Fetching contributors…
Cannot retrieve contributors at this time
314 lines (216 sloc) 11.4 KB

YAJL C Bindings for Ruby

This gem is a C binding to the excellent YAJL JSON parsing and generation library.

You can read more info at the projects website or check out it's codes at


  • JSON parsing and encoding directly to and from an IO stream (file, socket, etc) or String. Compressed stream parsing and encoding supported for Bzip2, Gzip and Deflate.

  • Parse and encode multiple JSON objects to and from streams or strings continuously.

  • JSON gem compatibility API - allows yajl-ruby to be used as a drop-in replacement for the JSON gem

  • Basic HTTP client (only GET requests supported for now) which parses JSON directly off the response body *as it's being received*

  • ~3.5x faster than JSON.generate

  • ~1.9x faster than JSON.parse

  • ~4.5x faster than YAML.load

  • ~377.5x faster than YAML.dump

  • ~1.5x faster than Marshal.load

  • ~2x faster than Marshal.dump

How to install

Install it like any other gem hosted at the Githubs like so:

(more instructions here:

sudo gem install brianmario-yajl-ruby

Example of use

NOTE: I'm building up a collection of small examples in the examples ( folder.

First, you're probably gonna want to require it:

require 'yajl'


Then maybe parse some JSON from:

a File IO

json ='test.json', 'r')
parser =
hash = parser.parse(json)

or maybe a StringIO

json ="...some JSON...")
parser =
hash = parser.parse(json)

or maybe STDIN

cat someJsonFile.json | ruby -ryajl -e "puts Yajl::Parser.parse(STDIN).inspect"

Or lets say you didn't have access to the IO object that contained JSON data, but instead only had access to chunks of it at a time. No problem!

(Assume we're in an EventMachine::Connection instance)

def post_init
   @parser =

def object_parsed(obj)
   puts "Sometimes one pays most for the things one gets for nothing. - Albert Einstein"
   puts obj.inspect

def connection_completed
  # once a full JSON object has been parsed from the stream
  # object_parsed will be called, and passed the constructed object
  @parser.on_parse_complete = method(:object_parsed)

def receive_data(data)
  # continue passing chunks
  @parser << data

Or if you don't need to stream it, it'll just return the built object from the parse when it's done.

NOTE: if there are going to be multiple JSON strings in the input, you *must* specify a block or callback as this
is how yajl-ruby will hand you (the caller) each object as it's parsed off the input.

obj = Yajl::Parser.parse(str_or_io)

Or how about a JSON API HTTP request? This actually makes a request using a raw TCPSocket, then parses the JSON body right off the socket. While it's being received over the wire!

require 'uri'
require 'yajl/http_stream'

url = URI.parse("")
results = Yajl::HttpStream.get(url)

Or do the same request, with Gzip and Deflate output compression support (also supports Bzip2, if loaded): (this does the same raw socket Request, but transparently parses the compressed response body)

require 'uri'
require 'yajl/gzip'
require 'yajl/deflate'
require 'yajl/http_stream'

url = URI.parse("")
results = Yajl::HttpStream.get(url)

Since yajl-ruby parses JSON as a stream, supporting API's like Twitter's Streaming API are a piece-of-cake. You can simply supply a block to Yajl::HttpStream.get, which is used as the callback for when a JSON object has been unserialized off the stream. For the case of this Twitter Streaming API call, the callback gets fired a few times a second (depending on your connection speed). The code below is all that's needed to make the request and stream unserialized Ruby hashes off the response, continuously. You'll note that I've enabled the :symbolize_keys parser option as well. Doing so is much more efficient for parsing JSON streams with lots of repetitive keys - for things like result sets or multiple API requests - than the same parse with string keys. This is because Ruby will reuse (and never GC) it's symbol table. Be that as it may, if you want to parse JSON strings with random key names it's much better to leave string keys enabled (the default), so they can get GC'd later.

require 'uri'
require 'yajl/http_stream'

uri = URI.parse("http://#{username}:#{password}")
Yajl::HttpStream.get(uri, :symbolize_keys => true) do |hash|
  puts hash.inspect

Or how about parsing directly from a compressed file?

require 'yajl/bzip2'

file ='some.json.bz2', 'r')
result = Yajl::Bzip2::StreamReader.parse(file)


Since yajl-ruby does everything using streams, you simply need to pass the object to encode, and the IO to write the stream to (this happens in chunks).

