jawn: A Scala repository from julien-lafont

Jawn

"Jawn is for parsing jay-sawn."

Origin

The term "jawn" comes from the Philadelphia area. It conveys about as much information as "thing" does. I chose the name because I had moved to Montreal so I was remembering Philly fondly. Also, there isn't a better way to describe objects encoded in JSON than "things". Finally, we get a catchy slogan.

Jawn was designed to parse JSON into an AST as quickly as possible.

Overview

Jawn consists of three parts:

A fast, generic JSON parser
A small, somewhat anemic AST
Support packages which parse to third-party ASTs

Currently Jawn is competitive with the fastest Java JSON libraries (GSON and Jackson) and in the author's benchmarks it often wins. It seems to be faster than any other Scala parser that exists (as of July 2014).

Given the plethora of really nice JSON libraries for Scala, the expectation is that you are here for (1) and (3) not (2).

Quick Start

Jawn supports Scala 2.10 and 2.11. Here's a build.sbt snippet that shows you how to depend on Jawn for your project:

resolvers += Resolver.sonatypeRepo("releases")

// use this if you just want jawn's parser, and will implement your own facade
libraryDependencies += "org.spire-math" %% "jawn-parser" % "0.8.3"

// use this if you want jawn's parser and also jawn's ast
libraryDependencies += "org.spire-math" %% "jawn-ast" % "0.8.3"

If you want to use Jawn's parser with another project's AST, see the "Supporting external ASTs with Jawn" section. For example, with Spray you would say:

libraryDependencies += "org.spire-math" %% "jawn-spray" % "0.8.3"

There are a few reasons you might want to do this:

The library's built-in parser is significantly slower than Jawn
Jawn supports more input types (ByteBuffer, File, etc.)
You need asynchronous JSON parsing

(NOTE: previous to version 0.8.3 the support libraries would have been named "spray-support" instead of "jawn-spray".)

Dependencies

jawn-parser has no dependencies other than Scala.

jawn-ast depends on jawn-parser but nothing else.

The various support projects (e.g. jawn-argonaut) depend on the library they are supporting.

Parsing

Jawn's parser is both fast and relatively featureful. Assuming you want to get back an AST of type J and you have a Facade[J] defined, you can use the following parse signatures:

Parser.parseUnsafe[J](String) → J
Parser.parseFromString[J](String) → Try[J]
Parser.parsefromPath[J](String) → Try[J]
Parser.parseFromFile[J](File) → Try[J]
Parser.parseFromChannel[J](ReadableByteChannel) → Try[J]
Parser.parseFromByteBuffer[J](ByteBuffer) → Try[J]

Jawn also supports asynchronous parsing, which allows users to feed the parser with data as it is available. There are three modes:

SingleValue waits to return a single J value once parsing is done.
UnwrapArray if the top-level element is an array, return values as they become available.
ValueStream parser one-or-more json values separated by whitespace

Here's an example:

import jawn.ast
import jawn.AsyncParser
import jawn.ParseException

val p = ast.JParser.async(mode = AsyncParser.UnwrapArray)

def chunks: Stream[String] = ???
def sink(j: ast.JValue): Unit = ???

def loop(st: Stream[String]): Either[ParseException, Unit] =
  st match {
    case s #:: tail =>
      p.absorb(s) match {
        case Right(js) =>
          js.foreach(sink)
          loop(tail)
        case Left(e) =>
          Left(e)
      }
    case _ =>
      p.finish().right.map(_.foreach(sink))
  }
  
loop(chunks)

You can also call jawn.Parser.async[J] to use async parsing with an arbitrary data type (provided you also have an implicit Facade[J]).

Supporting external ASTs with Jawn

Jawn currently supports six external ASTs directly:

Argonaut (6.0.4)
Json4s (3.2.11)
Play (2.3.6)
Rojoma (2.4.3)
Rojoma-v3 (3.2.1)
Spray (1.3.1)

Each of these subprojects provides a Parser object (an instance of SupportParser[J]) that is parameterized on the given project's AST (J). The following methods are available:

Parser.parseUnsafe(String) → J
Parser.parseFromString(String) → Try[J]
Parser.parsefromPath(String) → Try[J]
Parser.parseFromFile(File) → Try[J]
Parser.parseFromChannel(ReadableByteChannel) → Try[J]
Parser.parseFromByteBuffer(ByteBuffer) → Try[J]

These methods parallel those provided by jawn.Parser.

For the following snippets, XYZ is one of (argonaut, json4s, play, rojoma, rojoma-v3 or spray):

This is how you would include the subproject in build.sbt:

resolvers += Resolver.sonatypeRepo("releases")

libraryDependencies += "org.spire-math" %% jawn-"XYZ" % "0.8.3"

This is an example of how you might use the parser into your code:

import jawn.support.XYZ.Parser

val myResult = Parser.parseFromString(myString)

Do-It-Yourself Parsing

Jawn supports building any JSON AST you need via type classes. You benefit from Jawn's fast parser while still using your favorite Scala JSON library. This mechanism is also what allows Jawn to provide "support" for other libraries' ASTs.

To include Jawn's parser in your project, add the following snippet to your build.sbt file:

resolvers += Resolver.sonatypeRepo("releases")

libraryDependencies += "org.spire-math" %% "jawn-parser" % "0.8.3"

To support your AST of choice, you'll want to define a Facade[J] instance, where the J type parameter represents the base of your JSON AST. For example, here's a facade that supports Spray:

import spray.json._
object Spray extends SimpleFacade[JsValue] {
  def jnull() = JsNull
  def jfalse() = JsFalse
  def jtrue() = JsTrue
  def jnum(s: String) = JsNumber(s)
  def jint(s: String) = JsNumber(s)
  def jstring(s: String) = JsString(s)
  def jarray(vs: List[JsValue]) = JsArray(vs)
  def jobject(vs: Map[String, JsValue]) = JsObject(vs)
}

Most ASTs will be easy to define using the SimpleFacade or MutableFacade traits. However, if an ASTs object or array instances do more than just wrap a Scala collection, it may be necessary to extend Facade directly.

