{-# LANGUAGE ConstraintKinds #-} {-# LANGUAGE FlexibleContexts #-} {-# LANGUAGE FlexibleInstances #-} {-# LANGUAGE MultiParamTypeClasses #-} {-# LANGUAGE NamedFieldPuns #-} {-# LANGUAGE TypeFamilies #-} {-| Module: Data.Capnp.Message Description: Cap'N Proto messages This module provides support for working directly with Cap'N Proto messages. -} module Data.Capnp.Message ( -- * Reading and writing messages hPutMsg , hGetMsg , putMsg , getMsg -- * Limits on message size , maxSegmentSize , maxSegments -- * Converting between messages and 'ByteString's , encode , decode -- * Message type class , Message(..) -- * Immutable messages , empty , ConstMsg -- * Reading data from messages , getSegment , getWord -- * Mutable Messages , MutMsg , newMessage -- ** Allocating space in messages , alloc , allocInSeg , newSegment -- ** Modifying messages , setWord , setSegment , WriteCtx(..) ) where import Prelude hiding (read) import Data.Bits (shiftL) import Control.Monad (void, when, (>=>)) import Control.Monad.Catch (MonadThrow(..)) import Control.Monad.Primitive (PrimMonad, PrimState) import Control.Monad.State (evalStateT, get, put) import Control.Monad.Trans.Class (lift) import Control.Monad.Writer (execWriterT, tell) import Data.Bytes.Get (getWord32le, runGetS) import Data.ByteString.Internal (ByteString(..)) import Data.Either (fromRight) import Data.Maybe (fromJust) import Data.Primitive (MutVar, newMutVar, readMutVar, writeMutVar) import Data.Word (Word32, Word64) import System.Endian (fromLE64, toLE64) import System.IO (Handle, stdin, stdout) import qualified Data.ByteString as BS import qualified Data.ByteString.Builder as BB import qualified Data.Vector as V import qualified Data.Vector.Generic.Mutable as GMV import qualified Data.Vector.Mutable as MV import qualified Data.Vector.Storable as SV import qualified Data.Vector.Storable.Mutable as SMV import Data.Capnp.Address (WordAddr(..)) import Data.Capnp.Bits (WordCount(..), hi, lo) import Data.Capnp.TraversalLimit (LimitT, MonadLimit(invoice), evalLimitT) import Data.Mutable (Mutable(..)) import Internal.AppendVec (AppendVec) import Internal.Util (checkIndex) import qualified Internal.AppendVec as AppendVec -- | The maximum size of a segment supported by this libarary, in words. maxSegmentSize :: Int maxSegmentSize = 1 `shiftL` 28 -- 2 GiB. -- | The maximum number of segments allowed in a message by this library. maxSegments :: Int maxSegments = 1024 -- | A 'Message' is a (possibly read-only) capnproto message. It is -- parameterized over a monad in which operations are performed. class Monad m => Message m msg where -- | The type of segments in the message. data Segment msg -- | 'numSegs' gets the number of segments in a message. numSegs :: msg -> m Int -- | @'internalGetSeg' message index@ gets the segment at index 'index' -- in 'message'. Most callers should use the 'getSegment' wrapper, instead -- of calling this directly. internalGetSeg :: msg -> Int -> m (Segment msg) -- | Get the length of the segment, in units of 64-bit words. numWords :: Segment msg -> m Int -- | @'slice' start length segment@ extracts a sub-section of the segment, -- starting at index @start@, of length @length@. slice :: Int -> Int -> Segment msg -> m (Segment msg) -- | @'read' segment index@ reads a 64-bit word from the segement at the -- given index. Consider using 'getWord' on the message, instead of -- calling this directly. read :: Segment msg -> Int -> m Word64 -- | Convert a ByteString to a segment. fromByteString :: ByteString -> m (Segment msg) -- | Convert a segment to a byte string. toByteString :: Segment msg -> m ByteString -- | @'getSegment' message index@ fetches the given segment in the message. -- It throws a @BoundsError@ if the address is out of bounds. getSegment :: (MonadThrow m, Message m msg) => msg -> Int -> m (Segment msg) getSegment msg i = do checkIndex i =<< numSegs msg internalGetSeg msg i -- | @'getWord' msg addr@ returns the word at @addr@ within @msg@. It throws a -- @BoundsError@ if the address is out of bounds. getWord :: (MonadThrow m, Message m msg) => msg -> WordAddr -> m Word64 getWord msg WordAt{wordIndex=wordIndex@(WordCount i), segIndex} = do seg <- getSegment msg segIndex checkIndex i =<< numWords seg seg `read` i -- | @'setSegment' message index segment@ sets the segment at the given index -- in the message. It throws a @BoundsError@ if the address is out of bounds. setSegment :: (WriteCtx m s, MonadThrow m) => MutMsg s -> Int -> Segment (MutMsg s) -> m () setSegment msg i seg = do checkIndex i =<< numSegs msg internalSetSeg msg i seg -- | @'setWord' message address value@ sets the word at @address@ in the -- message to @value@. If the address is not valid in the message, a -- @BoundsError@ will be thrown. setWord :: (WriteCtx m s, MonadThrow m) => MutMsg s -> WordAddr -> Word64 -> m () setWord msg WordAt{wordIndex=WordCount i, segIndex} val = do seg <- getSegment msg segIndex checkIndex i =<< numWords seg write seg i val -- | A read-only capnproto message. -- -- 'ConstMsg' is an instance of the generic 'Message' type class. its -- implementations of 'toByteString' and 'fromByteString' are O(1); -- the underlying bytes are not copied. newtype ConstMsg = ConstMsg (V.Vector (Segment ConstMsg)) instance Monad m => Message m ConstMsg where newtype Segment ConstMsg = ConstSegment { constSegToVec :: SV.Vector Word64 } numSegs (ConstMsg vec) = pure $ V.length vec internalGetSeg (ConstMsg vec) i = vec `V.indexM` i numWords (ConstSegment vec) = pure $ SV.length vec slice start len (ConstSegment vec) = pure $ ConstSegment (SV.slice start len vec) read (ConstSegment vec) i = fromLE64 <$> vec `SV.indexM` i -- FIXME: Verify that the pointer is actually 64-bit aligned before casting. fromByteString (PS fptr offset len) = pure $ ConstSegment (SV.unsafeCast $ SV.unsafeFromForeignPtr fptr offset len) toByteString (ConstSegment vec) = pure $ PS fptr offset len where (fptr, offset, len) = SV.unsafeToForeignPtr (SV.unsafeCast vec) -- | 'decode' decodes a message from a bytestring. -- -- The segments will not be copied; the resulting message will be a view into -- the original bytestring. Runs in O(number of segments in the message). decode :: MonadThrow m => ByteString -> m ConstMsg decode bytes = fromByteString bytes >>= decodeSeg -- | 'encode' encodes a message as a bytestring builder. encode :: Monad m => ConstMsg -> m BB.Builder encode msg = -- We use Maybe as the MonadThrow instance required by -- writeMessage/toByteString, but we know this can't actually fail, -- so we ignore errors. TODO: we should get rid of the Monad constraint -- on this function and just have the tyep be ConstMsg -> BB.Builder, -- but that will have some cascading api effects, so we're deferring -- that for a bit. pure $ fromJust $ execWriterT $ writeMessage msg (tell . BB.word32LE) (toByteString >=> tell . BB.byteString) -- | 'decodeSeg' decodes a message from a segment, treating the segment as if -- it were raw bytes. -- -- this is mostly here as a helper for 'decode'. decodeSeg :: MonadThrow m => Segment ConstMsg -> m ConstMsg decodeSeg seg = do len <- numWords seg flip evalStateT (Nothing, 0) $ evalLimitT len $ -- Note: we use the traversal limit to avoid needing to do bounds checking -- here; since readMessage invoices the limit before reading, we can rely -- on it not to read past the end of the blob. -- -- TODO: while this works, it means that we throw 'TraversalLimitError' -- on failure, which makes for a confusing API. readMessage read32 readSegment where read32 = do (cur, idx) <- get case