Derived `CirceEnum` codecs fail roundtrip on `case class` values

[VKFisher]

Tested on 1.7.3, scala 2.13.10

Given this type definition

sealed trait SomeType extends EnumEntry

object SomeType extends Enum[SomeType] with CirceEnum[SomeType] {
  val values: IndexedSeq[SomeType] = findValues

  case class SomeValue(value: Int) extends SomeType
}

Running this example

private val example = {
  val initialValue      = SomeType.SomeValue(value = 5)
  val encodedJson       = SomeType.circeEncoder(initialValue)
  val encodedJsonString = encodedJson.spaces2
  val decodedJson       = parser.parse(encodedJsonString).toOption.get
  val decodedJsonString = decodedJson.spaces2
  val decodingResult    = SomeType.circeDecoder.decodeJson(decodedJson)

  println("=== initialValue ===")
  println(initialValue)
  println("=== encodedJsonString ===")
  println(encodedJsonString)
  println("=== decodedJsonString ===")
  println(decodedJsonString)
  println("=== decodingResult ===")
  println(decodingResult)
}

produces this output

=== initialValue ===
SomeValue(5)
=== encodedJsonString ===
"SomeValue(5)"
=== decodedJsonString ===
"SomeValue(5)"
=== decodingResult ===
Left(DecodingFailure('SomeValue(5)' is not a member of enum example.SomeType$@4d910fd6, List()))

[lloydmeta]

1. Depending on where you get the definition of an enum, you might get different answers, but ultimately, they are either defined as "a closed/finite set of values" or set of constant values, etc. With compiled static langs, the additional implication that this set is defined at compile time.

   An instance of a case class created at compile time would not be part of the enum set. Enumeratum (deliberately) ignores any inherited class that is not an object

2. Circe encoding-to-JSON for a type with an encoder is non-fallible, hence the ability to encode non-members of an enum type

I would need a lot of convincing to budge from (1) (ie. SomeValue case class will remain not an enum member for the foreseeable future), and I don't really know how we can simply solve for (2), either, since it's a Circe thing (and I would likewise need heavy convincing to code something in the encoder that checks for members throws)

[VKFisher]

Huh, I see. I was mistaken about enumeratum's scope then.

In any case, I think that codecs that fail a roundtrip at runtime on some subset of possible values is an issue that needs some sort of resolution. What do you think about these?

• Updating the documentation with an explicit statement that enumeratum enums are just simple sets of constant values and not full-fledged ADTs (some languages like Rust or Swift name their ADTs Enums so this is a possible source of confusion)
• Emitting a warning or an error at compile time if there's a non-object value type that enumeratum or circe won't be able to handle

[lloydmeta]

Updating the documentation with an explicit statement that enumeratum enums are just simple sets of constant values and not full-fledged ADTs (some languages like Rust or Swift name their ADTs Enums so this is a possible source of confusion)

While I can appreciate both Rust and the potential confusion, I think it's sufficiently clear that enums != ADTs when it comes to the Scala context and enumeratum: we have a values: ImmutableSeq[EnumType] as a requirement/main feature, and ADTs are implemented via sealed. That Rust and Swift decided to use enum as a keyword for ADTs/tagged unions (all enums are tagged unions but not all tagged unions are enums) is "interesting", but I don't feel that should require all other enumeration implementations to explicitly call out that they are , in fact, implementing enumerable enumerations.

Emitting a warning or an error at compile time if there's a non-object value type that enumeratum or circe won't be able to handle

This would break a lot of usages of intermediate classes or case classes that are used to construct valid object members.

IMO it would be more sensible to update the findValue macros to emit a warning that there were no members found (though this should probably be a minor version bump).

[VKFisher]

@lloydmeta fair enough. Closing this issue since no changes are warranted. Thanks for your responses!