CODEWITHSUNDEEP
scalascalazshapelessscodec
Chris Ridmann
Chris Ridmann

Reputation: 2896

scodec decode/encode split length field

I am using scodec: https://github.com/scodec/scodec to decode/encode a binary protocol.

I am struggling with a part of the spec where a "length" field is split into two parts by a "moreflag". The moreflag indicates if the length field needs more space.

Example:
Byte 1:  identifier bits 8-7, moreFlag bit 6, length bits 5-0    // first length field filled with 0's if moreFlag is false
Byte 2:  moreFlag bit 8, length bits 7-0  
Byte 3:  otherJunk bits 8-0

My problem is I want to encode/decode both of these length fields into a single case class field:

case class Header(totalLength: Int, otherJunk: Int)

I have tried a few different things, but nothing has worked out so far:

implicit val headerCodec: Codec[Header] = (
  ("identifier" | uint2) :~>:
  ("moreFlag" | bool).compact >>:~ { meta => 
    if (meta.last) {
      // more flag enabled, combine lengths somehow
      ("first length part" | uint(5)) :: ("moreFlag2DontCare" | uint(1) :~>: ("second length part - how to combine?" | uint(7)) :: ("otherJunk" | uint8) 
    }
    else {
      ("first length part always 0s" | constant(bin"00000")) :: ("moreFlag2DontCare" | uint(1) :~>: ("fullLength" | uint(7)) :: ("otherJunk" | uint8)
    }
  }
).as[Header]

Am I on the right track here? Thanks!

Upvotes: 2

Views: 656

Answers (1)

mpilquist
mpilquist

Reputation: 3855

One way to accomplish this is to use the predefined combinators to define the structure of the binary format -- not the combination logic. Then put the combination logic inside a function passed to xmap.

import scodec._
import bits._
import codecs._
import shapeless._

case class Header(totalLength: Int, otherJunk: Int)

object Header {
  implicit val codec: Codec[Header] = {
    type Struct = Int :: Boolean :: BitVector :: Boolean :: BitVector :: Int :: HNil
    val struct: Codec[Struct] = ("identifier" | uint2) :: ("moreFlag" | bool) :: ("upper length bits" | codecs.bits(5)) :: ("moreFlag2" | bool) :: ("lower length bits" | codecs.bits(7)) :: ("other junk" | uint8)
    def to(header: Header): Struct = {
      val lengthBits = uint(12).encodeValid(header.totalLength)
      val more = !(lengthBits startsWith bin"00000")
      0 :: more :: lengthBits.take(5) :: false :: lengthBits.drop(5) :: header.otherJunk :: HNil
    }
    def from(struct: Struct): Header = struct match {
      case id :: moreFlag :: upperLengthBits :: moreFlag2 :: lowerLengthBits :: otherJunk :: HNil =>
        val length =
          if (moreFlag) uint(12).decodeValidValue(upperLengthBits ++ lowerLengthBits)
          else uint(7).decodeValidValue(lowerLengthBits)
        Header(length, otherJunk)
    }
    struct.xmap[Header](from, to)
  }
}

Note that the upper and lower length bits are encoded as bits(5) and bits(7) and then manually combined in the from function and decoded. The decoding is safe, despite using the unsafe decodeValidValue, because the uint(12) codec is total on 12-bit inputs -- it is impossible for it to return a left.

However, the to function uses encodeValid, which is clearly unsafe -- e.g., Int.MaxValue will cause an exception. We can fix this with a slightly more complicated version, where the to function returns an Err \/ Struct, and we call widen instead of xmap:

object Header {
  implicit val codec: Codec[Header] = {
    type Struct = Int :: Boolean :: BitVector :: Boolean :: BitVector :: Int :: HNil
    val struct: Codec[Struct] = ("identifier" | uint2) :: ("moreFlag" | bool) :: ("upper length bits" | codecs.bits(5)) :: ("moreFlag2" | bool) :: ("lower length bits" | codecs.bits(7)) :: ("other junk" | uint8)
    def to(header: Header): Err \/ Struct = {
      uint(12).encode(header.totalLength) map { lengthBits =>
        val more = !(lengthBits startsWith bin"00000")
        0 :: more :: lengthBits.take(5) :: false :: lengthBits.drop(5) :: header.otherJunk :: HNil
      }
    }
    def from(struct: Struct): Header = struct match {
      case id :: moreFlag :: upperLengthBits :: moreFlag2 :: lowerLengthBits :: otherJunk :: HNil =>
        val length =
          if (moreFlag) uint(12).decodeValidValue(upperLengthBits ++ lowerLengthBits)
          else uint(7).decodeValidValue(lowerLengthBits)
        Header(length, otherJunk)
    }
    struct.widen[Header](from, to)
  }
}

Upvotes: 4

Related Questions