Consider checking the type of a bad row before parsing directly with the right decoder

[benjben]

At the moment we use .as[BadRow] to parse a Json to its bad row case class.
This is based on ADTs encoding and decoding.
Given the number of classes and fields in the bad rows it might be more performant to first check the schema of the bad row (in "schema" field) and then use directly the right Decoder, e.g.

schema match {
  case Schemas.EnrichmentFailures.toSchemaUri =>
    json.as[EnrichmentFailures]
  ...
}

[chuwy]

Yup, that's a good idea. But I believe it is duplicated ticket: #28

[chuwy]

At the same time, I think it would be better to leave Decoder[BadRow] like it is now, because .as[A] decoding implies that structure is compatible with A, while what we really can decode it SelfDescribingData[BadRow]

[benjben]

But I believe it is duplicated ticket

Arf it is indeed, sorry.

because .as[A] decoding implies that structure is compatible with A, while what we really can decode it SelfDescribingData[BadRow]

Good point. So we would need a parse function as you suggested in #28

def parse(json: Json): Either[String, BadRow] =
  for {
    sdj <- SelfDescribingData
      .parse(json)
      .leftMap(e => s"Cannot parse ${json.noSpaces} as self-describing JSON, ${e.code}")
    br <-  sdj.schema match {
      case Schemas.EnrichmentFailures.toSchemaUri ->
        sdj.data.as[EnrichmentFailures].leftMap(_.getMessage)
     case Schemas.SchemaViolations.toSchemaUri ->
        sdj.data.as[SchemaViolations].leftMap(_.getMessage)
     ...
  } yield br

We should also add one to parse directly from a String.