Decoding a sealed trait in Argonaut based on JSON structure?

Question

Given the following example:

sealed trait Id

case class NewId(prefix: String, id: String) extends Id
case class RevisedId(prefix: String, id: String, rev: String) extends Id

case class User(key: Id, name: String)

val json = """
{
  "key": {
    "prefix": "user",
    "id": "Rt01",
    "rev": "0-1"
  },
  "name": "Bob Boberson"
}
"""

implicit val CodecUser: CodecJson[User] = casecodec2(User.apply, User.unapply)("key", "name")

implicit val CodecId: CodecJson[Id] = ???

json.decodeOption[User]

I need to write a CodecJson for Id that will decode an object when it has the proper structure.

Adding a discriminator field of some sort is a common suggestion for this, but I don't want to change the JSON I'm already producing/consuming with spray-json and json4s.

In those libraries your encoders/decoders are basically just PartialFunction[JValue, A] and PartialFunction[A, JValue]. If your value isn't defined in the domain it's a failure. It's a really simple, elegant solution I think. In addition to that you've got Extractors for the JSON types, so it's easy to match an object on fields/structure.

Rapture goes a step further, making field order unimportant and ignoring the presence of non-matching fields, so you could just do something like:

case json"""{ "prefix": $prefix, "id": $id, "rev": $rev }""" => 
  RevisedId(prefix, id, rev)

That's really simple/powerful.

I'm having trouble figuring out how to do something similar with argonaut. This is the best I've come up with so far:

val CodecNewId = casecodec2(NewId.apply, NewId.unapply)("prefix", "id")
val CodecRevisedId = casecodec3(RevisedId.apply, RevisedId.unapply)("prefix", "id", "rev")

implicit val CodecId: CodecJson[Id] =
  CodecJson.derived[Id](
    EncodeJson {
      case id: NewId => CodecNewId(id)
      case id: IdWithRev => RevisedId(id)
    },
    DecodeJson[Id](c => {
      val q = RevisedId(c).map(a => a: Id)
      q.result.fold(_ => CodecNewId(c).map(a => a: Id), _ => q)
    })
  )

So there's a few problems with that. I have to define extra codecs I don't intend to use. Instead of using the case-class extractors in the EncodeJson for CodecJson[Id], I'm delegating to the other encoders I've defined. Just just doesn't feel very straight-forward or clean for classes that only have 2 or 3 fields.

The code for the DecodeJson section is also pretty messy. Aside from an extra type-cast in the ifEmpty side of the fold it's identical to the code in DecodeJson.|||.

Does anyone have a more idiomatic way to write a basic codec for Sum-types in argonaut that doesn't require a discriminator and can instead match on the structure of the json?

Answer 1

This is the best I've been able to come up with. It doesn't have that fundamental feel of elegance that the partial functions do, but it is a good bit more terse and easier to decipher than my first attempt.

val CodecNewId = casecodec2(NewId.apply, NewId.unapply)("prefix", "id")
val CodecRevisedId = casecodec3(RevisedId.apply, RevisedId.unapply)("prefix", "id", "rev")

implicit val CodecId: CodecJson[Id] = CodecJson(
  {
    case id: NewId => CodecNewId(id)
    case id: RevisedId => CodecRevisedId(id)
  },
   (CodecRevisedId ||| CodecNewId.map(a => a: Id))(_))

We're still using the "concrete" codecs for each sub-type. But we've gotten away from the CodecJson.derive call, we don't need to wrap our encode function in an EncodeJson, and we can map our DecodeJson function instead of type-casting so we can go back to using ||| instead of copying it's implementation, which makes the code a lot more readable.

This is definitely a usable solution, if not quite what I'd hoped for.