writeNWith (newAs ps)
2165+writeNAs ps = createOfWith (newAs ps)
18222166
1823−-- | @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
18242168-- 'MutArray'.
18252169--
1826−-- >>> writeN = MutArray.writeNWith MutArray.new
1827−-- >>> writeN n = Fold.take n (MutArray.writeNUnsafe n)
1828−-- >>> 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)
18292173--
1830−{-# 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 #-}
18312181writeN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
1832−writeN = writeNAs Unpinned
2182+writeN = createOf
2183+
2184+-- | Like 'createOf' but creates a pinned array.
2185+{-# INLINE_NORMAL pinnedCreateOf #-}
2186+pinnedCreateOf ::
2187+ forall m a. (MonadIO m, Unbox a)
2188+ => Int
2189+ -> Fold m a (MutArray a)
2190+pinnedCreateOf = writeNAs Pinned
18332191
1834−-- | Like 'writeN' but creates a pinned array.
1835−{-# INLINE_NORMAL pinnedWriteN #-}
2192+{-# DEPRECATED pinnedWriteN "Please use pinnedCreateOf instead." #-}
2193+{-# INLINE pinnedWriteN #-}
18362194pinnedWriteN ::
18372195 forall m a. (MonadIO m, Unbox a)
18382196 => Int
18392197 -> Fold m a (MutArray a)
1840−pinnedWriteN = writeNAs Pinned
2198+pinnedWriteN = pinnedCreateOf
18412199
1842−-- | Like writeNWithUnsafe but writes the array in reverse order.
2200+-- | Like unsafeCreateOfWith but writes the array in reverse order.
18432201--
18442202-- /Internal/
18452203{-# INLINE_NORMAL writeRevNWithUnsafe #-}
@@ -1860,7 +2218,7 @@ writeRevNWithUnsafe alloc n = fromArrayUnsafe <$> FL.foldlM' step initial
18602218 return
18612219 $ ArrayUnsafe contents ptr end
18622220
1863−-- | Like writeNWith but writes the array in reverse order.
2221+-- | Like createOfWith but writes the array in reverse order.
18642222--
18652223-- /Internal/
18662224{-# INLINE_NORMAL writeRevNWith #-}
@@ -1868,31 +2226,36 @@ writeRevNWith :: forall m a. (MonadIO m, Unbox a)
18682226 => (Int -> m (MutArray a)) -> Int -> Fold m a (MutArray a)
18692227writeRevNWith alloc n = FL.take n (writeRevNWithUnsafe alloc n)
18702228
1871−-- | Like writeN but writes the array in reverse order.
2229+-- | Like 'createOf' but writes the array in reverse order.
18722230--
18732231-- /Pre-release/
1874−{-# 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 #-}
18752238writeRevN :: forall m a. (MonadIO m, Unbox a) => Int -> Fold m a (MutArray a)
1876−writeRevN = writeRevNWith new
2239+writeRevN = revCreateOf
18772240
18782241-- | @pinnedWriteNAligned align n@ folds a maximum of @n@ elements from the
18792242-- input stream to a 'MutArray' aligned to the given size.
18802243--
1881−-- >>> pinnedWriteNAligned align = MutArray.writeNWith (MutArray.pinnedNewAligned align)
1882−-- >>> 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)
18832246--
18842247-- /Pre-release/
18852248--
18862249{-# INLINE_NORMAL pinnedWriteNAligned #-}
18872250pinnedWriteNAligned :: forall m a. (MonadIO m, Unbox a)
18882251 => Int -> Int -> Fold m a (MutArray a)
1889−pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
2252+pinnedWriteNAligned align = createOfWith (pinnedNewAligned align)
18902253
18912254-- XXX Buffer to a list instead?
18922255
18932256-- | Buffer a stream into a stream of arrays.
18942257--
1895−-- >>> writeChunks n = Fold.many (MutArray.writeN n) Fold.toStreamK
2258+-- >>> buildChunks n = Fold.many (MutArray.createOf n) Fold.toStreamK
18962259--
18972260-- Breaking an array into an array stream can be useful to consume a large
18982261-- array sequentially such that memory of the array is released incrementatlly.