This allows you to encode JSON as a stream, writing directly to a socket

socket =, 9000)
hash = {:foo => 12425125, :bar => "some string", ... }
encoder =
Yajl::Encoder.encode(hash, socket)

Or what if you wanted to compress the stream over the wire?

require 'yajl/gzip'
socket =, 9000)
hash = {:foo => 12425125, :bar => "some string", ... }
Yajl::Gzip::StreamWriter.encode(hash, socket)

Or what about encoding multiple objects to JSON over the same stream? This example will encode and send 50 JSON objects over the same stream, continuously.

socket =, 9000)
encoder =
50.times do
  hash = {:current_time =>, :foo => 12425125}
  encoder.encode(hash, socket)

Using EventMachine and you want to encode and send in chunks? (Assume we're in an EventMachine::Connection instance)

def post_init
   # Passing a :terminator character will let us determine when the encoder
   # is done encoding the current object
   @encoder = => nil)
   motd_contents ="/path/to/motd.txt")
   status ="/path/to/huge/status_file.txt")
   @motd = {:motd => motd_contents, :system_status => status}

def connection_completed
  # The encoder will do it's best to hand you data in chunks that
  # are around 8kb (but you may see some that are larger)
  # It should be noted that you could have also assigned the _on_progress_ callback
  # much like you can assign the _on_parse_complete_ callback with the parser class.
  # Passing a block (like below) essentially tells the encoder to use that block
  # as the callback normally assigned to _on_progress_.
  # Send our MOTD and status
  @encoder.encode(@motd) do |chunk|
    if chunk.nil? # got our terminator, encoding is done

But to make things simple, you might just want to let yajl-ruby do all the hard work for you and just hand back a string when it's finished. In that case, just don't provide and IO or block (or assign the on_progress callback).

str = Yajl::Encoder.encode(obj)

You can also use Yajl::Bzip2::StreamWriter and Yajl::Deflate::StreamWriter. So you can pick whichever fits your CPU/bandwidth sweet-spot.

JSON gem Compatibility API

The JSON gem compatibility API isn't enabled by default. You have to explicitly require it like so:

require 'yajl/json_gem'

That's right, you can just replace “require 'json'” with the line above and you're done!

This will require yajl-ruby itself, as well as enable it's JSON gem compatibility API.

This includes the following API:

JSON.parse, JSON.generate, JSON.pretty_generate, JSON.load, JSON.dump
and all of the #to_json instance method overrides for Ruby's primitive objects

Once the compatibility API is enabled, your existing or new project should work as if the JSON gem itself were being used. Only you'll be using Yajl ;)

There are a lot more possibilities that I'd love to see other gems/plugins for someday.

Some ideas are:


After I finished implementation - this library performs close to the same as the current JSON.parse (C gem) does on small/medium files.

But on larger files, and higher amounts of iteration, this library was around 2x faster than JSON.parse.

The main benefit of this library is in it's memory usage. Since it's able to parse the stream in chunks, it's memory requirements are very, very low.

Here's what parsing a 2.43MB JSON file off the filesystem 20 times looks like:

Memory Usage


  • Yajl::Parser#parse: 32MB

  • JSON.parse: 54MB

  • ActiveSupport::JSON.decode: 63MB


  • Yajl::Parser#parse: 32MB

  • JSON.parse: 57MB

  • ActiveSupport::JSON.decode: 67MB

Parse Time

  • Yajl::Parser#parse: 4.54s

  • JSON.parse: 5.47s

  • ActiveSupport::JSON.decode: 64.42s

Encode Time

  • Yajl::Encoder#encode: 3.59s

  • JSON#to_json: 6.2s

  • ActiveSupport::JSON.encode: 45.58s

Compared to YAML

NOTE: I converted the 2.4MB JSON file to YAML for this test.

Parse Time (from their respective formats)

  • Yajl::Parser#parse: 4.33s

  • JSON.parse: 5.37s

  • YAML.load: 19.47s

Encode Time (to their respective formats)

  • Yajl::Encoder#encode: 3.47s

  • JSON#to_json: 6.6s

  • YAML.dump(obj, io): 1309.93s

Compared to Marshal.load/Marshal.dump

NOTE: I converted the 2.4MB JSON file to a Hash and a dump file from Marshal.dump for this test.

Parse Time (from their respective formats)

  • Yajl::Parser#parse: 4.54s

  • JSON.parse: 7.40s

  • Marshal.load: 7s

Encode Time (to their respective formats)

  • Yajl::Encoder#encode: 2.39s

  • JSON#to_json: 8.37s

  • Marshal.dump: 4.66s

Third Party Sources Bundled

This project includes code from the BSD licensed yajl project, copyright 2007-2009 Lloyd Hilaiel

Special Thanks

For those of you using yajl-ruby out in the wild, please hit me up on Twitter (brianmario) or send me a message here on the Githubs describing the site and how you're using it. I'd love to get a list going!

I've had a lot of inspiration, and a lot of help. Thanks to everyone who's been a part of this and those to come!

  • Lloyd Hilaiel - - for writing Yajl!!

  • Josh Ferguson - - for peer-pressuring me into getting back into C; it worked ;) Also tons of support over IM

  • Jonathan Novak - - pointer-hacking help

  • Tom Smith - - pointer-hacking help

  • Rick - for making an ActiveSupport patch with support for this library and teasing me that it might go into Rails 3. You sure lit a fire under my ass and I got a ton of work done because of it! :)

  • The entire Github Crew - - my inspiration, time spent writing this, finding Yajl, So many-MANY other things wouldn't have been possible without this awesome service. I owe you guys some whiskey at Kilowatt.

  • benburkert -

  • Aman Gupta - - tons of suggestions and inspiration for the most recent features, and hopefully more to come ;)