Facade for serialization of supported ASTs?

[rossabaker]

Escalating Twitter discussion discussion beyond 140 characters:

I created jawn-streamz for parsing scalaz-streams of ByteVectors to any Jawn-supported JSON AST. I think it could make a good Jawn module, but I also have the use case of serializing JSON back to ByteVectors. It would be splendid if this could also be done through a facade, and keep streamz support shielded from the Great JSON Framework Holy War.

Open questions:

1. Would serialization be in scope for Jawn, or is that bloat?
2. If in scope, what types are targeted for serialization? Parser supports String, File, Channel, ByteBuffer, and scalaz.stream.Process. Symmetry is neat.
3. What additions to the facade would be necessary to support those targets?
4. Should a serialization facade offer options for pretty-printing, compact, etc.?

[casualjim]

I'm going to try and put this parser into muster, which in turn will be used in json4s. https://github.com/json4s/muster

[non]

I am totally open to serialization from "raw" ASTs (I don't want to handle any object-mapping in Jawn). So it's definitely in-scope if we think we can do it well.

I think symmetry around targets makes sense, with the caveat that I am not sure if there is a canonical byte vector implementation?

As far as facade additions, I feel like we'd probably need two layers (sort of like how we have Parser[J] but also Facade[J]). I don't have a totally clear picture but I don't think it would necessarily be that hard, with that caveat that using "the most efficient" method of iteration over collections might necessitate more verbose definition.

What do you think? I guess my biggest question for you is about byte vectors.

[bryce-anderson]

The scodec ByteVector is gaining traction. Its used in a few typelevel projects right now, and we are using it in http4s and loving it. I suppose that doesn't make it canonical, but its the best thing around that I'm aware of.

[non]

Ah, OK, I'll definitely take a look. Thanks!

[rossabaker]

ByteVector was the obvious choice for jawn-streamz, because scalaz-stream already uses it extensively. A reactive-stream of Akka's ByteStrings might appeal to a different audience. I like that jawn-parser's lone dependency is scala-library, with more exotic targets integrated via modules. That same delineation makes sense to me for serialization.

And +1 to avoiding object mapping in Jawn.

[non]

We might choose to provide something simpler that could be mapped to ByteVector. For example, List[ByteBuffer] would not be a bad choice (since we would not be doing any truncation, but only appending). I'd probably implement that using ListBuffer and then converting to List.

We could also provide Stream[ByteBuffer] (or similar) for rendering-as-you-go.

[casualjim]

I understand it's not the goal of jawn, but you can now also do object mapping. https://github.com/json4s/muster/blob/jawn/codecs/jawn/src/main/scala/muster/codec/jawn/jawn_input.scala

[non]

Awesome! Eventually I'll add a section to the README linking to muster.

[lihaoyi]

I'd like this too =) I just C&Ped the bulk of the jawn serializer in upickle, just because I didn't want to pull in the bulk of Spire just to serialize some ASTs

[rossabaker]

Note: muster is now unmaintained, but it does implement this rendering for the ASTs: https://github.com/json4s/muster/blob/4816c14045e9d9609959e453636d6324b7dbf720/core/src/main/scala/muster/output/output.scala

If you're still willing to consider scope creep from abstract parser to abstract codec, I can look into scavenging that into a pull request. That would let me finish the other half of the jawn-streamz project.

[non]

I have something in-progress working on this. I'll look at it a bit more and comment here.

[lihaoyi]

For the record I ended up forking Jawn into uJson, which has support for serializing every AST supported by jawn via the Facade interface (renamed to Visitor)