@@ -1901,37 +2264,43 @@ pinnedWriteNAligned align = writeNWith (pinnedNewAligned align)
19012264--
19022265-- /Unimplemented/
19032266--
1904−{-# INLINE_NORMAL writeChunks #-}
2267+{-# INLINE_NORMAL buildChunks #-}
2268+buildChunks :: (MonadIO m, Unbox a) =>
2269+ Int -> Fold m a (StreamK n (MutArray a))
2270+buildChunks n = FL.many (createOf n) FL.toStreamK
2271+
2272+{-# DEPRECATED writeChunks "Please use buildChunks instead." #-}
2273+{-# INLINE writeChunks #-}
19052274writeChunks :: (MonadIO m, Unbox a) =>
19062275 Int -> Fold m a (StreamK n (MutArray a))
1907−writeChunks n = FL.many (writeN n) FL.toStreamK
2276+writeChunks = buildChunks
19082277
19092278{-# INLINE_NORMAL writeWithAs #-}
19102279writeWithAs :: forall m a. (MonadIO m, Unbox a)
19112280 => PinnedState -> Int -> Fold m a (MutArray a)
1912−-- writeWithAs ps n = FL.rmapM rightSize $ writeAppendWith (* 2) (newAs ps n)
2281+-- writeWithAs ps n = FL.rmapM rightSize $ appendWith (* 2) (newAs ps n)
19132282writeWithAs ps elemCount =
19142283 FL.rmapM extract $ FL.foldlM' step initial
19152284
19162285 where
19172286
19182287 initial = do
1919− when (elemCount < 0) $ error "writeWith: elemCount is negative"
1920− liftIO $ newAs ps elemCount
2288+ when (elemCount < 0) $ error "createWith: elemCount is negative"
2289+ newAs ps elemCount
19212290
19222291 step arr@(MutArray _ start end bound) x
19232292 | INDEX_NEXT(end,a) > bound = do
19242293 let oldSize = end - start
19252294 newSize = max (oldSize * 2) 1
1926− arr1 <- liftIO $ reallocExplicit (SIZE_OF(a)) newSize arr
2295+ arr1 <- liftIO $ reallocExplicitAs ps (SIZE_OF(a)) newSize arr
19272296 snocUnsafe arr1 x
19282297 step arr x = snocUnsafe arr x
19292298
19302299 extract = liftIO . rightSize
19312300
1932−-- XXX Compare writeWith with fromStreamD which uses an array of streams
1933−-- implementation. We can write this using writeChunks above if that is faster.
1934−-- If writeWith is faster then we should use that to implement
2301+-- XXX Compare createWith with fromStreamD which uses an array of streams
2302+-- implementation. We can write this using buildChunks above if that is faster.
2303+-- If createWith is faster then we should use that to implement
19352304-- fromStreamD.
19362305--
19372306-- XXX The realloc based implementation needs to make one extra copy if we use
@@ -1946,39 +2315,56 @@ writeWithAs ps elemCount =
19462315-- XXX check if GHC's memory allocator is efficient enough. We can try the C
19472316-- malloc to compare against.
19482317
1949−-- | @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
19502319-- starts at a size big enough to hold minCount elements, the size is doubled
19512320-- every time the array needs to be grown.
19522321--
19532322-- /Caution! Do not use this on infinite streams./
19542323--
1955−-- >>> f n = MutArray.writeAppendWith (* 2) (MutArray.new n)
1956−-- >>> writeWith n = Fold.rmapM MutArray.rightSize (f n)
1957−-- >>> writeWith n = Fold.rmapM MutArray.fromArrayStreamK (MutArray.writeChunks 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)
19582327--
19592328-- /Pre-release/
1960−{-# 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 #-}
19612337writeWith :: forall m a. (MonadIO m, Unbox a)
19622338 => Int -> Fold m a (MutArray a)
1963−-- writeWith n = FL.rmapM rightSize $ writeAppendWith (* 2) (new n)
1964−writeWith = writeWithAs Unpinned
2339+writeWith = createWith
19652340
19662341-- | Fold the whole input to a single array.
19672342--
1968−-- Same as 'writeWith' using an initial array size of 'arrayChunkBytes' bytes
2343+-- Same as 'createWith' using an initial array size of 'arrayChunkBytes' bytes
19692344-- rounded up to the element size.
