rallyhealth / weepickle

A fork of uPickle, shaded for backwards and forwards compatibility.

Version Matrix

weePickle version

A stable JSON, YAML, MsgPack, XML, etc. serialization framework based on uPickle.

We're tired of dependency hell!

weePickle exists to fulfill two promises:

  1. We will not break compatibility in v1.x.y, as enforced by MiMa.
  2. When we release v2.0.0, you can use it immediately without waiting for your other library dependencies to update. We achieve this by shading.

Shading

Both weepickle-v1.jar and weepickle-v2.jar (in the future) will coexist on the classpath peacefully by applying shading at multiple levels.

  1. All artifact names are suffixed with the major version number (e.g. -v1), which prevents evictions.
  2. All packages are prefixed with the major version number (e.g. com.rallyhealth.v1), which prevents classpath conflicts.

Shading allows libraries to depend directly on weePickle-v1 without fear of causing incompatible evictions and runtime failures.

For more background, see shading.md.

Features

weePickle combines some of the best parts of the serialization ecosystem.

  • Zero-overhead conversion of uPickle
  • jackson-core: async parsing and broad format support (JSON, YAML, XML, CBOR, SMILE, Ion, etc.)
  • case class support (any number of fields) through customizable macros
  • scala json AST interop (circe, json4s, play-json, argonaut)
  • Fast serialization to/from MessagePack

sbt

resolvers += "Rally Health" at "https://dl.bintray.com/rallyhealth/maven"
libraryDependencies += "com.rallyhealth" %% "weepickle-v1" % "version"

version: Download

Getting Started

JSON to Scala:

FromJson("[1,2,3]").transform(ToScala[List[Int]])    ==> List(1, 2, 3)

Scala to JSON:

FromScala(List(1, 2, 3)).transform(ToJson.string)    ==> "[1,2,3]"

JSON to pretty JSON:

FromJson("[1,2,3]").transform(ToPrettyJson.string)   ==>
[
    1,
    2,
    3
]

Files & YAML:

sbt

resolvers += "Rally Health" at "https://dl.bintray.com/rallyhealth/maven"
libraryDependencies ++= Seq(
  "com.rallyhealth" %% "weepickle-v1" % "version",
  "com.rallyhealth" %% "weeyaml-v1" % "version"
)
import com.rallyhealth.weejson.v1.yaml.{FromYaml, ToYaml}
import com.rallyhealth.weejson.v1.jackson.FromJson
import java.nio.file.Files
import java.nio.file.Paths

val jsonFile = Files.newInputStream(Paths.get("file.json"))
val yamlFile = Files.newOutputStream(Paths.get("file.yml"))

FromJson(jsonFile).transform(ToYaml.outputStream(yamlFile))

Case Classes:

import com.rallyhealth.weepickle.v1.WeePickle
case class Foo(i: Int)

object Foo {
  implicit val rw = WeePickle.macroFromTo[Foo]
}

FromScala(Foo(1)).transform(ToJson.string)           ==> """{"i":1}"""
FromScala(Foo(1)).transform(ToXml.string)            ==> """<root><i>1</i></root>"""
FromJson("""{"i":1}""").transform(ToScala[Foo])      ==> Foo(1)

Pick Any Two

You can convert directly between any From/To types. See Zero-Overhead Tree Processing with the Visitor Pattern for how this works.

The following is a non-exhaustive map of type support: From and To convertable types

Supported Types

  • Boolean, Byte, Char, Short, Int, Long, Float, Double
  • Tuples from 1 to 22
  • Immutable Seq, List, Vector, Set, SortedSet, Option, Array, Maps, and all other collections with a reasonable CanBuildFrom implementation
  • Duration, Either,
  • Date, Instant, LocalDate, LocalTime, LocalDateTime, OffsetDateTime, ZonedDateTime
  • Stand-alone case classes and case objects, and their generic equivalents,
  • Non-generic case classes and case objects that are part of a sealed trait or sealed class hierarchy
  • sealed trait and sealed classes themselves, assuming that all subclasses are picklable
  • UUIDs
  • null

Readability/writability is recursive: a container such as a Tuple or case class is only readable if all its contents are readable, and only writable if all its contents are writable. That means that you cannot serialize a List[Any], since weePickle doesn't provide a generic way of serializing Any.

