Update (18.11.2015): added spray-json-shapeless library
Update (06.11.15): added circe library
Some time ago I wrote a post on relational database access in Scala since I was looking for a library and there were many of them available, making it hard to make a choice. It turns out that the situation is similar if not worse when it comes to JSON libraries in Scala.
There are just plenty of them. You have no idea. (me neither until I wrote this post)
The following is an attempt to provide a quick overview at how a subset of the libraries I found does a few of the most common things one would likely need to do in regard to JSON:
- parsing it from a raw string
- browsing the AST
- building an AST
- mapping to a case class
There are of course plenty more valid use-cases but this post is not going to cover those.
Let’s get started!
Scala JSON libraries
play-json
The Play Framework comes with a JSON library that covers most of the common use-cases one would encounter when building web applications:
Parsing raw JSON
|
1 2 3 4 5 6 7 8 9 |
scala> import play.api.libs.json._ import play.api.libs.json._ scala> val rawJson = """{"hello": "world", "age": 42}""" rawJson: String = {"hello": "world", "age": 42} scala> Json.parse(rawJson) res0: play.api.libs.json.JsValue = {"hello":"world","age":42} |
Browsing the AST
|
1 2 3 |
scala> (res0 \ "hello").as[String] res1: String = world |
Building a JSON AST
|
1 2 3 |
scala> Json.obj("hello" -> "world", "age" -> 42) res2: play.api.libs.json.JsObject = {"hello":"world","age":42} |
Mapping to a case class
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 |
scala> case class Model(hello: String, age: Int) defined class Model scala> implicit val modelFormat = Json.format[Model] modelFormat: play.api.libs.json.OFormat[Model] = play.api.libs.json.OFormat$$anon$1@81116d scala> Json.fromJson[Model](res0) res3: play.api.libs.json.JsResult[Model] = JsSuccess(Model(world,42),) scala> res3.get res4: Model = Model(world,42) scala> Json.toJson(Model("bar", 23)) res5: play.api.libs.json.JsValue = {"hello":"bar","age":23} |
lift-json
lift-json is the JSON library of the Lift framework. It is one of the oldest one out there if I’m not mistaken.
Parsing raw JSON
|
1 2 3 4 5 6 7 8 9 |
scala> import net.liftweb.json._ import net.liftweb.json._ scala> val rawJson = """{"hello": "world", "age": 42, "nested": { "deeper": { "treasure": true }}}""" rawJson: String = {"hello": "world", "age": 42, "nested": { "deeper": { "treasure": true }}} scala> parse(rawJson) res0: net.liftweb.json.JValue = JObject(List(JField(hello,JString(world)), JField(age,JInt(42)), JField(nested,JObject(List(JField(deeper,JObject(List(JField(treasure,JBool(true)))))))))) |
Browsing the AST
|
1 2 3 |
scala> res0 \ "nested" \ "deeper" \ "treasure" res1: net.liftweb.json.JsonAST.JValue = JBool(true) |
Building a JSON AST
|
1 2 3 4 5 6 |
scala> import net.liftweb.json.JsonDSL._ import net.liftweb.json.JsonDSL._ scala> ("hello" -> "world") ~ ("age" -> 42) res2: net.liftweb.json.JsonAST.JObject = JObject(List(JField(hello,JString(world)), JField(age,JInt(42)))) |
Mapping to a case class
|
1 2 3 4 5 6 7 8 9 10 11 12 |
object LiftJsonExample { def main(args: Array[String]): Unit = { import net.liftweb.json._ implicit val formats = DefaultFormats case class Model(hello: String, age: Int) val rawJson = """{"hello": "world", "age": 42}""" println(parse(rawJson).extract[Model]) } } |
spray-json
spray rols its own JSON library that focuses on working with case classes:
Parsing raw JSON
|
1 2 3 4 5 6 7 8 9 10 11 12 |
scala> import spray.json._ import spray.json._ scala> import DefaultJsonProtocol._ import DefaultJsonProtocol._ scala> val rawJson = """{"hello": "world", "age": 42}""" rawJson: String = {"hello": "world", "age": 42} scala> rawJson.parseJson res0: spray.json.JsValue = {"hello":"world","age":42} |
Browsing the AST
No can do?
Building a JSON AST
No can do?