19702345--
19712346-- /Caution! Do not use this on infinite streams./
19722347--
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." #-}
19732354{-# INLINE write #-}
19742355write :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
1975−write = writeWith (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2356+write = create
2357+
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)
19762363
1977−-- | Like 'write' but creates a pinned array.
2364+{-# DEPRECATED pinnedWrite "Please use pinnedCreate instead." #-}
19782365{-# INLINE pinnedWrite #-}
19792366pinnedWrite :: forall m a. (MonadIO m, Unbox a) => Fold m a (MutArray a)
1980−pinnedWrite =
1981− writeWithAs Pinned (allocBytesToElemCount (undefined :: a) arrayChunkBytes)
2367+pinnedWrite = pinnedCreate
19822368
19832369-------------------------------------------------------------------------------
19842370-- construct from streams, known size
@@ -1988,7 +2374,7 @@ pinnedWrite =
19882374fromStreamDNAs :: forall m a. (MonadIO m, Unbox a)
19892375 => PinnedState -> Int -> D.Stream m a -> m (MutArray a)
19902376fromStreamDNAs ps limit str = do
1991− (arr :: MutArray a) <- liftIO $ newAs ps limit
2377+ (arr :: MutArray a) <- newAs ps limit
19922378 end <- D.foldlM' (fwrite (arrContents arr)) (return $ arrEnd arr) $ D.take limit str
19932379 return $ arr {arrEnd = end}
19942380
@@ -1998,15 +2384,21 @@ fromStreamDNAs ps limit str = do
19982384 liftIO $ pokeAt ptr arrContents x
19992385 return $ INDEX_NEXT(ptr,a)
20002386
2001−-- | Use the 'writeN' fold instead.
2387+-- | Use the 'createOf' fold instead.
20022388--
2003−-- >>> fromStreamDN n = Stream.fold (MutArray.writeN n)
2389+-- >>> fromStreamN n = Stream.fold (MutArray.createOf n)
20042390--
2005−{-# INLINE_NORMAL fromStreamDN #-}
2391+{-# INLINE_NORMAL fromStreamN #-}
2392+fromStreamN :: forall m a. (MonadIO m, Unbox a)
2393+ => Int -> D.Stream m a -> m (MutArray a)
2394+-- fromStreamDN n = D.fold (createOf n)
2395+fromStreamN = fromStreamDNAs Unpinned
2396+
2397+{-# DEPRECATED fromStreamDN "Please use fromStreamN instead." #-}
2398+{-# INLINE fromStreamDN #-}
20062399fromStreamDN :: forall m a. (MonadIO m, Unbox a)
20072400 => Int -> D.Stream m a -> m (MutArray a)
2008−-- fromStreamDN n = D.fold (writeN n)
2009−fromStreamDN = fromStreamDNAs Unpinned
2401+fromStreamDN = fromStreamN
20102402
20112403-- | Create a 'MutArray' from the first N elements of a list. The array is
20122404-- allocated to size N, if the list terminates before N elements then the
@@ -2026,16 +2418,70 @@ pinnedFromListN n xs = fromStreamDNAs Pinned n $ D.fromList xs
20262418-- /Pre-release/
20272419{-# INLINE fromListRevN #-}
20282420fromListRevN :: (MonadIO m, Unbox a) => Int -> [a] -> m (MutArray a)
2029−fromListRevN n xs = D.fold (writeRevN n) $ D.fromList xs
2421+fromListRevN n xs = D.fold (revCreateOf n) $ D.fromList xs
2422+
2423+-- | Convert a pure stream in Identity monad to a mutable array.
2424+{-# INLINABLE fromPureStreamN #-}
2425+fromPureStreamN :: (MonadIO m, Unbox a) =>
2426+ Int -> Stream Identity a -> m (MutArray a)
2427+fromPureStreamN n xs =
2428+ D.fold (createOf n) $ D.morphInner (return . runIdentity) xs
20302429
20312430-- | Convert a pure stream in Identity monad to a mutable array.
20322431{-# INLINABLE fromPureStream #-}
20332432fromPureStream :: (MonadIO m, Unbox a) => Stream Identity a -> m (MutArray a)
20342433fromPureStream xs =
2035− liftIO $ D.fold write $ D.morphInner (return . runIdentity) xs
2434+ D.fold create $ D.morphInner (return . runIdentity) xs
2435+
2436+{-# INLINABLE fromPtrN #-}
2437+fromPtrN :: MonadIO m => Int -> Ptr Word8 -> m (MutArray Word8)
2438+fromPtrN len addr = do
2439+ -- memcpy is better than stream copy when the size is known.