Case classes are serialized using the apply and unapply methods on their companion objects. This means that you can make your own classes serializable by giving them companions apply and unapply. sealed hierarchies are serialized as tagged unions: whatever the serialization of the actual object, together with the fully-qualified name of its class, so the correct class in the sealed hierarchy can be reconstituted later.

Anything else is not supported by default, but you can add support using Custom Picklers.

Defaults

If a field is missing upon deserialization, weePickle uses the default value if one exists.

case class Dflt(i: Int = 42)

FromJson("""{}""").transform(ToScala[Dflt])          ==> Dflt(42)
FromJson("""{"i": 999}""").transform(ToScala[Dflt])  ==> Dflt(999)

If a field at serialization time has the same value as the default, it will be written unless annotated with @dropDefault.

FromScala(Dflt(42)).transform(ToJson.string)         ==> """{"i": 42}"""
case class Dflt2(@dropDefault i: Int = 42)
FromScala(Dflt2(42)).transform(ToJson.string)        ==> """{}"""

If a class is annotated with @dropDefault, all fields with default values will not be written.

@dropDefault case class Dflt3(i: Int = 42, j: Int = 43, k: Int = 45)
FromScala(Dflt2(42, 43, 0)).transform(ToJson.string)        ==> """{"k": 0}"""

Options

Option[T] is unwrapped when the Option is Some (rationale):

case class Maybe1(i: Option[Int])
object Maybe1 {
  implicit val rw = WeePickle.macroFromTo[Maybe1]
}

FromScala(Maybe1(Some(42))).transform(ToJson.string) ==> """{"i":42}"""
FromJson("""{"i":42}""").transform(ToScala[Maybe1])  ==> Maybe1(Some(42))

None is translated as null (rationale):

FromScala(Maybe1(None)).transform(ToJson.string) ==> """{"i":null}"""
FromJson("""{"i":null}""").transform(ToScala[Maybe1]) ==> Maybe1(None)
FromJson("""{}""").transform(ToScala[Maybe1]) ==> Maybe1(None)

If you want to suppress the field entirely on None, you can use Defaults.

case class Maybe2(@dropDefault i: Option[Int] = None)

FromScala(Maybe2(None)).transform(ToJson.string)     ==> """{}"""

But Option types are a special case where None is an assumed default if a default is not provided explicitly. So putting @dropDefault at the class level will apply to all Option types in the class, whether a default is provided explicitly or not.

@dropDefault case class Maybe3(i: Option[Int], j: Option[Int], k: Option[Int] = Some(0))

FromScala(Maybe3(None, None, Some(0))).transform(ToJson.string)     ==> """{}"""

Custom Keys

weePickle allows you to specify the key with which a field is serialized via a @key annotation.

case class KeyBar(@key("hehehe") kekeke: Int)
object KeyBar{
  implicit val rw = WeePickle.macroFromTo[KeyBar]
}

FromScala(KeyBar(10)).transform(ToJson.string)             ==> """{"hehehe":10}"""
FromJson("""{"hehehe": 10}""").transform(ToScala[KeyBar])  ==> KeyBar(10)

Sealed Hierarchies

Sealed hierarchies are serialized as tagged values, the serialized object tagged with the full name of the instance's class:

sealed trait Outcome
case class Success(value: Int) extends Outcome
case class DeferredVictory(excuses: Seq[String]) extends Outcome

object Success {
  implicit val rw = WeePickle.macroFromTo[Success]
}
object DeferredVictory {
  implicit val rw = WeePickle.macroFromTo[DeferredVictory]
}
// order matters: the trait's companion object must come at the end for implicit resolution to work
object Outcome {
  implicit val rw = WeePickle.macroFromTo[Outcome]
}

FromScala(DeferredVictory(Seq("My json AST is too slow."))).transform(ToJson.string))  ==>
  """{"$type":"com.example.DeferredVictory","excuses":["My json AST is too slow."]}"""

// You can read tagged value without knowing its
// type in advance, just use type of the sealed trait
FromJson("""{"$type":"com.example.Success","value":42}""").transform(ToScala[Outcome]) ==> Success(42)

