0.99.2

• Included workaround for a C++11 issue in GCC 4.8, contributed by Alex Merry
• Fixed operator== so that json() == json(json::object())
• Fixed issue with json assignment to initializer list

• Fixed issue with assignment to empty json object with multiple keys, e.g.

  json val;
  val["key1"]["key2"] = 1;

0.99.1

• Fix to json_filter class
• Fix to readme_examples

0.99

• Fixes to deprecated json parse functions (deprecated, but still supposed to work)
• The Visual C++ specific implementation for reading floating point numbers should have freed a _locale_t object, fixed
• Added json_type_traits specialization to support assignment from non-const strings
• When parsing fractional numbers in text, floating point number precision is retained, and made available to serialization to preserve round-trip. The default output precision has been changed from 15 to 16.
• Added json std::initializer_list constructor for constructing arrays
• The deprecated json member constants null, an_object, and an_array have been removed
• Microsoft VC++ versions earlier than 2013 are no longer supported

0.98.4

• Fixes issues with compilation with clang

0.98.3

New features

• Supports Stefan Goessner's JsonPath. See example below and documentation.
• json member function find added
• json member function count added
• json array range accessor elements() added, which supports range-based for loops over json arrays, and replaces begin_elements and end_elements
• json object range accessor members() added, which supports range-based for loops over json objects, and replaces begin_members and end_members
• New version of json add member function that takes a parameter array_iterator
• json member function shrink_to_fit added

API Changes

• The json internal representation of signed and unsigned integers has been changed from long long and unsigned long long to int64_t and uint64_t. This should not impact you unless you've implemented your own json_input_handler or json_output_handler, in which case you'll need to change your json_input_handler function signatures

  void do_longlong_value(long long value, const basic_parsing_context& context) override
  void do_ulonglong_integer_value(unsigned long long value, const basic_parsing_context& context) override

to

void do_integer_value(int64_t value, const basic_parsing_context<Char>& context) override
void do_uinteger_value(uint64_t value, const basic_parsing_context<Char>& context) override

and your json_output_handler function signatures from

void do_longlong_value(long long value) override
void do_ulonglong_integer_value(unsigned long long value) override

to

void do_integer_value(int64_t value) override
void do_uinteger_value(uint64_t value) override

• output_format drops support for floatfield property

Non-beaking API Changes

• remove_range has been deprecated, use erase(array_iterator first, array_iterator last) instead
• remove has been deprecated, use erase(const std::string& name ) instead
• json::parse_string has been renamed to json::parse, parse_string is deprecated but still works
• json member functionis_emptyhas been renamed toempty,is_empty` is deprecated but still works. Rationale: consistency with C++ containers
• json member functions begin_elements and end_elements have been deprecated, instead use elements().begin() and elements.end()
• json member functions begin_members and end_members have been deprecated, instead use members().begin() and members.end()
• json member function has_member has been deprecated, instead use count. Rationale: consistency with C++ containers
• json member function remove_member has been deprecated, instead use remove. Rationale: only member function left with _element or _member suffix
• json_parse_exception renamed to parse_exception, json_parse_exception typedef to parse_exception
• json::parse(std::istream& is) renamed to json::parse_stream. json::parse(std::istream is) is deprecated but still works.

0.98.2 Release

• json constructor is now templated, so constructors now accept extended types
• Following RFC7159, json_parser now accepts any JSON value, removing the constraint that it be an object or array.
• The member json_type_traits member functions is, as, and assign have been changed to static functions. if you have implemented your own type specializations, you will also have to change your is, as and assign functions to be static.
• Removed json deprecated functions custom_data, set_custom_data, add_custom_data
• json_reader member function max_depth has been renamed to max_nesting_depth, the former name is still supported.
• json member function resize_array has been renamed to resize, the former name is still supported.

jsoncons supports alternative ways for constructing null, object, and array values.

null:

json a = jsoncons::null_type();  // Using type constructor
json b = json::null_type();      // Using alias
json c(json::null);              // From static data member prototype

object:

json a();                 // Default is empty object
json b = json::object();  // Using type constructor
json c(json::an_object);  // From static data member prototype

array:

json a = json::array();      // Using type constructor
json b = json::make_array(); // Using factory method
json c(json::an_array);      // From static data member prototype

Since C++ has possible order issues with static data members, the jsoncons examples and documentation have been changed to consistently use the other ways, and json::null, json::an_object and json::an_array have been, while still usable, deprecated.