2440+ -- XXX We can implement a stream copy in a similar way by streaming Word64
2441+ -- first and then remaining Word8.
2442+ arr <- new len
2443+ _ <- unsafeAsPtr arr
2444+ (\ptr -> liftIO $ c_memcpy ptr addr (fromIntegral len))
2445+ return (arr {arrEnd = len})
2446+
2447+{-# INLINABLE fromByteStr# #-}
2448+fromByteStr# :: MonadIO m => Addr# -> m (MutArray Word8)
2449+fromByteStr# addr = do
2450+ -- It is better to count the size first and allocate exact space.
2451+ -- Also, memcpy is better than stream copy when the size is known.
2452+ -- C strlen compares 4 bytes at a time, so is better than the stream
2453+ -- version. https://github.com/bminor/glibc/blob/master/string/strlen.c
2454+ -- XXX We can possibly use a stream of Word64 to do the same.
2455+ -- fromByteStr# addr = fromPureStream (D.fromByteStr# addr)
2456+ len <- liftIO $ c_strlen (Ptr addr)
2457+ let lenInt = fromIntegral len
2458+ arr <- new lenInt
2459+ _ <- unsafeAsPtr arr (\ptr -> liftIO $ c_memcpy ptr (Ptr addr) len)
2460+ return (arr {arrEnd = lenInt})
20362461
20372462-------------------------------------------------------------------------------
2038−-- convert stream to a single array
2463+-- convert a stream of arrays to a single array by reallocating and copying
2464+-------------------------------------------------------------------------------
2465+
2466+-- XXX Both of these implementations of splicing seem to perform equally well.
2467+-- We need to perform benchmarks over a range of sizes though.
2468+
2469+-- | Also see 'fromChunksK'.
2470+{-# INLINE fromChunksRealloced #-}
2471+fromChunksRealloced :: forall m a. (MonadIO m, Unbox a)
2472+ => Stream m (MutArray a) -> m (MutArray a)
2473+fromChunksRealloced s = do
2474+ res <- D.uncons s
2475+ case res of
2476+ Just (a, strm) -> do
2477+ arr <- D.foldlM' spliceExp (pure a) strm
2478+ -- Reallocation is exponential so there may be 50% empty space in
2479+ -- worst case. One more reallocation to reclaim the space.
2480+ rightSize arr
2481+ Nothing -> pure nil
2482+
2483+-------------------------------------------------------------------------------
2484+-- convert a stream of arrays to a single array by buffering arrays first
20392485-------------------------------------------------------------------------------
20402486
20412487{-# INLINE arrayStreamKLength #-}
@@ -2045,34 +2491,72 @@ arrayStreamKLength as = K.foldl' (+) 0 (K.map length as)
20452491-- | Convert an array stream to an array. Note that this requires peak memory
20462492-- that is double the size of the array stream.
20472493--
2494+{-# INLINE fromChunkskAs #-}
2495+fromChunkskAs :: (Unbox a, MonadIO m) =>
2496+ PinnedState -> StreamK m (MutArray a) -> m (MutArray a)
2497+fromChunkskAs ps as = do
2498+ len <- arrayStreamKLength as
2499+ arr <- newAs ps len
2500+ -- XXX is StreamK fold faster or StreamD fold?
2501+ K.foldlM' spliceUnsafe (pure arr) as
2502+ -- fromStreamDN len $ D.unfoldMany reader $ D.fromStreamK as
2503+
2504+-- XXX Need to compare this with fromChunks and fromChunkList and keep the
2505+-- fastest or simplest one if all are equally fast.
2506+
2507+-- | Convert an array stream to an array. Note that this requires peak memory
2508+-- that is double the size of the array stream.
2509+--
2510+-- Also see 'fromChunksRealloced'.
