Use typeclasses instead of KNumber/KString/KairosCompatibleType

[thomastoye]

Currently, supported KairosDB types are represented by classes that extend the sealed trait KairosCompatibleType, with implementations KNumber and KString. We might use a typeclass instead so users can specify their own types (for when they have custom data types).

Here's one idea:

trait KairosCompatibleTypeTypeClass[A] extends Format[A] {
  /** The KairosDB name for the type */
  val typeName: String
}

object KairosString extends KairosCompatibleTypeTypeClass[String] {
  override val typeName: String = "string"
  override def writes(o: String): JsValue = JsString(o)
  override def reads(json: JsValue): JsResult[String] = json.validate[String]
}

object KairosBigDecimal extends KairosCompatibleTypeTypeClass[BigDecimal] {
  override val typeName: String = "number"
  override def writes(o: BigDecimal): JsValue = JsNumber(o)
  override def reads(json: JsValue): JsResult[BigDecimal] = json.validate[BigDecimal]
}

case class DataPointWithTypeclassValue[A](metricName: MetricName, value: A, timestamp: Instant, tags: Seq[Tag], ttl: Option[FiniteDuration] = None) extends DataPoint

  def addDataPointsWithTypeClass[A](dataPoints: Seq[DataPointWithTypeclassValue[A]])(implicit evidence: KairosCompatibleTypeTypeClass[A]): Future[Unit] = {
    val body = Json.toJson(dataPoints)

    wsClient
      .url(uri / "api" / "v1" / "datapoints")
      .applyKairosDBAuth
      .post(body)
      .map { res =>
        res.status match {
          case 204 => ()
          case 400 => throw KairosDBResponseBadRequestException(errors = getErrors(res))
          case 500 => throw KairosDBResponseInternalServerErrorException(errors = getErrors(res))
        }
      }
  }

And then using it:

addDataPointsWithTypeClass[String](Seq(DataPointWithTypeclassValue(MetricName("aoeu"), "aoue", Instant.now, Seq())))(KairosString)

I'm not sure, but I believe that this will make it impossible to mix datapoints with values of different types (e.g. a datapoint with a number and a datapoint with a string). Perhaps using some Scala type magic?