0.98.1 Release

• Enhances parser for CSV files that outputs JSON, see example below.
• Adds get_result member function to json_deserializer, which returns the json value v stored in a json_deserializer as std::move(v). The root() member function has been deprecated but is still supported.
• Adds is_valid member function to json_deserializer
• Enhances json::any class, adds type checks when casting back to original value
• Fixes some warning messages

0.98 Release

Bug fixes:

• Fixes the noexcept specification (required for Visual Studio 2015 and later.) Fix contributed by Rupert Steel.
• Fixes bug with proxy operator== when comparing object member values, such as in val["field"] == json("abc")

Enhancements:

• Refines error codes and improves error messages

• Renames json_reader method read to read_next, reflecting the fact that it supports reading a sequence of JSON texts from a stream. The
  former name is deprecated but still works.

• Adds json_reader method check_done that throws if there are unconsumed non-whitespace characters after one or more calls to read_next.

• Adds getter and setter max_depth methods to allow setting the maximum JSON parse tree depth if desired, by default
  it is arbitrarily large (limited by heap memory.)

• Modifies json static methods parse_string, parse_file, and parse_stream behaviour to throw if there are unconsumed non-whitespace characters after reading one JSON text.

Changes to extensions:

• Changes the top level namespace for the extensions from jsoncons_ext to jsoncons, e.g. jsoncons_ext::csv::csv_reader becomes jsoncons::csv::csv_reader
• Modifies csv_reader and csv_serializer so that the constructors are passed parameters in a csv_parameters object rather than a json object.
• Adds more options to csv_reader

0.97.2 Release

• Incorporates test suite files from http://www.json.org/JSON_checker/ into test suite
• The jsoncons parser accepts all of the JSON_checker files that its supposed to accept.
• Failures to reject incorrect exponential notation (e.g. [0e+-1]) have been fixed.

• The jsoncons parser now rejects all of the JSON_checker files that its supposed to reject except ones with stuff after the end of the document, e.g.

  ["Extra close"]]

  (Currently the jsoncons parser stops after reading a complete JSON text, and supports reading a sequence of JSON texts.)

• Incorporates a fix to operator== on json objects, contributed by Alex Merry

0.97.1 Release

• "Transforming JSON with filters" example fixed
• Added a class-specific in-place new to the json class that is implemented in terms of the global version (required to create json objects with placement new operator.)
• Reorganized header files, removing unnecessary includes.
• Incorporates validation contributed by Alex Merry for ensuring that there is an object or array on parse head.
• Incorporates fix contributed by Milan Burda for “Switch case is in protected scope” clang build error

0.97 Release

• Reversion of 0.96 change:

The virtual methods do_float_value, do_integer_value, and do_unsigned_value of json_input_handler and json_outputhandler have been restored to do_double_value, do_longlong_value and do_ulonglong_value, and their typedefed parameter types float_type, integer_type, and unsigned_type have been restored to double, long long, and unsigned long long.

The rationale for this reversion is that the change doesn't really help to make the software more flexible, and that it's better to leave out the typedefs. There will be future enhancements to support greater numeric precision, but these will not affect the current method signatures.

• Fix for "unused variable" warning message

0.96 Release

This release includes breaking changes to interfaces. Going forward, the interfaces are expected to be stable.

Breaking changes:

• Renamed error_handler to parse_error_handler.

• Renamed namespace json_parser_error to json_parser_errc

• Renamed value_adapter to json_type_traits, if you have implemented your own type specializations,
  you will have to rename value_adapter also.

• Only json arrays now support operator[](size_t) to loop over values, this is no longer supported for json objects. Use a json object iterator instead.

• The virtual methods do_double_value, do_integer_value and do_uinteger_value of json_input_handler and json_outputhandler have been renamed to do_float_value, do_integer_value, and do_unsigned_value,
  and their parameters have been changed from double, long long, and unsigned long long to typedefs float_type, integer_type, and unsigned_type.
  The rationale for this change is to allow different configurations for internal number types (reversed in 0.97.)

General changes

• json member function begin_object now returns a bidirectional iterator rather than a random access iterator.

• Static singleton instance methods have been added to default_parse_error_handler
  and empty_json_input_handler.

• Added to the json class overloaded static methods parse, parse_string
  and parse_file that take a parse_error_handler as a parameter.

• Added methods last_char() and eof() to parsing_context.

• Enhancements to json parsing and json parse event error notification.

• Added to json_input_handler and json_output_handler a non virtual method value that takes a null terminated string.

• Added methods is_integer, is_unsigned and is_float to json to replace is_longlong, is_ulonglong and is_double, which have been deprecated.

• Added methods as_integer, as_unsigned and as_float to json to replace is_longlong, is_ulonglong and is_double, which have been deprecated.