2511+--
2512+{-# INLINE fromChunksK #-}
2513+fromChunksK :: (Unbox a, MonadIO m) =>
2514+ StreamK m (MutArray a) -> m (MutArray a)
2515+fromChunksK = fromChunkskAs Unpinned
2516+
2517+{-# DEPRECATED fromArrayStreamK "Please use fromChunksK instead." #-}
20482518{-# INLINE fromArrayStreamK #-}
20492519fromArrayStreamK :: (Unbox a, MonadIO m) =>
20502520 StreamK m (MutArray a) -> m (MutArray a)
2051−fromArrayStreamK as = do
2052− len <- arrayStreamKLength as
2053− fromStreamDN len $ D.unfoldMany reader $ D.fromStreamK as
2521+fromArrayStreamK = fromChunksK
20542522
20552523{-# INLINE fromStreamDAs #-}
20562524fromStreamDAs ::
20572525 (MonadIO m, Unbox a) => PinnedState -> D.Stream m a -> m (MutArray a)
2058−fromStreamDAs ps m = arrayStreamKFromStreamDAs ps m >>= fromArrayStreamK
2526+fromStreamDAs ps m =
2527+ arrayStreamKFromStreamDAs Unpinned m >>= fromChunkskAs ps
2528+
2529+-- | Create an 'Array' from a stream. This is useful when we want to create a
2530+-- single array from a stream of unknown size. 'createOf' is at least twice
2531+-- as efficient when the size is already known.
2532+--
2533+-- Note that if the input stream is too large memory allocation for the array
2534+-- may fail. When the stream size is not known, `chunksOf` followed by
2535+-- processing of indvidual arrays in the resulting stream should be preferred.
2536+--
2537+-- /Pre-release/
2538+{-# INLINE fromStream #-}
2539+fromStream :: (MonadIO m, Unbox a) => Stream m a -> m (MutArray a)
2540+fromStream = fromStreamDAs Unpinned
20592541
2542+-- fromStream (Stream m) = P.fold create m
20602543-- CAUTION: a very large number (millions) of arrays can degrade performance
20612544-- due to GC overhead because we need to buffer the arrays before we flatten
20622545-- all the arrays.
20632546--
2064−-- XXX Compare if this is faster or "fold write".
2547+-- XXX Compare if this is faster or "fold create".
20652548--
20662549-- | We could take the approach of doubling the memory allocation on each
20672550-- overflow. This would result in more or less the same amount of copying as in
20682551-- the chunking approach. However, if we have to shrink in the end then it may
20692552-- result in an extra copy of the entire data.
20702553--
2071−-- >>> fromStreamD = StreamD.fold MutArray.write
2554+-- >>> fromStreamD = StreamD.fold MutArray.create
20722555--
20732556{-# INLINE fromStreamD #-}
2557+{-# DEPRECATED fromStreamD "Please use fromStream instead." #-}
20742558fromStreamD :: (MonadIO m, Unbox a) => D.Stream m a -> m (MutArray a)
2075−fromStreamD = fromStreamDAs Unpinned
2559+fromStreamD = fromStream
20762560
20772561-- | Create a 'MutArray' from a list. The list must be of finite size.
20782562--
@@ -2097,6 +2581,10 @@ fromListRev xs = fromListRevN (Prelude.length xs) xs
20972581-- Cloning
20982582-------------------------------------------------------------------------------
20992583
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+
21002588{-# INLINE cloneAs #-}
21012589cloneAs ::
21022590 ( MonadIO m
@@ -2106,7 +2594,7 @@ cloneAs ::
21062594 )
21072595 => PinnedState -> MutArray a -> m (MutArray a)
21082596cloneAs ps src =
2109− liftIO $ do
2597+ do
21102598 let startSrc = arrStart src
21112599 srcLen = arrEnd src - startSrc
21122600 newArrContents <-
@@ -2145,17 +2633,16 @@ spliceCopy :: forall m a. MonadIO m =>
21452633 Unbox a =>
21462634#endif
21472635 MutArray a -> MutArray a -> m (MutArray a)
2148−spliceCopy arr1 arr2 = liftIO $ do
2636+spliceCopy arr1 arr2 = do
21492637 let start1 = arrStart arr1
21502638 start2 = arrStart arr2
21512639 len1 = arrEnd arr1 - start1
21522640 len2 = arrEnd arr2 - start2
2153− let newLen = len1 + len2
2641+ let len = len1 + len2
21542642 newArrContents <-
21552643 if Unboxed.isPinned (arrContents arr1)
2156− then Unboxed.pinnedNew newLen
2157− else Unboxed.new newLen
2158− let len = len1 + len2