You can also look at the facades used by the support projects to help you create your own. This could also be useful if you wanted to use an older version of a supported library.

Using the AST

Access

For accessing atomic values, JValue supports two sets of methods: get-style methods and as-style methods.

The get-style methods return Some(_) when called on a compatible JSON value (e.g. strings can return Some[String], numbers can return Some[Double], etc.), and None otherwise:

getBoolean → Option[Boolean]
getString → Option[String]
getLong → Option[Long]
getDouble → Option[Double]
getBigInt → Option[BigInt]
getBigDecimal → Option[BigDecimal]

In constrast, the as-style methods will either return an unwrapped value (instead of returning Some(_)) or throw an exception (instead of returning None):

asBoolean → Boolean // or exception
asString → String // or exception
asLong → Long // or exception
asDouble → Double // or exception
asBigInt → BigInt // or exception
asBigDecimal → BigDecimal // or exception

To access elements of an array, call get with an Int position:

get(i: Int) → JValue // returns JNull if index is illegal

To access elements of an object, call get with a String key:

get(k: String) → JValue // returns JNull if key is not found

Both of these methods also return JNull if the value is not the appropraite container. This allows the caller to chain lookups without having to check that each level is correct:

val v: JValue = ???

// returns JNull if a problem is encountered in structure of 'v'.
val t: JValue = v.get("novels").get(0).get("title")

// if 'v' had the right structure and 't' is JString(s), then Some(s).
// otherwise, None.
val titleOrNone: Option[String] = t.getString

// equivalent to titleOrNone.getOrElse(throw ...)
val titleOrDie: String = t.asString

Updating

The atomic values (JNum, JBoolean, JNum, and JString) are immutable.

Objects are fully-mutable and can have items added, removed, or changed:

set(k: String, v: JValue) → Unit
remove(k: String) → Option[JValue]

If set is called on a non-object, an exception will be thrown. If remove is called on a non-object, None will be returned.

Arrays are semi-mutable. Their values can be changed, but their size is fixed:

set(i: Int, v: JValue) → Unit

If set is called on a non-array, or called with an illegal index, an exception will be thrown.

(A future version of Jawn may provide an array whose length can be changed.)

Profiling

Jawn uses JMH along with the sbt-jmh plugin.

Running Benchmarks

The benchmarks are located in the benchmark project. You can run the benchmarks by typing benchmark/run from SBT. There are many supported arguments, so here are a few examples:

Run all benchmarks, with 10 warmups, 10 iterations, using 3 threads:

benchmark/run -wi 10 -i 10 -f1 -t3

Run just the CountriesBench test (5 warmups, 5 iterations, 1 thread):

benchmark/run -wi 5 -i 5 -f1 -t1 .*CountriesBench

Benchmark Issues

Currently, the benchmarks are a bit fiddily. The most obvious symptom is that if you compile the benchmarks, make changes, and compile again, you may see errors like:

[error] (benchmark/jmh:generateJavaSources) java.lang.NoClassDefFoundError: jawn/benchmark/Bla25Bench

The fix here is to run benchmark/clean and try again.

You will also see intermittent problems like:

[error] (benchmark/jmh:compile) java.lang.reflect.MalformedParameterizedTypeException

The solution here is easier (though frustrating): just try it again. If you continue to have problems, consider cleaning the project and trying again.

(In the future I hope to make the benchmarking here a bit more resilient. Suggestions and pull requests gladly welcome!)

Files

The benchmarks use files located in benchmark/src/main/resources. If you want to test your own files (e.g. mydata.json), you would:

Copy the file to benchmark/src/main/resources/mydata.json.
Add the following code to JmhBenchmarks.scala:

class MyDataBench extends JmhBenchmarks("mydata.json")

Jawn has been tested with much larger files, e.g. 100M - 1G, but these are obviously too large to ship with the project.

With large files, it's usually easier to comment out most of the benchmarking methods and only test one (or a few) methods. Some of the slower JSON parsers get much slower for large files.

Interpreting the results

Remember that the benchmarking results you see will vary based on:

Hardware
Java version
JSON file size
JSON file structure
JSON data values

I have tried to use each library in the most idiomatic and fastest way possible (to parse the JSON into a simple AST). Pull requests to update library versions and improve usage are very welcome.

Future Work

More support libraries could be added.

It's likely that some of Jawn's I/O could be optimized a bit more, and also made more configurable. The heuristics around all-at-once loading versus input chunking could definitely be improved.

In cases where the user doesn't need fast lookups into JSON objects, an even lighter AST could be used to improve parsing and rendering speeds.

Strategies to cache/intern field names of objects could pay big dividends in some cases (this might require AST changes).

If you have ideas for any of these (or other ideas) please feel free to open an issue or pull request so we can talk about it.

Disclaimers

Jawn only supports UTF-8 when parsing bytes. This might change in the future, but for now that's the target case. You can always decode your data to a string, and handle the character set decoding using Java's standard tools.

Jawn's AST is intended to be very lightweight and simple. It supports simple access, and limited mutable updates. It intentionally lacks the power and sophistication of many other JSON libraries.

Copyright and License

All code is available to you under the MIT license, available at http://opensource.org/licenses/mit-license.php.

julien-lafont/jawn