Bug fixes:

• Fixed issue with column number reported by json_reader

• Where &s[0] and s.length() were passed to methods, &s[0] has been replaced with s.c_str().
  While this shouldn't be an issue on most implementations, VS throws an exception in debug modes when the string has length zero.

• Fixes two issues in 0.95 reported by Alex Merry that caused errors with GCC: a superfluous typename has been removed in csv_serializer.hpp, and a JSONCONS_NOEXCEPT specifier has been added to the json_error_category_impl name method.

• Fixed a number of typename issues in the 0.96 candidate identifed by Ignatov Serguei.

• Fixes issues with testsuite cmake and scons reported by Alex Merry and Ignatov Serguei

0.95

Enhancements:

• Added template method any_cast to json class.

• The allocator type parameter in basic_json is now supported, it allows you to supply a
  custom allocator for dynamically allocated, fixed size small objects in the json container.
  The allocator type is not used for structures including vectors and strings that use large
  or variable amounts of memory, these always use the default allocators.

Non-breaking Change:

• json_filter method parent has been renamed to input_handler (old name still works)

Breaking change (if you've implemented your own input and output handlers, or if you've
passed json events to input and output handlers directly):

• The input handler virtual method
  name(const std::string& name, const parsing_context& context)
  has been changed to
  do_name(const char* p, size_t length, const parsing_context& context)

• The output handler virtual method
  name(const std::string& name)
  has been changed to
  do_name(const char* p, size_t length)

• The input handler virtual method
  string_value(const std::string& value, const parsing_context& context)
  has been changed to
  do_string_value(const char* p, size_t length, const parsing_context& context)

• The output handler virtual method
  string_value(const std::string& value)
  has been changed to
  do_string_value(const char* p, size_t length)

The rationale for the method parameter changes is to allow different internal
representations of strings but preserve efficiency.

• The input and output handler virtual implementation methods begin_json, end_json,
  begin_object, end_object, begin_array, end_array, name, string_value,
  longlong_value, ulonglong_value, double_value, bool_value and null_value
  have been renamed to do_begin_json, do_end_json, do_begin_object, do_end_object,
  do_begin_array, do_end_array, do_name, do_string_value, do_longlong_value,
  do_ulonglong_value, do_double_value, do_bool_value and do_null_value and have been
  made private.

• Public non-virtual interface methods begin_json, end_json,
  begin_object, end_object, begin_array, end_array, name
  have been added to json_input_handler and json_output_handler.

The rationale for these changes is to follow best C++ practices by making the
json_input_handler and json_output_handler interfaces public non-virtual and
the implementations private virtual. Refer to the documentation and tutorials for details.

• The error_handler virtual implementation methods have been renamed to do_warning and do_error, and made private. Non virtual public interface methods warning and error have been added. Error handling now leverages std::error_code to communicate parser error events in an extendable way.

Bug fixes:

• Fixed bug in csv_reader

0.94.1

Bug fixes:

• Incorporates fix from Alex Merry for comparison of json objects

0.94

Bug fixes

• Incorporates contributions from Cory Fields for silencing some compiler warnings
• Fixes bug reported by Vitaliy Gusev in json object operator[size_t]
• Fixes bug in json is_empty method for empty objects

Changes

• json constructors that take string, double etc. are now declared explicit (assignments and defaults to get and make_array methods have their own implementation and do not depend on implicit constructors.)
• make_multi_array renamed to make_array (old name is still supported)
• Previous versions supported any type values through special methods set_custom_data, add_custom_data, and custom_data. This version introduces a new type json::any that wraps any values and works with the usual accessors set, add and as, so the specialized methods are no longer required.

Enhancements

• json get method with default value now accepts extended types as defaults
• json make_array method with default value now accepts extended types as defaults

New extensions

• Added jsoncons_ext/boost/type_extensions.hpp to collect extensions traits for boost types, in particular, for boost::gregorian dates.

0.93 Release

New features

• Supports wide character strings and streams with wjson, wjson_reader etc. Assumes UTF16 encoding if sizeof(wchar_t)=2 and UTF32 encoding if sizeof(wchar_t)=4.
• The empty class null_type is added to the jsoncons namespace, it replaces the member type json::null_type (json::null_type is typedefed to jsoncons::null_type for backward compatibility.)

Defect fixes:

• The ascii character 0x7f (del) was not being considered a control character to be escaped, this is fixed.
• Fixed two issues with serialization when the output format property escape_all_non_ascii is enabled. One, the individual bytes were being checked if they were non ascii, rather than first converting to a codepoint. Two, continuations weren't being handled when decoding.

0.92a Release

Includes contributed updates for valid compilation and execution in gcc and clang environments

0.92 Release

Breaking change (but only if you have subclassed json_input_handler or json_output_handler)

• For consistency with other names, the input and output handler methods new to 0.91 - value_string, value_double, value_longlong, value_ulonglong and value_bool - have been renamed to string_value, double_value, longlong_value, ulonglong_value and bool_value.

Non breaking changes (previous features are deprecated but still work)

• name_value_pair has been renamed to member_type (typedefed to previous name.)

• as_string(output_format format) has been deprecated, use the existing to_string(output_format format) instead

Enhancements:

• json now has extensibilty, you can access and modify json values with new types, see the tutorial Extensibility

Preparation for allocator support:

• The basic_json and related classes now have an Storage template parameter, which is currently just a placeholder, but will later provide a hook to allow users to control how json storage is allocated. This addition is transparent to users of the json and related classes.

0.91 Release

This release should be largely backwards compatible with 0.90 and 0.83 with two exceptions:

1. If you have used object iterators, you will need to replace uses of std::pair with name_value_pair, in particular, first becomes name() and second becomes value().

2. If you have subclassed json_input_handler, json_output_handler, or json_filter, and have implemented value(const std::string& ..., value(double ..., etc., you will need to modify the names to value_string(const std::string& ..., value_double(double ... (no changes if you are feeding existing implementations.)

The changes are

• Replaced std::pairstd::string,json with name_value_pair that has accessors name() and value()

• In json_input_handler and json_output_handler, allowed for overrides of the value methods by making them non-virtual and adding virtual methods value_string, value_double, value_longlong, value_ulonglong, and value_bool

Other new features:

• Changed implementation of is and as, the current implementation should be user extensible

• make_multi_array makes a multidimensional array with the number of dimensions specified as a template parameter. Replaces make_2d_array and make_3d_array, which are now deprecated.

• Added support for isstd::vector<T> and asstd::vector<T>

• Removed JSONCONS_NO_CXX11_RVALUE_REFERENCES, compiler must support move semantics

Incorporates a number of contributions from Pedro Larroy and the developers of the clearskies_core project:

• build system for posix systems
• GCC to list of supported compilers
• Android fix
• fixed virtual destructors missing in json_input_handler, json_output_handler and parsing_context
• fixed const_iterator should be iterator in json_object implementation

To clean up the interface and avoid too much duplicated functionality, we've deprecated some json methods (but they still work)

make_array

Use json val(json::an_array) or json::make_multi_array<1>(...) instead (but make_array will continue to work)

make_2d_array
make_3d_array

Use make_multi_array<2> and make_multi_array<3> instead

as_vector

Use asstd::vector<int> etc. instead

as_int
as_uint
as_char

Use as, as, and as instead

Release 0.90a

Fixed issue affecting clang compile

Release 0.90

This release should be fully backwards compatible with 0.83.

Includes performance enhancements to json_reader and json_deserializer

Fixes issues with column numbers reported with exceptions

Incorporates a number of patches contributed by Marc Chevrier:

• Fixed issue with set member on json object when a member with that name already exists
• clang port
• cmake build files for examples and test suite
• json template method is for examining the types of json values
• json template method as for accessing json values

Release 0.83

Optimizations (very unlikely to break earlier code)

• get(const std::name& name) const now returns const json& if keyed value exists, otherwise a const reference to json::null

• get(const std::string& name, const json& default_val) const now returns const json (actually a const proxy that evaluates to json if read)

Bug fixes

• Line number not incremented within multiline comment - fixed

Deprecated features removed

• Removed deprecated output_format properties (too much bagage to carry around)

Release 0.82a

• The const version of the json operator didn't need to return a proxy, the return value has been changed to const json& (this change is transparent to the user.)

• get(const std::name& name) has been changed to return a copy (rather than a reference), and json::null if there is no member with that name (rather than throw.) This way both get methods return default values if no keyed value exists.

• non-const and const methods json& at(const std::name& name) have been added to replace the old single argument get method. These have the same behavior as the corresponding operator[] functions, but the non-const at is more efficient.

Release 0.81

• Added accessor and modifier methods floatfield to output_format to provide a supported way to set the floatfield format flag to fixed or scientific with a specified number of decimal places (this can be done in older versions, but only with deprecated methods.)

• The default constructor now constructs an empty object (rather than a null object.) While this is a change, it's unlikely to break exisitng code (all test cases passed without modification.)

This means that instead of

json obj(json::an_object);
obj["field"] = "field";

you can simply write

json obj;
obj["field"] = "field";

The former notation is still supported, though.

• Added a version of 'resize_array' to json that resizes the array to n elements and initializes them to a specified value.

