2{-# LANGUAGE BangPatterns #-}
3{-# LANGUAGE MagicHash #-}
4{-# LANGUAGE RecordWildCards #-}
5{-# LANGUAGE UnboxedTuples #-}
10-- Module : Streamly.FileSystem.FD
11-- Copyright : (c) 2019 Composewell Technologies
14-- Maintainer : streamly@composewell.com
15-- Stability : experimental
18-- This module is a an experimental replacement for
19-- "Streamly.FileSystem.Handle". The former module provides IO facilities based
20-- on the GHC Handle type. The APIs in this module avoid the GHC handle layer
21-- and provide more explicit control over buffering.
23-- Read and write data as streams and arrays to and from files.
25-- This module provides read and write APIs based on handles. Before reading or
26-- writing, a file must be opened first using 'openFile'. The 'Handle' returned
27-- by 'openFile' is then used to access the file. A 'Handle' is backed by an
28-- operating system file descriptor. When the 'Handle' is garbage collected the
29-- underlying file descriptor is automatically closed. A handle can be
30-- explicitly closed using 'closeFile'.
32-- Reading and writing APIs are divided into two categories, sequential
33-- streaming APIs and random or seekable access APIs. File IO APIs are quite
34-- similar to "Streamly.Memory.Array" read write APIs. In that regard, arrays can
35-- be considered as in-memory files or files can be considered as on-disk
38-- > import qualified Streamly.FileSystem.FD as FD
41module Streamly.FileSystem.FD
50 -- TODO file path based APIs
55 -- | Streaming APIs read or write data to or from a file or device
56 -- sequentially, they never perform a seek to a random location. When
57 -- reading, the stream is lazy and generated on-demand as the consumer
58 -- consumes it. Read IO requests to the IO device are performed in chunks
59 -- of 32KiB, this is referred to as @defaultChunkSize@ in the
60 -- documentation. One IO request may or may not read the full chunk. If the
61 -- whole stream is not consumed, it is possible that we may read slightly
62 -- more from the IO device than what the consumer needed. Unless specified
63 -- otherwise in the API, writes are collected into chunks of
64 -- @defaultChunkSize@ before they are written to the IO device.
66 -- Streaming APIs work for all kind of devices, seekable or non-seekable;
67 -- including disks, files, memory devices, terminals, pipes, sockets and
68 -- fifos. While random access APIs work only for files or devices that have
69 -- random access or seek capability for example disks, memory devices.
70 -- Devices like terminals, pipes, sockets and fifos do not have random
73 -- ** Read File to Stream
88 -- ** Write File from Stream
98 , writeArraysPackedUpto
100 -- XXX these are incomplete
102 -- , writevArraysPackedUpto
104 -- -- * Random Access (Seek)
105 -- -- | Unlike the streaming APIs listed above, these APIs apply to devices or
106 -- files that have random access or seek capability. This type of devices
107 -- include disks, files, memory devices and exclude terminals, pipes,
108 -- sockets and fifos.
113 -- , readAt -- read from a given position to th end of file
114 -- , readSliceArrayUpto
115 -- , readSliceArrayOf
120 -- , writeAt -- start writing at the given position
125import Control.Monad.IO.Class (MonadIO(..))
126import Data.Word (Word8)
127import Foreign.ForeignPtr (withForeignPtr)
128-- import Foreign.ForeignPtr.Unsafe (unsafeForeignPtrToPtr)
129import Foreign.Ptr (plusPtr, castPtr)
130import Foreign.Storable (Storable(..))