Mapping to a case class
|
1 2 3 4 5 6 7 8 9 |
scala> case class Model(hello: String, age: Int) defined class Model scala> implicit val modelFormat = jsonFormat2(Model) modelFormat: spray.json.RootJsonFormat[Model] = spray.json.ProductFormatsInstances$$anon$2@7bc880f8 scala> res1.convertTo[Model] res4: Model = Model(world,42) |
spray-json-shapeless
spray-json-shapeless derives spray-json’s JsonFormat-s automatically using shapeless (no need to define an implicit JsonFormat
Mapping to a case class
|
1 2 3 4 5 6 7 8 9 10 11 12 |
scala> import spray.json._ import spray.json._ scala> import fommil.sjs.FamilyFormats._ import fommil.sjs.FamilyFormats._ scala> case class Model(hello: String, age: Int) defined class Model scala> Model("hello", 42).toJson res0: spray.json.JsValue = {"hello":"hello","age":42} |
This is quite useful, it removes the boilerplate formats hanging around
json4s
json4s is a bit like slf4j in the sense that it tries to unite all kind of rogue libraries serving the same purpose by providing a common interface. But not every library uses it, which means that chances are high that your project will contain json4s in addition to another (few) Scala JSON libraries. Hopefully it will one day succeed with its slogan “One AST to rule them all”.
json4s has its roots in lift-json so this will look familiar:
Parsing raw JSON
|
1 2 3 4 5 6 7 8 9 10 11 12 |
scala> import org.json4s._ import org.json4s._ scala> import org.json4s.native.JsonMethods._ import org.json4s.native.JsonMethods._ scala> val rawJson = """{"hello": "world", "age": 42}""" rawJson: String = {"hello": "world", "age": 42} scala> parse(rawJson) res0: org.json4s.JValue = JObject(List((hello,JString(world)), (age,JInt(42)))) |
Browsing the AST
|
1 2 3 |
scala> res0 \ "hello" res1: org.json4s.JValue = JString(world) |
Building a JSON AST tree
|
1 2 3 |
scala> ("hello" -> "world") ~ ("age" -> 42) res2: org.json4s.JsonAST.JObject = JObject(List((hello,JString(world)), (age,JInt(42)))) |
Mapping to a case class
|
1 2 3 4 5 6 7 8 9 10 11 12 13 |
object Json4sExample { def main(args: Array[String]): Unit = { import org.json4s._ import org.json4s.native.JsonMethods._ implicit val formats = DefaultFormats case class Model(hello: String, age: Int) val rawJson = """{"hello": "world", "age": 42}""" println(parse(rawJson).extract[Model]) } } |
argonaut
Argonaut promotes “purely functional JSON in Scala”. It uses scalaz under the hood and
Parsing raw JSON
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 |
import scalaz._, Scalaz._ import scalaz._ import Scalaz._ scala> import argonaut._, Argonaut._ import argonaut._ import Argonaut._ scala> val rawJson = """{"hello": "world", "age": 42, "nested": { "deeper": { "treasure": true }}}""" rawJson: String = {"hello": "world", "age": 42, "nested": { "deeper": { "treasure": true }}} scala> rawJson.parseOption res0: Option[argonaut.Json] = Some({"hello":"world","age":42,"nested":{"deeper":{"treasure":true}}}) |
Browsing the AST
There are several mechanisms available, let’s use a lense. Those are funky:
|
1 2 3 4 5 6 |
scala> val ohMyLens = jObjectPL >=> jsonObjectPL("nested") >=> jObjectPL >=> jsonObjectPL("deeper") >=> jObjectPL >=> jsonObjectPL("treasure") >=> jBoolPL ohMyLens: scalaz.PLensFamily[argonaut.Json,argonaut.Json,Boolean,Boolean] = scalaz.PLensFamilyFunctions$$anon$2@8c894ab scala> ohMyLens.get(res0.get) res1: Option[Boolean] = Some(true) |
Building a JSON AST tree
|
1 2 3 |
scala> ("hello", jString("world")) ->: ("age", jNumber(42)) ->: jEmptyObject res2: argonaut.Json = {"age":42,"hello":"world"} |
Mapping to a case class
|
1 2 3 4 5 6 7 8 9 |
scala> case class Model(hello: String, age: Int) defined class Model scala> implicit def ModelCodecJson: CodecJson[Model] = casecodec2(Model.apply, Model.unapply)("hello", "age") ModelCodecJson: argonaut.CodecJson[Model] scala> rawJson.decodeOption[Model] res3: Option[Model] = Some(Model(world,42)) |
circe
Circe is a fork of Argonaut that uses cats instead of scalaz and uses shapeless to generate codecs.
Parsing raw JSON
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 |
scala> import io.circe._, io.circe.generic.auto._, io.circe.parse._, io.circe.syntax._ import io.circe._ import io.circe.generic.auto._ import io.circe.parse._ import io.circe.syntax._ scala> val rawJson = """{"hello": "world", "age": 42, "nested": { "deeper": { "treasure": true }}}""" rawJson: String = {"hello": "world", "age": 42, "nested": { "deeper": { "treasure": true }}} scala> parse(rawJson).getOrElse(Json.empty) res0: io.circe.Json = { "hello" : "world", "age" : 42, "nested" : { "deeper" : { "treasure" : true } } } |
Browsing the AST
So far there’s only support for cursors which let you move around the JSON tree and do changes if you would like, not for Lenses yet:
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 |
scala> val cursor = res0.cursor cursor: io.circe.Cursor = CJson({ "hello" : "world", "age" : 42, "nested" : { "deeper" : { "treasure" : true } } }) scala> for { | nested <- cursor.downField("nested") | deeper <- nested.downField("deeper") | treasure <- deeper.downField("treasure") | } yield treasure.as[Boolean] res1: Option[io.circe.Decoder.Result[Boolean]] = Some(Right(true)) |
Mapping to a case class hierarchy
I didn’t get this to work with the example below. The example in the documentation works though so here is that one:
|
1 2 3 4 5 6 7 8 9 10 11 12 13 |
scala> sealed trait Foo defined trait Foo scala> case class Bar(xs: List[String]) extends Foo defined class Bar scala> case class Qux(i: Int, d: Option[Double]) extends Foo defined class Qux scala> val foo: Foo = Qux(13, Some(14.0)) foo: Foo = Qux(13,Some(14.0)) scala> foo.asJson.noSpaces res0: String = {"Qux":{"d":14.0,"i":13}} scala> decode[Foo](foo.asJson.spaces4) res1: cats.data.Xor[io.circe.Error,Foo] = Right(Qux(13,Some(14.0))) |
sphere-json
The sphere-json library focuses on providing an easy way to get de/serializers for entire families of case classes. This is really useful when working with any kind of protocol-related system. I am using it in a CQRS system where commands and events are travelling back and forth between the server and a Javascript UI. Instead of having to define a codec for each one of my case classes, I simply have them extend a common abstract class and let the library take care of the rest. Watch for yourselves:
Mapping to a case class hierarchy
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 |
import io.sphere.json.generic._ import io.sphere.json._ object Messages { sealed abstract class Message case class Hello(hello: String) extends Message case class Age(age: Int) extends Message } object SphereJsonExample { import Messages._ val rawHello = """{ "hello": "world", "type": "Hello" }""" val rawAge = """{ "age": 42, "type": "Age" }""" implicit val allYourJson = deriveJSON[Message] def magic(json: String) = fromJSON[Message](json).fold( fail => println("Oh noes: " + fail), hi => println(hi) ) def main(args: Array[String]): Unit = { magic(rawHello) magic(rawAge) } } |
jawn
jawn is a library that focuses on speed. It defines a lightweight AST and is compatible with all kind of other ASTs that we have seen here.
Parsing raw JSON
|
1 2 3 4 5 6 7 8 9 10 11 12 |
scala> import jawn._ import jawn._ scala> import jawn.ast._ import jawn.ast._ scala> val rawJson = """{"hello": "world", "age": 42}""" rawJson: String = {"hello": "world", "age": 42} scala> jawn.ast.JParser.parseFromString(rawJson).get res0: jawn.ast.JValue = {"age":42,"hello":"world"} |
rapture
rapture’s json library is the ultimate Scala JSON library. It doesn’t really do anything with JSON itself, instead, it abstracts over the following JSON libraries (which it calls backends):
- Argonaut
- Jackson
- Jawn
- JSON4S
- Lift
- Play
- Scala standard library JSON
- Spray
Parsing raw JSON
|
1 2 3 4 5 6 7 8 |
scala> import rapture.json._ scala> import rapture.json.jsonBackends.play._ import rapture.json.jsonBackends.play._ scala> Json.parse(rawJson) res2: rapture.json.Json = {"hello":"world","age":42} |
Browsing the AST
Rapture is using Scala’s Dynamic trait, which makes this fun:
|
1 2 3 |
scala> res2.hello.as[String] res3: String = world |
Building a JSON AST tree
|
1 2 3 |
scala> json"""{ "hello": "world", "age": 42}""" res6: rapture.json.Json = {"hello":"world","age":42} |
The Scala standard library
Up until recently I did not know that Scala had a JSON utility in its standard library. But here it is!
