A recent project needed a web interface to display Scala case classes populated by scodec. The UI is simply a developer level dump of the data, so there are no specific constraints placed on the Json needed to populate the pages.
The domain model is fairly complex and needs to support the non trivial codecs. The design makes extensive use of package objects, traits and case objects. The last two require a type indicator in the Json.
Our Java projects made extensive use of xstream to implement similar non-Java specific serialization. Xstream “just works”, at the cost of using reflection.
The first attempt used upickle. That was initially successful, and correctly handled the traits and the objects. However, I quickly ran into a known bug which originates from a long standing scalac bug
The next attempt tried json4s (nee Lift Json). That code base is fairly old (and evidently no longer actively developed). Some custom code was required to handle the package object and case object class names.
This attempt also failed when the compiler was unable to resolve all the implicits.
The final attempt applied the Play Json functionality. That did require manual effort to define the Writers, but was ultimately successful. That implementation deserves its own post.
The above experience unfortunately implies that the “code generation” tools of Scala (macros, implicits, etc) exceed the current capabilities of scalac for non trivial codebases.