131import GHC.ForeignPtr (mallocPlainForeignPtrBytes)
132-- import System.IO (Handle, hGetBufSome, hPutBuf)
133import System.IO (IOMode)
134import Prelude hiding (read)
136import qualified GHC.IO.FD as FD
137import qualified GHC.IO.Device as RawIO
139import Streamly.Internal.Memory.Array.Types (Array(..), byteLength, defaultChunkSize)
140import Streamly.Internal.Data.Stream.Serial (SerialT)
141import Streamly.Internal.Data.Stream.StreamK.Type (IsStream, mkStream)
143#if !defined(mingw32_HOST_OS)
144import Streamly.Internal.Memory.Array.Types (groupIOVecsOf)
145import Streamly.Internal.Data.Stream.StreamD (toStreamD)
146import Streamly.Internal.Data.Stream.StreamD.Type (fromStreamD)
147import qualified Streamly.FileSystem.FDIO as RawIO hiding (write)
149-- import Streamly.Data.Fold (Fold)
150-- import Streamly.String (encodeUtf8, decodeUtf8, foldLines)
152import qualified Streamly.Memory.Array as A
153import qualified Streamly.Internal.Memory.ArrayStream as AS
154import qualified Streamly.Prelude as S
155import qualified Streamly.Internal.Data.Stream.StreamD.Type as D
157-------------------------------------------------------------------------------
159-------------------------------------------------------------------------------
161-- The following references may be useful to build an understanding about the
164-- http://www.linux-mag.com/id/308/ for blocking/non-blocking IO on linux.
165-- https://lwn.net/Articles/612483/ Non-blocking buffered file read operations
166-- https://en.wikipedia.org/wiki/C_file_input/output for C APIs.
167-- https://docs.oracle.com/javase/tutorial/essential/io/file.html for Java API.
168-- https://www.w3.org/TR/FileAPI/ for http file API.
170-------------------------------------------------------------------------------
172-------------------------------------------------------------------------------
174-- XXX attach a finalizer
175-- | A 'Handle' is returned by 'openFile' and is subsequently used to perform
176-- read and write operations on a file.
178newtype Handle = Handle FD.FD
180-- | File handle for standard input
182stdin = Handle FD.stdin
184-- | File handle for standard output
186stdout = Handle FD.stdout
188-- | File handle for standard error
190stderr = Handle FD.stderr
192-- XXX we can support all the flags that the "open" system call supports.
193-- Instead of using RTS locking mechanism can we use system provided locking
196-- | Open a file that is not a directory and return a file handle.
197-- 'openFile' enforces a multiple-reader single-writer locking on files. That
198-- is, there may either be many handles on the same file which manage input, or
199-- just one handle on the file which manages output. If any open handle is
200-- managing a file for output, no new handle can be allocated for that file. If
201-- any open handle is managing a file for input, new handles can only be
202-- allocated if they do not manage output. Whether two files are the same is
203-- implementation-dependent, but they should normally be the same if they have
204-- the same absolute path name and neither has been renamed, for example.
206openFile :: FilePath -> IOMode -> IO Handle
207openFile path mode = fmap (Handle . fst) $ FD.openFile path mode True
209-------------------------------------------------------------------------------
211-------------------------------------------------------------------------------
213-- | Read a 'ByteArray' from a file handle. If no data is available on the
214-- handle it blocks until some data becomes available. If data is available
215-- then it immediately returns that data without blocking. It reads a maximum
216-- of up to the size requested.
217{-# INLINABLE readArrayUpto #-}
218readArrayUpto :: Int -> Handle -> IO (Array Word8)
219readArrayUpto size (Handle fd) = do
220 ptr <- mallocPlainForeignPtrBytes size
221 -- ptr <- mallocPlainForeignPtrAlignedBytes size (alignment (undefined :: Word8))
222 withForeignPtr ptr $ \p -> do
223 -- n <- hGetBufSome h p size
224 n <- RawIO.read fd p size
227 , aEnd = p `plusPtr` n
228 , aBound = p `plusPtr` size
230 -- XXX shrink only if the diff is significant
234-------------------------------------------------------------------------------
236-------------------------------------------------------------------------------
238-- | Write an 'Array' to a file handle.
241{-# INLINABLE writeArray #-}
242writeArray :: Storable a => Handle -> Array a -> IO ()
243writeArray _ arr | A.length arr == 0 = return ()
244writeArray (Handle fd) arr = withForeignPtr (aStart arr) $ \p -> do
245 -- RawIO.writeAll fd (castPtr p) aLen
246 RawIO.write fd (castPtr p) aLen
248 -- Experiment to compare "writev" based IO with "write" based IO.