Parsing raw JSON
|
1 2 3 4 5 6 7 8 9 10 |
scala> import scala.util.parsing.json._ import scala.util.parsing.json._ scala> val rawJson = """{"hello": "world", "age": 42}""" rawJson: String = {"hello": "world", "age": 42} scala> JSON.parseFull(rawJson) warning: there was one deprecation warning; re-run with -deprecation for details res0: Option[Any] = Some(Map(hello -> world, age -> 42.0)) |
Browsing the “AST”
|
1 2 3 |
scala> res0.get.asInstanceOf[Map[String, Any]]("hello").asInstanceOf[String] res4: String = world |
Even more!
I am forgetting a ton of libraries here I am sure. That, and I am tired and my glass of Uigeadail is getting empty.
So let me mention a few more:
- Julien Richard-Foy built the play-json-variants which add the root hierarchy capability to play-json
- Pascal Voitot added a few functional manipulations to play-json in play-json-zipper
- Robert J. Macomber built the rojoma-json library which I only discovered now and am too tired to cover (sorry)
Great, now which one to pick?
Honestly I don’t have any good advice here. These days I am sticking to play-json and sphere-json, but that’s for no particular reason other than these libraries being there and doing what I need to do (parse and write JSON, traverse and some rare times transform the AST, bind to case classes, make it possible to support custom Java types). If play-json had support for hierarchies out of the box I would probably not have even looked for anything else.
Because for all the joy there seems to be in implementing JSON libraries in Scala, one thing has to be said: JSON de/serialization is boring. It’s this annoying thing that you have to do in order to get your application to talk to another computerized system, period.
I have never met a developer who told me how much pleasure they derived from turning classes into JSON strings and back. That’s just not a thing.
I have, however, met more than one developer that has run into trouble getting library X to cover one of the simple use-cases outlined above. Believe me, there is nothing more frustrating than having to spend time on the annoying task of setting up JSON de/serialization in order to do the boring thing of tossing strings back and forth the network. That’s time you will never get back.
Good night, and good luck.
Liked this post? Subscribe to the mailing list to get regular updates on similar topics.
12 Comments on “A quick tour of JSON libraries in Scala”
Nice analysis – thanks! I wanted to point out that in Argonaut you should also check out CodecJson.derive, eg
implicit val codec = CodecJson.derive[CaseClass]
Thank you so much for this list and the examples. Especially considering the current discussion about adding an AST to the scala stdlib.
Nice set of examples. You might want to also check out my presentation comparing performance of the various Scala Json Libraries.
http://www.slideshare.net/nestorhp/scala-json-features-and-performance
Very nice review
Nice post, wondered what your thoughts were on my post regarding xml processing in scala, the post is a bit old now. http://joncook.github.io/blog/2013/11/03/xml-processing-with-scala/ and on githib https://github.com/JonCook/scala-xml-parsing-example
One other important note regarding spray-json – I believie it is the only library listed which preserves field order in objects.
check out our new lib, also preserves order!
https://github.com/MediaMath/scala-json
Great article! I’ve got a new one for your list:
https://github.com/MediaMath/scala-json
Painfully tried to match the feature set of existing scala and scala-js JSON libraries, and we’ve done it and more!
Cross compiles for 2.10-2.12 X scala/scala-js-0.6
Pingback: Useful play librariesSimplybook Wiki | Simplybook Wiki
Thanks for the great article.
Actually I am having a little problem with play-json 2.4.8 and spark 1.6.1. Because of Spark 1.6.1 uses jackson-databind 2.4.4 and play-json 2.4.8 uses jackson-databind 2.5.4.
I have added this line to build.sbt to force which version to use.
dependencyOverrides ++= Set(
“com.fasterxml.jackson.core” % “jackson-databind” % “2.4.4”
)
It works on my local machine. But when I submit it to EMR runnin Spark 1.6.1, an error occured like below:
java.lang.NoClassDefFoundError: play/api/libs/json/JsValue
Do you have any idea about what’s happening?
Great overview!
A correction – spray-json does support AST navigation (though not in the most pleasant way). Example:
Constructing an AST:
Nice analysis. One use case I am missing: Partial JSON objects. Often for updates a client only sends the changed fields of a JSON structure back. Some libraries support this use case and can transform such a response into a function that updates exactly the given fields in a matching case class. This is sometimes goes under the name incomplete decoders.