Encoding Scala None to JSON value using circe

Circe have added a method dropNullValues on Json that uses what Travis Brown mentioned above.

def dropNulls[A](encoder: Encoder[A]): Encoder[A] =
    encoder.mapJson(_.dropNullValues)

implicit val entityEncoder: Encoder[Entity] = dropNulls(deriveEncoder)

The best way to do this is probably just to add a post-processing step to a semi-automatically derived encoder for B:

import io.circe.{ Decoder, JsonObject, ObjectEncoder }
import io.circe.generic.JsonCodec
import io.circe.generic.semiauto.{ deriveDecoder, deriveEncoder }

@JsonCodec
case class A(a1: String, a2: Option[String])
case class B(b1: Option[A], b2: Option[A], b3: Int)

object B {
  implicit val decodeB: Decoder[B] = deriveDecoder[B]
  implicit val encodeB: ObjectEncoder[B] = deriveEncoder[B].mapJsonObject(
    _.filter {
      case ("b1", value) => !value.isNull
      case _ => true
    }
  )
}

And then:

scala> import io.circe.syntax._
import io.circe.syntax._

scala> B(None, None, 1).asJson.noSpaces
res0: String = {"b2":null,"b3":1}

You can adjust the argument to the filter to remove whichever null-valued fields you want from the JSON object (here I'm just removing b1 in B).

It's worth noting that currently you can't combine the @JsonCodec annotation and an explicitly defined instance in the companion object. This isn't an inherent limitation of the annotation—we could check the companion object for "overriding" instances during the macro expansion, but doing so would make the implementation substantially more complicated (right now it's quite simple). The workaround is pretty simple (just use deriveDecoder explicitly), but of course we'd be happy to consider an issue requesting support for mixing and matching @JsonCodec and explicit instances.