I'm building a Conduit
that writes a binary file consisting of a header followed by a Double
matrix as a row-ordered list. Here's the code:
import Conduit ((.|), ConduitM, mapC, sinkFileBS, yield)
import Control.Monad.Trans.Except (ExceptT)
import Control.Monad.Trans.Resource (ResourceT)
import Data.ByteString (ByteString)
import Data.ByteString.Conversion (toByteString')
import Data.Serialize.IEEE754 (putFloat64be)
import Data.Serialize.Put (putListOf, runPut)
import Data.Void (Void)
import Numeric.LinearAlgebra.Data ((><), Matrix, toLists)
import System.FilePath (FilePath)
type FileWriter = ResourceT (ExceptT String IO)
matrixSink :: FilePath -> ConduitM (Matrix Double) Void FileWriter ()
matrixSink path = byteBuilder .| sinkFileBS path where
byteBuilder = do
yield $ toByteString' "header"
mapC fromDoubleMatrix
fromDoubleMatrix :: Matrix Double -> ByteString
fromDoubleMatrix matrix = runPut $
putListOf putFloat64be (concat toLists matrix)
This almost works. If I test it using
runExceptT . runConduitRes $ yield matrix .| matrixSink "test.dat"
where matrix = (2 >< 2) [1, 2, 3, 4]
I get the expected file but with an extra byte between the header and the list of doubles. When displayed using show
the extra byte looks like this:
"\NUL\NUL\NUL\NUL\NUL\NUL\NUL\t"
Any idea how not to print this byte? Or if it's a canonical separator or something (so that I can ignore it in the reader)?
EDIT: The problem seems to occur in the putListOf
construction in fromDoubleMatrix
.
putListOf :: Putter a -> Putter [a]
putListOf pa = \l -> do
putWord64be (fromIntegral (length l))
mapM_ pa l
putListOf
encodes the length of the list before encoding the individual list elements. I think maybe you are dealing with fixed 2x2 matrices so you don't need that length, and you just want:
fromDoubleMatrix :: Matrix Double -> ByteString
fromDoubleMatrix matrix = runPut $
mapM_ putFloat64be (concat toLists matrix)