cur of Just n -> do put (Nothing, idx) return n Nothing -> do word <- lift $ lift $ read seg idx put (Just $ hi word, idx + 1) return (lo word) readSegment (WordCount len) = do (cur, idx) <- get put (cur, idx + len) lift $ lift $ slice idx len seg -- | @'readMessage' read32 readSegment@ reads in a message using the -- monadic context, which should manage the current read position, -- into a message. read32 should read a 32-bit little-endian integer, -- and @readSegment n@ should read a blob of @n@ 64-bit words. -- The size of the message (in 64-bit words) is deducted from the traversal, -- limit which can be used to set the maximum message size. readMessage :: (MonadThrow m, MonadLimit m) => m Word32 -> (WordCount -> m (Segment ConstMsg)) -> m ConstMsg readMessage read32 readSegment = do invoice 1 numSegs' <- read32 let numSegs = numSegs' + 1 invoice (fromIntegral numSegs `div` 2) segSizes <- V.replicateM (fromIntegral numSegs) read32 when (numSegs `mod` 2 == 0) $ void read32 V.mapM_ (invoice . fromIntegral) segSizes ConstMsg <$> V.mapM (readSegment . fromIntegral) segSizes -- | @'writeMesage' write32 writeSegment@ writes out the message. @write32@ -- should write a 32-bit word in little-endian format to the output stream. -- @writeSegment@ should write a blob. writeMessage :: MonadThrow m => ConstMsg -> (Word32 -> m ()) -> (Segment ConstMsg -> m ()) -> m () writeMessage (ConstMsg segs) write32 writeSegment = do let numSegs = V.length segs write32 (fromIntegral numSegs - 1) V.forM_ segs $ \seg -> write32 =<< fromIntegral <$> numWords seg when (numSegs `mod` 2 == 0) $ write32 0 V.forM_ segs writeSegment -- | @'hPutMsg' handle msg@ writes @msg@ to @handle@. If there is an exception, -- it will be an 'IOError' raised by the underlying IO libraries. hPutMsg :: Handle -> ConstMsg -> IO () hPutMsg handle msg = encode msg >>= BB.hPutBuilder handle -- | Equivalent to @'hPutMsg' 'stdout'@ putMsg :: ConstMsg -> IO () putMsg = hPutMsg stdout -- | @'hGetMsg' handle limit@ reads a message from @handle@ that is at most -- @limit@ 64-bit words in length. hGetMsg :: Handle -> Int -> IO ConstMsg hGetMsg handle size = evalLimitT size $ readMessage read32 readSegment where read32 :: LimitT IO Word32 read32 = lift $ do bytes <- BS.hGet handle 4 -- The only way we get a left is if we get less than 4 bytes, in which -- case hGet should have thrown: pure $ fromRight (error "impossible") (runGetS getWord32le bytes) readSegment n = lift $ BS.hGet handle (fromIntegral n * 8) >>= fromByteString -- | Equivalent to @'hGetMsg' 'stdin'@ getMsg :: Int -> IO ConstMsg getMsg = hGetMsg stdin -- | A 'MutMsg' is a mutable capnproto message. The type parameter @s@ is the -- state token for the instance of 'PrimMonad' in which the message may be -- modified. -- -- Due to mutabilty, the implementations of 'toByteString' and 'fromByteString' -- must make full copies, and so are O(n) in the length of the segment. newtype MutMsg s = MutMsg (MutVar s (AppendVec MV.MVector s (Segment (MutMsg s)))) deriving(Eq) -- | 'WriteCtx' is the context needed for most write operations. type WriteCtx m s = (PrimMonad m, s ~ PrimState m, MonadThrow m) instance (PrimMonad m, s ~ PrimState m) => Message m (MutMsg s) where newtype Segment (MutMsg s) = MutSegment (AppendVec SMV.MVector s Word64) numWords (MutSegment mseg) = pure $ GMV.length (AppendVec.getVector mseg) slice start len (MutSegment mseg) = pure $ MutSegment $ AppendVec.fromVector $ SMV.slice start len (AppendVec.getVector mseg) read (MutSegment mseg) i = fromLE64 <$> SMV.read (AppendVec.getVector mseg) i fromByteString bytes = do vec <- constSegToVec <$> fromByteString bytes MutSegment . AppendVec.fromVector <$> SV.thaw vec toByteString mseg = do seg <- freeze mseg toByteString (seg :: Segment ConstMsg) numSegs (MutMsg segVar) = GMV.length . AppendVec.getVector <$> readMutVar segVar internalGetSeg (MutMsg segVar) i = do segs <- AppendVec.getVector <$> readMutVar segVar MV.read segs i -- | @'internalSetSeg' message index segment@ sets the segment at the given -- index in the message. Most callers should use the 'setSegment' wrapper, -- instead of calling this directly. internalSetSeg :: WriteCtx m s => MutMsg s -> Int -> Segment (MutMsg s) -> m () internalSetSeg (MutMsg segVar) segIndex seg = do segs <- AppendVec.getVector <$> readMutVar segVar MV.write segs segIndex seg -- | @'write' segment index value@ writes a value to the 64-bit word -- at the provided index. Consider using 'setWord' on the message, -- instead of calling this directly. write :: WriteCtx m s => Segment (MutMsg s) -> Int -> Word64 -> m () write (MutSegment seg) i val = SMV.write (AppendVec.getVector seg) i (toLE64 val) -- | @'grow' segment amount@ grows the segment by the specified number -- of 64-bit words. The original segment should not be used afterwards. grow :: WriteCtx m s => Segment (MutMsg s) -> Int -> m (Segment (MutMsg s)) grow (MutSegment vec) amount = MutSegment <$> AppendVec.grow vec amount maxSegmentSize -- | @'newSegment' msg sizeHint@ allocates a new, initially empty segment in -- @msg@ with a capacity of @sizeHint@. It returns the a pair of the segment -- number and the segment itself. Amortized O(1). newSegment :: WriteCtx m s => MutMsg s -> Int -> m (Int, Segment (MutMsg s)) newSegment msg@(MutMsg segVar) sizeHint = do -- the next segment number will be equal to the *current* number of -- segments: segIndex <- numSegs msg -- make space for th new segment segs <- readMutVar segVar segs <- AppendVec.grow segs 1 maxSegments writeMutVar segVar segs newSeg <- MutSegment . AppendVec.makeEmpty <$> SMV.new sizeHint setSegment msg segIndex newSeg pure (segIndex, newSeg) -- | Like 'alloc', but the second argument allows the caller to specify the -- index of the segment in which to allocate the data. allocInSeg :: WriteCtx m s => MutMsg s -> Int -> WordCount -> m WordAddr allocInSeg msg segIndex (WordCount size) = do oldSeg@(MutSegment vec) <- getSegment msg segIndex let ret = WordAt { segIndex , wordIndex = WordCount $ GMV.length $ AppendVec.getVector vec } newSeg <- grow oldSeg size setSegment msg segIndex newSeg pure ret -- | @'alloc' size@ allocates 'size' words within a message. it returns the -- starting address of the allocated memory. alloc :: WriteCtx m s => MutMsg s -> WordCount -> m WordAddr alloc msg size = do segIndex <- pred <$> numSegs msg allocInSeg msg segIndex size -- | 'empty' is an empty message, i.e. a minimal message with a null pointer as -- its root object. empty :: ConstMsg empty = ConstMsg $ V.fromList [ ConstSegment $ SV.fromList [0] ] -- | Allocate a new empty message. newMessage :: WriteCtx m s => m (MutMsg s) newMessage = thaw empty instance Thaw (Segment ConstMsg) where type Mutable s (Segment ConstMsg) = Segment (MutMsg s) thaw = thawSeg thaw unsafeThaw = thawSeg unsafeThaw freeze = freezeSeg freeze unsafeFreeze = freezeSeg unsafeFreeze -- Helpers for @Segment ConstMsg@'s Thaw instance. thawSeg thaw (ConstSegment vec) = MutSegment <$> thaw (AppendVec.FrozenAppendVec vec) freezeSeg freeze (MutSegment mvec) = ConstSegment . AppendVec.getFrozenVector <$> freeze mvec instance Thaw ConstMsg where type Mutable s ConstMsg = MutMsg s thaw = thawMsg thaw unsafeThaw = thawMsg unsafeThaw freeze = freezeMsg freeze unsafeFreeze = freezeMsg unsafeFreeze -- Helpers for ConstMsg's Thaw instance. thawMsg thaw (ConstMsg vec) = do segments <- V.mapM thaw vec >>= V.unsafeThaw MutMsg <$> newMutVar (AppendVec.fromVector segments) freezeMsg freeze msg = do len <- numSegs msg ConstMsg <$> V.generateM len (internalGetSeg msg >=> freeze)