Scodec组合器：标头包含用于区分类型的幻数

Question

I am looking for a way to approach a protocol like the following example:

case class Request(bodyType: Int, foo: Int, bar: Int, body: RequestBody)

sealed trait RequestBody
case class Read(key: String) extends RequestBody
case class Write(key: String, value: Array[Byte]) extends RequestBody 

Here, bodyType == 0 will stand for Read , and bodyType != 0 will encode Write . Note that there are a few fields separating discriminator from discriminated value.

I've seen an example with byte-ordering . But as far as I understand this "squid" encoded discriminator would not round trip. What's the correct way to solve such a problem?

Answer 1

There are a few ways to do it, but this is one I've used:

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

case class Request(bodyType: Int, foo: Int, bar: Int, body: RequestBody)

sealed trait RequestBody
case class Read(key: String) extends RequestBody
object Read {
  implicit val codec: Codec[Read] = ("key" | utf8).as[Read]
  implicit val discriminator: Discriminator[RequestBody, Read, Int] = Discriminator(0)
}
case class Write(key: String, value: ByteVector) extends RequestBody
object Write {
  implicit val codec: Codec[Write] = {
    ("key"   | utf8  ) ::
    ("value" | bytes )
  }.as[Write]
  implicit val discriminator: Discriminator[RequestBody, Write, Int] = Discriminator(1)
}

object Request {
  implicit val codec: Codec[Request] = {
    ("bodyType" | uint16 ).flatPrepend { bodyType =>
    ("foo"      | uint16 ) ::
    ("bar"      | uint16 ) ::
    ("body"     | Codec.coproduct[RequestBody].discriminatedBy(provide(bodyType)).auto)
  }}.as[Request]
}