You can customize the "$type" key and values with annotations:

@discriminator("flavor")
sealed trait Outcome

@key("s")
case class Success(value: Int) extends Outcome

@key("dv")
case class DeferredVictory(excuses: Seq[String]) extends Outcome

FromScala(Success(42)).transform(ToJson.string)      ==> """{"flavor":"s",value:42}""" 

Enumerations

object Suit extends Enumeration {
  val Spades = Value("Spades")
  val Hearts = Value("Hearts")
  val Diamonds = Value("Diamonds")
  val Clubs = Value("Clubs")

  implicit val pickler = WeePickle.fromToEnumerationName(this)
}

FromScala(Suit.Spades).transform(ToJson.string)         ==> """"Spades""""
FromJson(""""Spades"""").transform(ToScala[Suit.Value]) ==> Suit.Spades

jackson-core

weePickle leans heavily on jackson-core for interop with JSON, YAML, and most other formats. Jackson-databind is not used.

Motivations

  1. jackson-core's JSON support is mature, widely used, and heavily optimized.
  2. The ecosystem of possible formats is huge: https://github.com/FasterXML/jackson#active-jackson-projects
  3. jackson-core has a solid track record of backward compatibility.

Buffer pooling

Internally, jackson-core uses buffer pooling to achieve some of its performance. Buffers return to the pool after calling close() on the underlying Parser/Generator. If this doesn't happen, new buffers get allocated for each message, and performance suffers slightly.

FromJson doesn't trust you and calls close() automatically after writing a single json text, which covers the vast majority of use cases. If you're working with multiple json texts separated by whitespace, jackson can handle it, but you have to drop down below the high level API and remember to close the parser/generator yourself.

Value AST

WeePickle includes its own AST named Value, largely unchanged from the upstream uJson.

val obj = Obj(
  "foo" -> Arr(
    42,
    "omg",
    true
  )
)

obj("foo")(0).num                  ==> 42

obj.toString                       ==> """{"foo":[42,"omg",true]}"""
obj.transform(ToPrettyJson.string) ==>
  """{
    "foo": [
      42,
      "omg",
      true
    ]
  }"""

FromJson("""{"foo":[42,"omg",true]}""").transform(Value) ==> obj

See:

MessagePack

weePack is weePickle's MessagePack implementation, largely unchanged from the upstream uPack.

sbt

resolvers += "Rally Health" at "https://dl.bintray.com/rallyhealth/maven"
libraryDependencies += "com.rallyhealth" %% "weepack-v1" % "version"

version: Download

Benchmarks

FromMsgPack/ToMsgPack perform exceptionally well under benchmarks, yielding higher throughput than JSON or the official jackson-dataformat-msgpack.

ParserBench

java 11:

Benchmark                    Mode  Cnt    Score    Error  Units
ParserBench.jsonBytes       thrpt   15  245.665 ±  3.202  ops/s
ParserBench.jsonString      thrpt   15  213.312 ±  5.250  ops/s
ParserBench.msgpackJackson  thrpt   15  205.738 ±  2.789  ops/s
ParserBench.msgpackScala    thrpt   15  422.313 ± 17.172  ops/s
ParserBench.smile           thrpt   15  271.947 ±  1.116  ops/s

GeneratorBench

java 11:

Benchmark                       Mode  Cnt    Score    Error  Units
GeneratorBench.jsonBytes       thrpt   15  238.335 ± 11.777  ops/s
GeneratorBench.jsonString      thrpt   15  240.125 ±  7.871  ops/s
GeneratorBench.msgpackJackson  thrpt   15  181.195 ±  5.774  ops/s
GeneratorBench.msgpackScala    thrpt   15  304.540 ±  2.225  ops/s
GeneratorBench.smile           thrpt   15  306.462 ±  3.134  ops/s

Limitations

  • ScalaJS is not supported (jackson-core is java-only)
  • Same macro limitations as uPickle
  • XML support is still rudimentary and contributions are welcome.

Developing

See developing.md for building, testing, and IDE support.

Upstream

uPickle: a simple Scala JSON and Binary (MessagePack) serialization library

If you use uPickle/weePickle and like it, please support it by donating to lihaoyi's Patreon: