oj 2.18.3 → 3.13.14

data/pages/Parser.md

@@ -0,0 +1,309 @@
+# How Oj Just Got Faster
+
+The original Oj parser is a performant parser that supports several
+modes. As of this writing Oj is almost 10 years old. A dinosaur by
+coding standards. It was time for an upgrade. Dealing with issues over
+the years it became clear that a few things could have been done
+better. The new `Oj::Parser` is a response that not only attempts to
+address some of the issues but also give the Oj parser a significant
+boost in performance. `Oj::Parser` takes a different approach to JSON
+parsing than the now legacy Oj parser. Not really a legacy parser yet
+since the `Oj::Parser` is not a drop-in replacement for the JSON gem
+but it is as much 3 times or more faster than the previous parser in
+some modes.
+
+## Address Issues
+
+There are a few features of the`Oj.load` parser that continue to be
+the reason for many of the issue on the project. The most significant
+area is compatibility with both Rails and the JSON gem as they battle
+it out for which behavior will win out in any particular
+situation. Most of the issues are on the writing or dumping side of
+the JSON packages but some are present on the parsing as
+well. Conversion of decimals is one area where the Rails and the JSON
+gem vary. The `Oj::Parser` addresses this by allowing for completely
+separate parser instances. Create a parser and configure it for the
+situation and leave the others parsers on their own.
+
+The `Oj::Parser` is mostly compatible with the JSON gem and Rails but
+no claims are made that the behavior will be the same as either.
+
+The most frequent issues that can addressed with the new parser are
+around the handling of options. For `Oj.load` there is a set of
+default options that can be set and the same options can be specified
+for each call to parse or load. This approach as a couple of
+downsides. One the defaults are shared across all calls to parse no
+matter what the desire mode is. The second is that having to provide
+all the options on each parse call incurs a performance penalty and is
+just annoying to repeat the same set of options over may calls.
+
+By localizing options to a specific parser instance there is never any
+bleed over to other instances.
+
+## How
+
+It's wonderful to wish for a faster parser that solves all the
+annoyances of the previous parser but how was it done is a much more
+interesting question to answer.
+
+At the core, the API for parsing was changed. Instead of a sinle
+global parser any number of parsers can be created and each is separate
+from the others. The parser itself is able to rip through a JSON
+string, stream, or file and then make calls to a delegate to process
+the JSON elements according to the delegate behavior. This is similar
+to the `Oj.load` parser but the new parser takes advantage of
+character maps, reduced conditional branching, and calling function
+pointers.
+
+### Options
+
+As mentioned, one way to change the options issues was to change the
+API. Instead of having a shared set of default options a separate
+parser is created and configured for each use case. Options are set
+with methods on the parser so no more guessing what options are
+available. With options isolated to individual parsers there is no
+unintended leakage to other parse use cases.
+
+### Structure
+
+A relative small amount of time is spent in the actual parsing of JSON
+in `Oj.load`. Most of the time is spent building the Ruby
+Objects. Even cutting the parsing time in half only gives a 10%
+improvement in performance but 10% is still an improvement.
+
+The `Oj::Parser` is designed to reduce conditional branching. To do
+that it uses character maps for the various states that the parser
+goes through when parsing. There is no recursion as the JSON elements
+are parsed. The use of a character maps for each parser state means
+the parser function can and is re-entrant so partial blocks of JSON
+can be parsed and the results combined.
+
+There are no Ruby calls in the parser itself. Instead delegates are
+used to implement the various behaviors of the parser which are
+currently validation (validate), callbacks (SAJ), or building Ruby
+objects (usual). The delegates are where all the Ruby calls and
+related optimizations take place.
+
+Considering JSON file parsing, `Oj.load_file` is able to read a file a
+block at a time and the new `Oj::Parser` does the same. There was a
+change in how that is done though. `Oj.load_file` sets up a reader
+that must be called for each character. Basically a buffered
+reader. `Oj::Parser` drops down a level and uses a re-entrant parser
+that takes a block of bytes at a time so there is no call needed for
+each character but rather just iterating over the block read from the
+file.
+
+Reading a block at a time also allows for an efficient second thread
+to be used for reading blocks. That feature is not in the first
+iteration of the `Oj::Parser` but the stage is set for it in the
+future. The same approach was used successfully in
+[OjC](https://github.com/ohler55/ojc) which is where the code for the
+parser was taken from.
+
+### Delegates
+
+There are three delegates; validate, SAJ, and usual.
+
+#### Validate
+
+The validate delegate is trivial in that does nothing other than let
+the parser complete. There are no options for the validate
+delegate. By not making any Ruby calls other than to start the parsing
+the validate delegate is no surprise that the validate delegate is the
+best performer.
+
+#### SAJ (Simple API for JSON)
+
+The SAJ delegate is compatible with the SAJ handlers used with
+`Oj.saj_parse` so it needs to keep track of keys for the
+callbacks. Two optimizations are used. The first is a reuseable key
+stack while the second is a string cache similar to the Ruby intern
+function.
+
+When parsing a Hash (JSON object) element the key is passed to the
+callback function if the SAJ handler responds to the method. The key
+is also provided when closing an Array or Hash that is part of a
+parent Hash. A key stack supports this.
+
+If the option is turned on a lookup is made and previously cached key
+VALUEs are used. This avoids creating the string for the key and
+setting the encoding on it. The cache used is a auto expanding hash
+implementation that is limited to strings less than 35 characters
+which covers most keys. Larger strings use the slower string creation
+approach. The use of the cache reduces object creation which save on
+both memory allocation and time. It is not appropriate for one time
+parsing of say all the keys in a dictionary but is ideally suited for
+loading similar JSON multiple times.
+
+#### Usual
+
+By far the more complex of the delegates is the 'usual' delegate. The
+usual delegate builds Ruby Objects when parsing JSON. It incorporates
+many options for configuration and makes use of a number of
+optimizations.
+
+##### Reduce Branching
+
+In keeping with the goal of reducing conditional branching most of the
+delegate options are implemented by changing a function pointer
+according to the option selected. For example when turning on or off
+`:symbol_keys` the function to calculate the key is changed so no
+decision needs to be made during parsing. Using this approach option
+branching happens when the option is set and not each time when
+parsing.
+
+##### Cache
+
+Creating Ruby Objects whether Strings, Array, or some other class is
+expensive. Well expensive when running at the speeds Oj runs at. One
+way to reduce Object creation is to cache those objects on the
+assumption that they will most likely be used again. This is
+especially true of Hash keys and Object attribute IDs. When creating
+Objects from a class name in the JSON a class cache saves resolving
+the string to a class each time. Of course there are times when
+caching is not preferred so caching can be turned on or off with
+option methods on the parser which are passed down to the delegate..
+
+The Oj cache implementation is an auto expanding hash. When certain
+limits are reached the hash is expanded and rehashed. Rehashing can
+take some time as the number of items cached increases so there is
+also an option to start with a larger cache size to avoid or reduce
+the likelihood of a rehash.
+
+The Oj cache has an advantage over the Ruby intern function
+(`rb_intern()`) in that several steps are needed for some cached
+items. As an example Object attribute IDs are created by adding an `@`
+character prefix to a string and then converting to a ID. This is done
+once when inserting into the cache and after that only a lookup is
+needed.
+
+##### Bulk Insert
+
+The Ruby functions available for C extension functions are extensive
+and offer many options across the board. The bulk insert functions for
+both Arrays and Hashes are much faster than appending or setting
+functions that set one value at a time. The Array bulk insert is
+around 15 times faster and for Hash it is about 3 times faster.
+
+To take advantage of the bulk inserts arrays of VALUEs are
+needed. With a little planning there VALUE arrays can be reused which
+leads into another optimization, the use of stacks.
+
+##### Stacks
+
+Parsing requires memory to keep track of values when parsing nested
+JSON elements. That can be done on the call stack making use of
+recursive calls or it can be done with a stack managed by the
+parser. The `Oj.load` method maintains a stack for Ruby object and
+builds the output as the parsing progresses.
+
+`Oj::Parser` uses three different stacks. One stack for values, one
+for keys, and one for collections (Array and Hash). By postponing the
+creation of the collection elements the bulk insertions for Array and
+Hash can be used. For arrays the use of a value stack and creating the
+array after all elements have been identified gives a 15x improvement
+in array creation.
+
+For Hash the story is a little different. The bulk insert for Hash
+alternates keys and values but there is a wrinkle to consider. Since
+Ruby Object creation is triggered by the occurrence of an element that
+matches a creation identifier the creation of a collection is not just
+for Array and Hash but also Object. Setting Object attributes uses an
+ID and not a VALUE. For that reason the keys should not be created as
+String or Symbol types as they would be ignored and the VALUE creation
+wasted when setting Object attributes. Using the bulk insert for Hash
+gives a 3x improvement for that part of the object building.
+
+Looking at the Object creation the JSON gem expects a class method of
+`#json_create(arg)`. The single argument is the Hash resulting from
+the parsing assuming that the parser parsed to a Hash first. This is
+less than ideal from a performance perspective so `Oj::Parser`
+provides an option to take that approach or to use the much more
+efficient approach of never creating the Hash but instead creating the
+Object and then setting the attributes directly.
+
+To further improve performance and reduce the amount of memory
+allocations and frees the stacks are reused from one call to `#parse`
+to another.
+
+## Results
+
+The results are even better than expected. Running the
+[perf_parser.rb](https://github.com/ohler55/oj/blob/develop/test/perf_parser.rb)
+file shows the improvements. There are four comparisons all run on a
+MacBook Pro with Intel processor.
+
+### Validation
+
+Without a comparible parser that just validates a JSON document the
+`Oj.saj_parse` callback parser with a nil handler is used for
+comparison to the new `Oj::Parser.new(:validate)`. In that case the
+comparison is:
+
+```
+             System  time (secs)  rate (ops/sec)
+-------------------  -----------  --------------
+Oj::Parser.validate       0.101      494369.136
+       Oj::Saj.none       0.205      244122.745
+```
+
+The `Oj::Parser.new(:validate)` is **2.03** times faster!
+
+### Callback
+
+Oj has two callback parsers. One is SCP and the other SAJ. Both are
+similar in that a handler is provided that implements methods for
+processing the various element types in a JSON document. Comparing
+`Oj.saj_parse` to `Oj::Parser.new(:saj)` with a all callback methods
+implemented handler gives the following raw results:
+
+```
+        System  time (secs)  rate (ops/sec)
+--------------  -----------  --------------
+Oj::Parser.saj       0.783       63836.986
+   Oj::Saj.all       1.182       42315.397
+```
+
+The `Oj::Parser.new(:saj)` is **1.51** times faster.
+
+### Parse to Ruby primitives
+
+Parsing to Ruby primitives and Array and Hash is possible with most
+parsers including the JSON gem parser. The raw results comparing
+`Oj.strict_load`, `Oj::Parser.new(:usual)`, and the JSON gem are:
+
+```
+          System  time (secs)  rate (ops/sec)
+----------------  -----------  --------------
+Oj::Parser.usual       0.452      110544.876
+ Oj::strict_load       0.699       71490.257
+       JSON::Ext       1.009       49555.094
+```
+
+The `Oj::Parser.new(:saj)` is **1.55** times faster than `Oj.load` and
+**2.23** times faster than the JSON gem.
+
+### Object
+
+Oj supports two modes for Object serialization and
+deserialization. Comparing to the JSON gem compatible mode
+`Oj.compat_load`, `Oj::Parser.new(:usual)`, and the JSON gem yields
+the following raw results:
+
+```
+          System  time (secs)  rate (ops/sec)
+----------------  -----------  --------------
+Oj::Parser.usual       0.071      703502.033
+ Oj::compat_load       0.225      221762.927
+       JSON::Ext       0.401      124638.859
+```
+
+The `Oj::Parser.new(:saj)` is **3.17** times faster than
+`Oj.compat_load` and **5.64** times faster than the JSON gem.
+
+## Summary
+
+With a performance boost of from 1.5x to over 3x over the `Oj.load`
+parser the new `Oj::Parser` is a big win in the performance arena. The
+isolation of options is another feature that should make life easier
+for developers.

data/pages/Rails.md

@@ -0,0 +1,167 @@
+# Rails Quickstart
+
+To universally replace Rails' use of the json gem with Oj, and also
+have Oj "take over" many methods on the JSON constant (`load`, `parse`, etc.) with
+their faster Oj counterparts, add this to an initializer:
+
+```ruby
+Oj.optimize_rails()
+```
+
+For more details and options, read on...
+
+# Oj Rails Compatibility
+
+The `:rails` mode mimics the ActiveSupport version 5 encoder. Rails and
+ActiveSupport are built around the use of the `as_json(*)` method defined for
+a class. Oj attempts to provide the same functionality by being a drop in
+replacement with a few exceptions.
+
+```ruby
+require 'oj'
+
+Oj::Rails.set_encoder()
+Oj::Rails.set_decoder()
+Oj::Rails.optimize()
+Oj::Rails.mimic_JSON()
+```
+
+or simply call
+
+```ruby
+Oj.optimize_rails()
+```
+
+Either of those steps will setup Oj to mimic Rails but it will not change the
+default mode type as the mode type is only used when calling the Oj encoding
+directly. If Rails mode is also desired then use the `Oj.default_options` to
+change the default mode.
+
+Some of the Oj options are supported as arguments to the encoder if called
+from `Oj::Rails.encode()` but when using the `Oj::Rails::Encoder` class the
+`encode()` method does not support optional arguments as required by the
+ActiveSupport compliance guidelines. The general approach Rails takes for
+configuring encoding options is to either set global values or to create a new
+instance of the Encoder class and provide options in the initializer.
+
+The globals that ActiveSupport uses for encoding are:
+
+ * `ActiveSupport::JSON::Encoding.use_standard_json_time_format`
+ * `ActiveSupport::JSON::Encoding.escape_html_entities_in_json`
+ * `ActiveSupport::JSON::Encoding.time_precision`
+ * `ActiveSupport::JSON::Encoding.json_encoder`
+
+Those globals are aliased to also be accessed from the ActiveSupport module
+directly so `ActiveSupport::JSON::Encoding.time_precision` can also be accessed
+from `ActiveSupport.time_precision`. Oj makes use of these globals in mimicking
+Rails after the `Oj::Rails.set_encode()` method is called. That also sets the
+`ActiveSupport.json_encoder` to the `Oj::Rails::Encoder` class.
+
+Options passed into a call to `to_json()` are passed to the `as_json()`
+methods. These are mostly ignored by Oj and simply passed on without
+modifications as per the guidelines. The exception to this are the options
+specific to Oj such as the `:circular` option which it used to detect circular
+references while encoding.
+
+By default Oj acts like the ActiveSupport encoder and honors any changes in
+the `as_json()` methods. There are some optimized Oj encoders for some
+classes. When the optimized encoder it toggled the `as_json()` methods will not
+be called for that class but instead the optimized version will be called. The
+optimized version is the same as the ActiveSupport default encoding for a
+given class. The optimized versions are toggled with the `optimize()` and
+`deoptimize()` methods. There is a default optimized version for every class
+that takes the visible attributes and encodes them but that may not be the
+same as what Rails uses. Trial and error is the best approach for classes not
+listed here.
+
+The classes that can be put in optimized mode and are optimized when
+`Oj::Rails.optimize` is called with no arguments are:
+
+ * Array
+ * BigDecimal
+ * Float
+ * Hash
+ * Range
+ * Regexp
+ * Time
+ * ActiveSupport::TimeWithZone
+ * ActionController::Parameters
+ * any class inheriting from ActiveRecord::Base
+ * any other class where all attributes should be dumped
+
+The ActiveSupport decoder is the `JSON.parse()` method. Calling the
+`Oj::Rails.set_decoder()` method replaces that method with the Oj equivalent.
+
+### Usage in Rails 3
+
+To support Rails 3 you can create a new module mixin to prepend to controllers:
+
+```ruby
+require 'oj'
+
+module OjJsonEncoder
+  def render(options = nil, extra_options = {}, &block)
+    if options && options.is_a?(Hash) && options[:json]
+      obj = options.delete(:json)
+      obj = Oj.dump(obj, :mode => :rails) unless obj.is_a?(String)
+      options[:text] = obj
+      response.content_type ||= Mime::JSON
+    end
+    super
+  end
+end
+```
+
+Usage:
+
+```ruby
+class MyController < ApplicationController
+  prepend OjJsonEncoder
+  def index
+    render :json => { :hello => 'world' }
+  end
+end
+```
+
+### Older Ruby Version Support (Pre 2.3.0)
+
+If you are using an older version of Ruby, you can pin `oj` to an earlier version in your Gemfile:
+
+```ruby
+gem 'oj', '3.7.12'
+```
+
+### Notes:
+
+1. Optimized Floats set the significant digits to 16. This is different than
+   Ruby which is used by the json gem and by Rails. Ruby varies the
+   significant digits which can be either 16 or 17 depending on the value.
+
+2. Optimized Hashes do not collapse keys that become the same in the output. As
+   an example, a non-String object that has a `to_s()` method will become the
+   return value of the `to_s()` method in the output without checking to see if
+   that has already been used. This could occur is a mix of String and Symbols
+   are used as keys or if a other non-String objects such as Numerics are mixed
+   with numbers as Strings.
+
+3. To verify Oj is being used turn on the Oj `:trace` option. Similar to the
+   Ruby Tracer Oj will then print out trace information. Another approach is
+   to turn on C extension tracing.  Set `tracer = TracePoint.new(:c_call) do
+   |tp| p [tp.lineno, tp.event, tp.defined_class, tp.method_id] end` or, in
+   older Rubies, set `Tracer.display_c_call = true`.
+
+   For example:
+
+     ```
+     require 'active_support/core_ext'
+     require 'active_support/json'
+     require 'oj'
+     Oj.optimize_rails
+     tracer.enable { Time.now.to_json }
+     # prints output including
+     ....
+     [20, :c_call, #<Class:Oj::Rails::Encoder>, :new]
+     [20, :c_call, Oj::Rails::Encoder, :encode]
+     ....
+     => "\"2018-02-23T12:13:42.493-06:00\""
+     ```

data/pages/Security.md

@@ -0,0 +1,20 @@
+# Security and Optimization
+
+Two settings in Oj are useful for parsing but do expose a vulnerability if used
+from an untrusted source. Symbolized keys can cause memory to be filled with
+previous versions of ruby. Ruby 2.1 and below does not garbage collect
+Symbols. The same is true for auto defining classes in all versions of ruby;
+memory will also be exhausted if too many classes are automatically
+defined. Auto defining is a useful feature during development and from trusted
+sources but it allows too many classes to be created in the object load mode and
+auto defined is used with an untrusted source. The `Oj.safe_load()` method
+sets and uses the most strict and safest options. It should be used by
+developers who find it difficult to understand the options available in Oj.
+
+The options in Oj are designed to provide flexibility to the developer. This
+flexibility allows Objects to be serialized and deserialized. No methods are
+ever called on these created Objects but that does not stop the developer from
+calling methods on them. As in any system, check your inputs before working with
+them. Taking an arbitrary `String` from a user and evaluating it is never a good
+idea from an unsecure source. The same is true for `Object` attributes as they
+are not more than `String`s. Always check inputs from untrusted sources.

data/pages/WAB.md

@@ -0,0 +1,13 @@
+# WAB mode
+
+The `:wab` mode ignores all options except the indent option. Performance of
+this mode is slightly faster than the :strict and :null modes. It is included
+to support the [WABuR](https://github.com/ohler55/wabur) project.
+
+Options other than the indentation are not supported since the encoding and
+formats are defined by the API that is used to encode data being passed from
+one components in a WAB system and allowing an option that would break the
+data exchange is best not supported.
+
+The mode encodes like the strict mode except the URI, Time, WAB::UUID, and
+BigDecimal are supported.

data/test/activerecord/result_test.rb

@@ -0,0 +1,32 @@
+#!/usr/bin/env ruby
+
+$: << File.dirname(__FILE__)
+$: << File.dirname(File.dirname(__FILE__))
+
+require 'helper'
+require "rails/all"
+
+Oj::Rails.set_encoder()
+Oj::Rails.optimize()
+
+Oj.default_options = { mode: :rails }
+
+class ActiveRecordResultTest < Minitest::Test
+  def test_hash_rows
+
+    result = ActiveRecord::Result.new(["one", "two"],
+				      [
+					["row 1 col 1", "row 1 col 2"],
+					["row 2 col 1", "row 2 col 2"],
+					["row 3 col 1", "row 3 col 2"],
+				      ])
+    #puts "*** result: #{Oj.dump(result, indent: 2)}"
+    json_result = if ActiveRecord.version >= Gem::Version.new("6")
+                    result.to_a
+                  else
+                    result.to_hash
+                  end
+
+    assert_equal Oj.dump(result, mode: :rails), Oj.dump(json_result)
+  end
+end

data/test/activesupport4/decoding_test.rb

@@ -0,0 +1,108 @@
+# encoding: utf-8
+require 'activesupport4/test_helper'
+require 'active_support/json'
+require 'active_support/time'
+
+class TestJSONDecoding < ActiveSupport::TestCase
+  class Foo
+    def self.json_create(object)
+      "Foo"
+    end
+  end
+
+  TESTS = {
+    %q({"returnTo":{"\/categories":"\/"}})        => {"returnTo" => {"/categories" => "/"}},
+    %q({"return\\"To\\":":{"\/categories":"\/"}}) => {"return\"To\":" => {"/categories" => "/"}},
+    %q({"returnTo":{"\/categories":1}})          => {"returnTo" => {"/categories" => 1}},
+    %({"returnTo":[1,"a"]})                    => {"returnTo" => [1, "a"]},
+    %({"returnTo":[1,"\\"a\\",", "b"]})        => {"returnTo" => [1, "\"a\",", "b"]},
+    %({"a": "'", "b": "5,000"})                  => {"a" => "'", "b" => "5,000"},
+    %({"a": "a's, b's and c's", "b": "5,000"})   => {"a" => "a's, b's and c's", "b" => "5,000"},
+    # multibyte
+    %({"matzue": "松江", "asakusa": "浅草"}) => {"matzue" => "松江", "asakusa" => "浅草"},
+    %({"a": "2007-01-01"})                       => {'a' => Date.new(2007, 1, 1)},
+    %({"a": "2007-01-01 01:12:34 Z"})            => {'a' => Time.utc(2007, 1, 1, 1, 12, 34)},
+    %(["2007-01-01 01:12:34 Z"])                 => [Time.utc(2007, 1, 1, 1, 12, 34)],
+    %(["2007-01-01 01:12:34 Z", "2007-01-01 01:12:35 Z"]) => [Time.utc(2007, 1, 1, 1, 12, 34), Time.utc(2007, 1, 1, 1, 12, 35)],
+    # no time zone
+    %({"a": "2007-01-01 01:12:34"})              => {'a' => "2007-01-01 01:12:34"},
+    # invalid date
+    %({"a": "1089-10-40"})                       => {'a' => "1089-10-40"},
+    # xmlschema date notation
+    %({"a": "2009-08-10T19:01:02Z"})             => {'a' => Time.utc(2009, 8, 10, 19, 1, 2)},
+    %({"a": "2009-08-10T19:01:02+02:00"})        => {'a' => Time.utc(2009, 8, 10, 17, 1, 2)},
+    %({"a": "2009-08-10T19:01:02-05:00"})        => {'a' => Time.utc(2009, 8, 11, 00, 1, 2)},
+    # needs to be *exact*
+    %({"a": " 2007-01-01 01:12:34 Z "})          => {'a' => " 2007-01-01 01:12:34 Z "},
+    %({"a": "2007-01-01 : it's your birthday"})  => {'a' => "2007-01-01 : it's your birthday"},
+    %([])    => [],
+    %({})    => {},
+    %({"a":1})     => {"a" => 1},
+    %({"a": ""})    => {"a" => ""},
+    %({"a":"\\""}) => {"a" => "\""},
+    %({"a": null})  => {"a" => nil},
+    %({"a": true})  => {"a" => true},
+    %({"a": false}) => {"a" => false},
+    %q({"bad":"\\\\","trailing":""}) => {"bad" => "\\", "trailing" => ""},
+    %q({"a": "http:\/\/test.host\/posts\/1"}) => {"a" => "http://test.host/posts/1"},
+    %q({"a": "\u003cunicode\u0020escape\u003e"}) => {"a" => "<unicode escape>"},
+    %q({"a": "\\\\u0020skip double backslashes"}) => {"a" => "\\u0020skip double backslashes"},
+    %q({"a": "\u003cbr /\u003e"}) => {'a' => "<br />"},
+    %q({"b":["\u003ci\u003e","\u003cb\u003e","\u003cu\u003e"]}) => {'b' => ["<i>","<b>","<u>"]},
+    # test combination of dates and escaped or unicode encoded data in arrays
+    %q([{"d":"1970-01-01", "s":"\u0020escape"},{"d":"1970-01-01", "s":"\u0020escape"}]) =>
+      [{'d' => Date.new(1970, 1, 1), 's' => ' escape'},{'d' => Date.new(1970, 1, 1), 's' => ' escape'}],
+    %q([{"d":"1970-01-01","s":"http:\/\/example.com"},{"d":"1970-01-01","s":"http:\/\/example.com"}]) =>
+      [{'d' => Date.new(1970, 1, 1), 's' => 'http://example.com'},
+       {'d' => Date.new(1970, 1, 1), 's' => 'http://example.com'}],
+    # tests escaping of "\n" char with Yaml backend
+    %q({"a":"\n"})  => {"a"=>"\n"},
+    %q({"a":"\u000a"}) => {"a"=>"\n"},
+    %q({"a":"Line1\u000aLine2"}) => {"a"=>"Line1\nLine2"},
+    # prevent json unmarshalling
+    %q({"json_class":"TestJSONDecoding::Foo"}) => {"json_class"=>"TestJSONDecoding::Foo"},
+    # json "fragments" - these are invalid JSON, but ActionPack relies on this
+    %q("a string") => "a string",
+    %q(1.1) => 1.1,
+    %q(1) => 1,
+    %q(-1) => -1,
+    %q(true) => true,
+    %q(false) => false,
+    %q(null) => nil
+  }
+
+  TESTS.each_with_index do |(json, expected), index|
+    test "json decodes #{index}" do
+      prev = ActiveSupport.parse_json_times
+      ActiveSupport.parse_json_times = true
+      silence_warnings do
+        if expected.nil?
+          assert_nil ActiveSupport::JSON.decode(json), "JSON decoding failed for #{json}"
+        else
+          assert_equal expected, ActiveSupport::JSON.decode(json), "JSON decoding failed for #{json}"
+        end
+      end
+      ActiveSupport.parse_json_times = prev
+    end
+  end
+
+  test "json decodes time json with time parsing disabled" do
+    prev = ActiveSupport.parse_json_times
+    ActiveSupport.parse_json_times = false
+    expected = {"a" => "2007-01-01 01:12:34 Z"}
+    assert_equal expected, ActiveSupport::JSON.decode(%({"a": "2007-01-01 01:12:34 Z"}))
+    ActiveSupport.parse_json_times = prev
+  end
+
+  def test_failed_json_decoding
+    assert_raise(ActiveSupport::JSON.parse_error) { ActiveSupport::JSON.decode(%(undefined)) }
+    assert_raise(ActiveSupport::JSON.parse_error) { ActiveSupport::JSON.decode(%({a: 1})) }
+    assert_raise(ActiveSupport::JSON.parse_error) { ActiveSupport::JSON.decode(%({: 1})) }
+    assert_raise(ActiveSupport::JSON.parse_error) { ActiveSupport::JSON.decode(%()) }
+  end
+
+  def test_cannot_pass_unsupported_options
+    assert_raise(ArgumentError) { ActiveSupport::JSON.decode("", create_additions: true) }
+  end
+end
+