Hi,
could you maybe write TypeInformation/TypeSerializer wrappers that lazily
instantiate a type class-based serializer. It might even work using a "lazy
val". Something like this:
class ScalaTypeSerializer[T] extends TypeSerializer[T] {
lazy val serializer = "create the scala serializer"
...
def serialize(value: T, out: DataOutputView): () = {
serializer.serialize(value, out) // not sure how the generated
serializers would be used, just a placeholder
}
}
Cheers,
Aljoscha
On Thu, 23 Jun 2016 at 16:23 Stefano Baghino <stefano.bagh...@radicalbit.io>
wrote:
> Hello everybody,
>
> in the past days, I've written batch input/output formats for MongoDB.
>
> Initially, I've tried to use the non-blocking ReactiveMongo
> <http://reactivemongo.org/> driver, which uses the type class pattern in
> Scala for the serialization logic. The library also exposes some pretty
> neat macros that automatically generate the type class instances for you,
> given a case class.
>
> Now, the problem is that these macros are not useful in Flink because the
> generated type class instances would have to be serializable (something
> that has been understandably left out from the macros). Has anyone ever
> faced a similar problem? I've encountered it again when using upickle
> <http://www.lihaoyi.com/upickle-pprint/upickle/>, which has a similar
> facility but for JSON serialization.
>
> In the end I've resorted to writing my own serialization logic and
> explicitly extending java.util.Serializable in the end but I feel there may
> be a way to not do this (without rewriting/extending the macros to make the
> generated classes serializable).
>
> --
> BR,
> Stefano Baghino
>
> Software Engineer @ Radicalbit
>