2644+ then liftIO $ Unboxed.pinnedNew len
2645+ else liftIO $ Unboxed.new len
21592646 putSliceUnsafe (arrContents arr1) start1 newArrContents 0 len1
21602647 putSliceUnsafe (arrContents arr2) start2 newArrContents len1 len2
21612648 return $ MutArray newArrContents 0 len len
@@ -2167,7 +2654,7 @@ spliceCopy arr1 arr2 = liftIO $ do
21672654spliceUnsafe :: MonadIO m =>
21682655 MutArray a -> MutArray a -> m (MutArray a)
21692656spliceUnsafe dst src =
2170− liftIO $ do
2657+ do
21712658 let startSrc = arrStart src
21722659 srcLen = arrEnd src - startSrc
21732660 endDst = arrEnd dst
@@ -2189,7 +2676,7 @@ spliceWith :: forall m a. (MonadIO m, Unbox a) =>
21892676 (Int -> Int -> Int) -> MutArray a -> MutArray a -> m (MutArray a)
21902677spliceWith sizer dst@(MutArray _ start end bound) src = do
21912678{-
2192− let f = writeAppendWith (`sizer` byteLength src) (return dst)
2679+ let f = appendWith (`sizer` byteLength src) (return dst)
21932680 in D.fold f (toStreamD src)
21942681-}
21952682 assert (end <= bound) (return ())
@@ -2205,7 +2692,7 @@ spliceWith sizer dst@(MutArray _ start end bound) src = do
22052692 $ "splice: newSize is less than the total size "
22062693 ++ "of arrays being appended. Please check the "
22072694 ++ "sizer function passed."
2208− liftIO $ realloc newSizeInBytes dst
2695+ realloc newSizeInBytes dst
22092696 else return dst
22102697 spliceUnsafe dst1 src
22112698
@@ -2242,14 +2729,24 @@ spliceExp = spliceWith (\l1 l2 -> max (l1 * 2) (l1 + l2))
22422729-- Splitting
22432730-------------------------------------------------------------------------------
22442731
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+
22452743-- | Drops the separator byte
22462744{-# INLINE breakOn #-}
22472745breakOn :: MonadIO m
22482746 => Word8 -> MutArray Word8 -> m (MutArray Word8, Maybe (MutArray Word8))
2249−breakOn sep arr@MutArray{..} = asPtrUnsafe arr $ \p -> liftIO $ do
2250− -- XXX Instead of using asPtrUnsafe (pinning memory) we can pass unlifted
2251− -- Addr# to memchr and it should be safe (from ghc 8.4).
2747+breakOn sep arr@MutArray{..} = unsafeAsPtr arr $ \p -> liftIO $ do
22522748 -- XXX We do not need memchr here, we can use a Haskell equivalent.
2749+ -- Need efficient stream based primitives that work on Word64.
22532750 loc <- c_memchr p sep (fromIntegral $ byteLength arr)
22542751 let sepIndex = loc `minusPtr` p
22552752 return $
@@ -2270,32 +2767,41 @@ breakOn sep arr@MutArray{..} = asPtrUnsafe arr $ \p -> liftIO $ do
22702767 }
22712768 )
22722769
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+
22732792-- | Create two slices of an array without copying the original array. The
22742793-- specified index @i@ is the first index of the second slice.
22752794--
2795+{-# INLINE splitAt #-}
22762796splitAt :: forall a. Unbox a => Int -> MutArray a -> (MutArray a, MutArray a)
2277−splitAt i arr@MutArray{..} =
2797+splitAt i arr =
22782798 let maxIndex = length arr - 1
22792799 in if i < 0
22802800 then error "sliceAt: negative array index"
22812801 else if i > maxIndex
22822802 then error $ "sliceAt: specified array index " ++ show i
22832803 ++ " is beyond the maximum index " ++ show maxIndex
2284− else let off = i * SIZE_OF(a)
2285− p = arrStart + off
2286− in ( MutArray
2287− { arrContents = arrContents
2288− , arrStart = arrStart
2289− , arrEnd = p
2290− , arrBound = p
2291− }
2292− , MutArray
2293− { arrContents = arrContents
2294− , arrStart = p
2295− , arrEnd = arrEnd
2296− , arrBound = arrBound
2297− }
2298− )
2804+ else unsafeSplitAt i arr
22992805
23002806-------------------------------------------------------------------------------
23012807-- Casting
@@ -2345,9 +2851,13 @@ cast arr =
23452851-- guarantees that unsafe calls will be performed in the calling thread. Making
23462852-- it safe to pass heap-allocated objects to unsafe functions.
