ScalaのJSONライブラリいくつかベンチマーク

case classに対するcodecを自動で定義してくれる系のJSONライブラリをいくつか試してみました

• argonaut-shapeless
• circe
• JSON4S
• spray-json-shapeless

コード: https://github.com/ocadaruma/benchmark-json-libs

それぞれの簡単な使い方

case class Foo(a: String, b: Double, c: Boolean)
val foo = Foo("foo", 42.0, true)

argonaut-shapeless

import argonaut.Argonaut._
import argonaut.Shapeless._

foo.asJson.nospaces // => {"c":true,"b":42.0,"a":"foo"}
"""{"c":true,"b":42.0,"a":"foo"}""".decode[Foo].toOption // => Some(Foo(foo,42.0,true))

circe

import io.circe.parser._
import io.circe.syntax._
import io.circe.generic.auto._

foo.asJson.noSpaces // => {"a":"foo","b":42.0,"c":true}
decode[Foo]("""{"a":"foo","b":42.0,"c":true}""").toOption // => Some(Foo(foo,42.0,true))

JSON4S

import org.json4s._
import org.json4s.jackson.{Serialization, JsonMethods}

implicit val formats = DefaultFormats

Serialization.write(foo) // => {"a":"foo","b":42.0,"c":true}
JsonMethods.parse("""{"a":"foo","b":42.0,"c":true}""").extract[Foo] // => Foo(foo,42.0,true)

spray-json-shapeless

import spray.json._
import fommil.sjs.FamilyFormats._

foo.toJson.compactPrint // => {"a":"foo","b":42.0,"c":true}
"""{"a":"foo","b":42.0,"c":true}""".parseJson.convertTo[Foo] // => Foo(foo,42.0,true)

ベンチマーク結果

sbt-jmh使いました。

[info] Benchmark                                         Mode  Cnt   Score   Error  Units
[info] BenchmarkJsonLibs.argonautShapelessDeserialize   thrpt   10  41.063 ± 1.909  ops/s
[info] BenchmarkJsonLibs.argonautShapelessSerialize     thrpt   10  37.530 ± 0.926  ops/s
[info] BenchmarkJsonLibs.circeDeserialize               thrpt   10  52.426 ± 4.896  ops/s
[info] BenchmarkJsonLibs.circeSerialize                 thrpt   10  46.619 ± 1.056  ops/s
[info] BenchmarkJsonLibs.json4sJacksonDeserialize       thrpt   10  50.363 ± 1.359  ops/s
[info] BenchmarkJsonLibs.json4sJacksonSerialize         thrpt   10  29.081 ± 0.509  ops/s
[info] BenchmarkJsonLibs.json4sNativeDeserialize        thrpt   10  37.430 ± 1.234  ops/s
[info] BenchmarkJsonLibs.json4sNativeSerialize          thrpt   10  28.723 ± 0.934  ops/s
[info] BenchmarkJsonLibs.sprayJsonShapelessDeserialize  thrpt   10  70.349 ± 2.470  ops/s
[info] BenchmarkJsonLibs.sprayJsonShapelessSerialize    thrpt   10  29.165 ± 0.468  ops/s

Serializeはcirceがもっとも速く、Deserializeではspray-json-shapelessが一番速いという結果になりました。

今度Parserを変えていろいろ試してみよう