250 let iov' = iov {aEnd = aBound iov}
251 A.writeIndex iov' 0 (RawIO.IOVec (castPtr p) (fromIntegral aLen))
252 RawIO.writevAll fd (unsafeForeignPtrToPtr (aStart iov')) 1
255 aLen = byteLength arr
257#if !defined(mingw32_HOST_OS)
258-- | Write an array of 'IOVec' to a file handle.
261{-# INLINABLE writeIOVec #-}
262writeIOVec :: Handle -> Array RawIO.IOVec -> IO ()
263writeIOVec _ iov | A.length iov == 0 = return ()
264writeIOVec (Handle fd) iov =
265 withForeignPtr (aStart iov) $ \p ->
266 RawIO.writevAll fd p (A.length iov)
269-------------------------------------------------------------------------------
270-- Stream of Arrays IO
271-------------------------------------------------------------------------------
273-- | @readArraysOfUpto size h@ reads a stream of arrays from file handle @h@.
274-- The maximum size of a single array is specified by @size@. The actual size
275-- read may be less than or equal to @size@.
276{-# INLINABLE _readArraysOfUpto #-}
277_readArraysOfUpto :: (IsStream t, MonadIO m)
278 => Int -> Handle -> t m (Array Word8)
279_readArraysOfUpto size h = go
281 -- XXX use cons/nil instead
282 go = mkStream $ \_ yld _ stp -> do
283 arr <- liftIO $ readArrayUpto size h
288{-# INLINE_NORMAL readArraysOfUpto #-}
289readArraysOfUpto :: (IsStream t, MonadIO m)
290 => Int -> Handle -> t m (Array Word8)
291readArraysOfUpto size h = D.fromStreamD (D.Stream step ())
293 {-# INLINE_LATE step #-}
295 arr <- liftIO $ readArrayUpto size h
301-- XXX read 'Array a' instead of Word8
303-- | @readArrays h@ reads a stream of arrays from file handle @h@.
304-- The maximum size of a single array is limited to @defaultChunkSize@.
305-- 'readArrays' ignores the prevailing 'TextEncoding' and 'NewlineMode'
308-- > readArrays = readArraysOfUpto defaultChunkSize
311{-# INLINE readArrays #-}
312readArrays :: (IsStream t, MonadIO m) => Handle -> t m (Array Word8)
313readArrays = readArraysOfUpto defaultChunkSize
315-------------------------------------------------------------------------------
316-- Read File to Stream
317-------------------------------------------------------------------------------
319-- TODO for concurrent streams implement readahead IO. We can send multiple
320-- read requests at the same time. For serial case we can use async IO. We can
321-- also control the read throughput in mbps or IOPS.
323-- | @readInChunksOf chunkSize handle@ reads a byte stream from a file handle,
324-- reads are performed in chunks of up to @chunkSize@. The stream ends as soon
325-- as EOF is encountered.
327{-# INLINE readInChunksOf #-}
328readInChunksOf :: (IsStream t, MonadIO m) => Int -> Handle -> t m Word8
329readInChunksOf chunkSize h = AS.concat $ readArraysOfUpto chunkSize h
332-- read :: (IsStream t, MonadIO m, Storable a) => Handle -> t m a
334-- > read = 'readByChunks' A.defaultChunkSize
335-- | Generate a stream of elements of the given type from a file 'Handle'. The
336-- stream ends when EOF is encountered.
340read :: (IsStream t, MonadIO m) => Handle -> t m Word8
341read = AS.concat . readArrays
343-------------------------------------------------------------------------------
345-------------------------------------------------------------------------------
347-- | Write a stream of arrays to a handle.
350{-# INLINE writeArrays #-}
351writeArrays :: (MonadIO m, Storable a) => Handle -> SerialT m (Array a) -> m ()
352writeArrays h m = S.mapM_ (liftIO . writeArray h) m
354-- | Write a stream of arrays to a handle after coalescing them in chunks of
355-- specified size. The chunk size is only a maximum and the actual writes could
356-- be smaller than that as we do not split the arrays to fit them to the
360{-# INLINE writeArraysPackedUpto #-}
361writeArraysPackedUpto :: (MonadIO m, Storable a)
362 => Int -> Handle -> SerialT m (Array a) -> m ()
363writeArraysPackedUpto n h xs = writeArrays h $ AS.compact n xs
365#if !defined(mingw32_HOST_OS)
366-- XXX this is incomplete
367-- | Write a stream of 'IOVec' arrays to a handle.
371writev :: MonadIO m => Handle -> SerialT m (Array RawIO.IOVec) -> m ()
372writev h m = S.mapM_ (liftIO . writeIOVec h) m
374-- XXX this is incomplete
375-- | Write a stream of arrays to a handle after grouping them in 'IOVec' arrays
376-- of up to a maximum total size. Writes are performed using gather IO via
377-- @writev@ system call. The maximum number of entries in each 'IOVec' group
381{-# INLINE _writevArraysPackedUpto #-}
382_writevArraysPackedUpto :: MonadIO m
383 => Int -> Handle -> SerialT m (Array a) -> m ()
384_writevArraysPackedUpto n h xs =
385 writev h $ fromStreamD $ groupIOVecsOf n 512 (toStreamD xs)
388-- GHC buffer size dEFAULT_FD_BUFFER_SIZE=8192 bytes.
391-- Note that if you use a chunk size less than 8K (GHC's default buffer
392-- size) then you are advised to use 'NOBuffering' mode on the 'Handle' in case you
393-- do not want buffering to occur at GHC level as well. Same thing applies to
396-- | Like 'write' but provides control over the write buffer. Output will
397-- be written to the IO device as soon as we collect the specified number of
401{-# INLINE writeInChunksOf #-}
402writeInChunksOf :: MonadIO m => Int -> Handle -> SerialT m Word8 -> m ()
403writeInChunksOf n h m = writeArrays h $ AS.arraysOf n m
405-- > write = 'writeInChunksOf' A.defaultChunkSize
407-- | Write a byte stream to a file handle. Combines the bytes in chunks of size
408-- up to 'A.defaultChunkSize' before writing. Note that the write behavior
409-- depends on the 'IOMode' and the current seek position of the handle.
413write :: MonadIO m => Handle -> SerialT m Word8 -> m ()
414write = writeInChunksOf defaultChunkSize
418write :: (MonadIO m, Storable a) => Handle -> SerialT m a -> m ()
419write = toHandleWith A.defaultChunkSize
422-------------------------------------------------------------------------------
423-- IO with encoding/decoding Unicode characters
424-------------------------------------------------------------------------------
428-- > readUtf8 = decodeUtf8 . read
430-- Read a UTF8 encoded stream of unicode characters from a file handle.
433{-# INLINE readUtf8 #-}
434readUtf8 :: (IsStream t, MonadIO m) => Handle -> t m Char
435readUtf8 = decodeUtf8 . read
438-- > writeUtf8 h s = write h $ encodeUtf8 s
440-- Encode a stream of unicode characters to UTF8 and write it to the given file
441-- handle. Default block buffering applies to the writes.
444{-# INLINE writeUtf8 #-}
445writeUtf8 :: MonadIO m => Handle -> SerialT m Char -> m ()
446writeUtf8 h s = write h $ encodeUtf8 s
448-- | Write a stream of unicode characters after encoding to UTF-8 in chunks
449-- separated by a linefeed character @'\n'@. If the size of the buffer exceeds
450-- @defaultChunkSize@ and a linefeed is not yet found, the buffer is written
451-- anyway. This is similar to writing to a 'Handle' with the 'LineBuffering'
455{-# INLINE writeUtf8ByLines #-}
456writeUtf8ByLines :: (IsStream t, MonadIO m) => Handle -> t m Char -> m ()
457writeUtf8ByLines = undefined
459-- | Read UTF-8 lines from a file handle and apply the specified fold to each
460-- line. This is similar to reading a 'Handle' with the 'LineBuffering' option.
463{-# INLINE readLines #-}
464readLines :: (IsStream t, MonadIO m) => Handle -> Fold m Char b -> t m b
465readLines h f = foldLines (readUtf8 h) f
467-------------------------------------------------------------------------------
468-- Framing on a sequence
469-------------------------------------------------------------------------------
471-- | Read a stream from a file handle and split it into frames delimited by
472-- the specified sequence of elements. The supplied fold is applied on each
476{-# INLINE readFrames #-}
477readFrames :: (IsStream t, MonadIO m, Storable a)
478 => Array a -> Handle -> Fold m a b -> t m b
479readFrames = undefined -- foldFrames . read
481-- | Write a stream to the given file handle buffering up to frames separated
482-- by the given sequence or up to a maximum of @defaultChunkSize@.