23472853
2348−-- 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
23492855-- operations. And we are safe from Haskell perspective because we will be
23502856-- 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.
23512861
23522862-- | Use a @MutArray a@ as @Ptr a@. This is useful when we want to pass an
23532863-- array as a pointer to some operating system call or to a "safe" FFI call.
@@ -2365,48 +2875,377 @@ cast arr =
23652875--
23662876-- /Pre-release/
23672877--
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." #-}
23682885{-# INLINE asPtrUnsafe #-}
23692886asPtrUnsafe :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2370−asPtrUnsafe arr f =
2371− Unboxed.asPtrUnsafe
2887+asPtrUnsafe = unsafePinnedAsPtr
2888+
2889+{-# INLINE unsafeAsPtr #-}
2890+unsafeAsPtr :: MonadIO m => MutArray a -> (Ptr a -> m b) -> m b
2891+unsafeAsPtr arr f =
2892+ Unboxed.unsafeAsPtr
23722893 (arrContents arr) (\ptr -> f (ptr `plusPtr` arrStart arr))
23732894
23742895-------------------------------------------------------------------------------
23752896-- Equality
23762897-------------------------------------------------------------------------------
23772898
2378−-- | Compare the length of the arrays. If the length is equal, compare the
2379−-- lexicographical ordering of two underlying byte arrays otherwise return the
2380−-- result of length comparison.
2899+-- | Byte compare two arrays. Compare the length of the arrays. If the length
2900+-- is equal, compare the lexicographical ordering of two underlying byte arrays
2901+-- otherwise return the result of length comparison.
2902+--
2903+-- /Unsafe/: Note that the 'Unbox' instance of sum types with constructors of
2904+-- different sizes may leave some memory uninitialized which can make byte
2905+-- comparison unreliable.
23812906--
23822907-- /Pre-release/
2908+{-# INLINE byteCmp #-}
2909+byteCmp :: MonadIO m => MutArray a -> MutArray a -> m Ordering
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
2930+
23832931{-# INLINE cmp #-}
2932+{-# DEPRECATED cmp "Please use byteCmp instead." #-}
23842933cmp :: MonadIO m => MutArray a -> MutArray a -> m Ordering
2385−cmp arr1 arr2 =
2386− liftIO
2387− $ do
2388− let marr1 = getMutableByteArray# (arrContents arr1)
2389− marr2 = getMutableByteArray# (arrContents arr2)
2390− !(I# st1#) = arrStart arr1
2391− !(I# st2#) = arrStart arr2
2392− !(I# len#) = byteLength arr1
2393− case compare (byteLength arr1) (byteLength arr2) of
2394− EQ -> do
2395− r <- liftIO $ IO $ \s# ->
2396− let res =
2397− I#
2398− (compareByteArrays#
2399− (unsafeCoerce# marr1)
2400− st1#
2401− (unsafeCoerce# marr2)
2402− st2#
2403− len#)
2404− in (# s#, res #)
2405− return $ compare r 0
2406− x -> return x
2934+cmp = byteCmp
24072935
2936+-- | Byte equality of two arrays.
2937+--
2938+-- >>> byteEq arr1 arr2 = (==) EQ $ MArray.byteCmp arr1 arr2
2939+--
2940+-- /Unsafe/: See 'byteCmp'.
2941+{-# INLINE byteEq #-}
2942+byteEq :: MonadIO m => MutArray a -> MutArray a -> m Bool
2943+byteEq arr1 arr2 = fmap (EQ ==) $ byteCmp arr1 arr2
2944+
2945+-------------------------------------------------------------------------------
2946+-- Compact
24082947-------------------------------------------------------------------------------
2409−-- NFData
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
24103249-------------------------------------------------------------------------------
24113250
24123251-- | Strip elements which match with predicate from both ends.