Regarding NewBinary... I think my challenge is how to add endian- conversion without duplicating all the put and get methods from NewBinary. I would still use that bit of TH code to figure out whether my platform is big or little endian. I don't care about cross- compilation and what that code does is analogous to #ifdef ... #define LITTLE_ENDIAN ... #endif, etc. I'm looking for architectural suggestions, though. Should I define LittleEndian Word32, BigEndian Word16, etc. or should I have Endian Word32 Big, Endian Word16 Little, etc. Thanks, Joel On Oct 5, 2005, at 11:42 AM, Udo Stenzel wrote:
Why don't you pull out 4 bytes and assemble them manually?
To that I'd like to add a snippet from NewBinary itself:
| instance Binary Word32 where | put_ h w = do | putByte h (fromIntegral (w `shiftR` 24)) | putByte h (fromIntegral ((w `shiftR` 16) .&. 0xff)) | putByte h (fromIntegral ((w `shiftR` 8) .&. 0xff)) | putByte h (fromIntegral (w .&. 0xff)) | get h = do | w1 <- getWord8 h | w2 <- getWord8 h | w3 <- getWord8 h | w4 <- getWord8 h | return $! ((fromIntegral w1 `shiftL` 24) .|. | (fromIntegral w2 `shiftL` 16) .|. | (fromIntegral w3 `shiftL` 8) .|. | (fromIntegral w4))
This obviously writes a Word32 in big endian format, also known as "network byte order", and doesn't care how the host platform stores integers. No need for `hton' and `ntoh'. To convert it to write little endian, just copy it and reorder some lines. (But I think, writing LE integers with no good reason and without an enclosing protocol that explicitly declares them (like IIOP) is a bad idea.)