485{-# INLINE writeByFrames #-}
486writeByFrames :: (IsStream t, MonadIO m, Storable a)
487 => Array a -> Handle -> t m a -> m ()
488writeByFrames = undefined
490-------------------------------------------------------------------------------
491-- Random Access IO (Seek)
492-------------------------------------------------------------------------------
494-- XXX handles could be shared, so we may not want to use the handle state at
495-- all for these APIs. we can use pread and pwrite instead. On windows we will
496-- need to use readFile/writeFile with an offset argument.
498-------------------------------------------------------------------------------
500-- | Read the element at the given index treating the file as an array.
503{-# INLINE readIndex #-}
504readIndex :: Storable a => Handle -> Int -> Maybe a
505readIndex arr i = undefined
507-- NOTE: To represent a range to read we have chosen (start, size) instead of
508-- (start, end). This helps in removing the ambiguity of whether "end" is
509-- included in the range or not.
511-- We could avoid specifying the range to be read and instead use "take size"
512-- on the stream, but it may end up reading more and then consume it partially.
514-- | @readSliceWith chunkSize handle pos len@ reads up to @len@ bytes
515-- from @handle@ starting at the offset @pos@ from the beginning of the file.
517-- Reads are performed in chunks of size @chunkSize@. For block devices, to
518-- avoid reading partial blocks @chunkSize@ must align with the block size of
519-- the underlying device. If the underlying block size is unknown, it is a good
520-- idea to keep it a multiple 4KiB. This API ensures that the start of each
521-- chunk is aligned with @chunkSize@ from second chunk onwards.
523{-# INLINE readSliceWith #-}
524readSliceWith :: (IsStream t, MonadIO m, Storable a)
525 => Int -> Handle -> Int -> Int -> t m a
526readSliceWith chunkSize h pos len = undefined
528-- | @readSlice h i count@ streams a slice from the file handle @h@ starting
529-- at index @i@ and reading up to @count@ elements in the forward direction
530-- ending at the index @i + count - 1@.
533{-# INLINE readSlice #-}
534readSlice :: (IsStream t, MonadIO m, Storable a)
535 => Handle -> Int -> Int -> t m a
536readSlice = readSliceWith defaultChunkSize
538-- | @readSliceRev h i count@ streams a slice from the file handle @h@ starting
539-- at index @i@ and reading up to @count@ elements in the reverse direction
540-- ending at the index @i - count + 1@.
543{-# INLINE readSliceRev #-}
544readSliceRev :: (IsStream t, MonadIO m, Storable a)
545 => Handle -> Int -> Int -> t m a
546readSliceRev h i count = undefined
548-- | Write the given element at the given index in the file.
551{-# INLINE writeIndex #-}
552writeIndex :: (MonadIO m, Storable a) => Handle -> Int -> a -> m ()
553writeIndex h i a = undefined
555-- | @writeSlice h i count stream@ writes a stream to the file handle @h@
556-- starting at index @i@ and writing up to @count@ elements in the forward
557-- direction ending at the index @i + count - 1@.
560{-# INLINE writeSlice #-}
561writeSlice :: (IsStream t, Monad m, Storable a)
562 => Handle -> Int -> Int -> t m a -> m ()
563writeSlice h i len s = undefined
565-- | @writeSliceRev h i count stream@ writes a stream to the file handle @h@
566-- starting at index @i@ and writing up to @count@ elements in the reverse
567-- direction ending at the index @i - count + 1@.
570{-# INLINE writeSliceRev #-}
571writeSliceRev :: (IsStream t, Monad m, Storable a)
572 => Handle -> Int -> Int -> t m a -> m ()
573writeSliceRev arr i len s = undefined
keysesc to close
- j k
- move down, up
- enter
- open
- esc
- back, or up a level
- 1 2 3 4
- readme, files, commits, refs
- /
- filter
- y
- copy the clone command
- c
- compare the chosen commit with another (or shift-click both)
- ?
- show or hide this list