• Added a version of the static method json::make_array that takes no arguments and makes an empty json array

Note that

json arr(json::an_array);

is equivalent to

json arr = json::make_array();

and

json arr(json::an_array);
arr.resize_array(10,0.0);

is equivalent to

json arr = json::make_array(10,0.0);

For consistency the json::make_array notation is now favored in the documentation.

Release 0.71

• Added resize_array method to json for resizing json arrays

• Fixed issue with remove_range method (templated code failed to compile if calling this method.)

• Added remove_member method to remove a member from a json object

• Fixed issue with multiline line comments, added test case

• Fixed issue with adding custom data to a json array using add_custom_data, added examples.

Release 0.70

• Since 0.50, jsoncons has used snprintf for default serialization of double values to string values. This can result in invalid json output when running on a locale like German or Spanish. The period character (‘.’) is now always used as the decimal point, non English locales are ignored.

• The output_format methods that support alternative floating point formatting, e.g. fixed, have been deprecated.

• Added a template method as_vector to the json class. If a json value is an array and conversion is possible to the template type, returns a std::vector of that type, otherwise throws an std::exception. Specializations are provided for std::string, bool, char, int, unsigned int, long, unsigned long, long long, unsigned long long, and double. For example

  std::string s("[0,1,2,3]");

  json val = json::parse_string(s);

  std::vector v = val.as_vector();

• Undeprecated the json member function precision

Release 0.60b

This release (0.60b) is fully backwards compatible with 0.50.

A change introduced with 0.60 has been reversed. 0.60 introduced an alternative method of constructing a json arrray or object with an initial default constructor, a bug with this was fixed in 0.60a, but this feature and related documentation has been removed because it added complexity but no real value.

Enhancements

• Added swap member function to json

• Added add and add_custom_data overrides to json that take an index value, for adding a new element at the specified index and shifting all elements currently at or above that index to the right.

• Added capacity member functions to json

0.60 extensions

• csv_serializer has been added to the csv extension

Release 0.50

This release is fully backwards compatible with 0.4*, and mostly backwards compatible to 0.32 apart from the two name changes in 0.41

Bug fixes

• When reading the escaped characters "\b", "\f", "\r" and "\t" appearing in json strings, json_reader was replacing them with the linefeed character, this has been fixed.

Deprecated

• Deprecated modifiers precision and fixed_decimal_places from output_format. Use set_floating_point_format instead.
• Deprecated constructor that takes indenting parameter from output_format. For pretty printing with indenting, use the pretty_print function or pass the indenting parameter in json_serializer.

Changes

• When serializing floating point values to a stream, previous versions defaulted to default floating point precision with a precision of 16. This has been changed to truncate trailing zeros but keep one if immediately after a decimal point.

New features

• For line reporting in parser error messages, json_reader now recognizes \r\n, \n alone or \r alone (\r alone is new.)
• Added set_floating_point_format methods to output_format to give more control over floating point notation.

Non functional enhancements

• json_reader now estimates the minimum capacity for arrays and objects, and reports that information for the begin_array and begin_object events. This greatly reduces reallocations.

Release 0.42

• Fixed another bug with multi line /**/ comments
• Minor fixes to reporting line and column number of errors
• Added fixed_decimal_places setter to output_format
• Added version of as_string to json that takes output_format as a parameter
• Reorganization of test cases and examples in source tree

Release 0.41

• Added non-member overload swap(json& a, json& b)
• Fixed bug with multi line /**/ comments
• Added begin_json and end_json methods to json_output_handler
• json_deserializer should now satisfy basic exception safety (no leak guarantee)
• Moved csv_reader.hpp to jsoncons_ext/csv directory
• Changed csv_reader namespace to jsoncons::csv
• json::parse_file no longer reads the entire file into memory before parsing (it now uses json_reader default buffering)

Release 0.40

• json_listener renamed to json_input_handler

• json_writer renamed to json_output_handler

• Added json_filter class

• json get method that takes default argument now returns a value rather than a reference

• Issue in csv_reader related to get method issue fixed
• Issue with const json operator[] fixed
• Added as_char method to json
• Improved exception safety, some opportunites for memory leaks in the presence of exceptions removed

Release 0.33

Added reserve method to json

Added static make_3d_array method to json

json_reader now configured for buffered reading

Added csv_reader class for reading CSV files and producing JSON events

Fixed bug with explicitly passing output_format in pretty_print.

Release 0.32

Added remove_range method, operator== and operator!= to proxy and json objects

Added static methods make_array and make_2d_array to json

Release 0.31

error_handler method content_error renamed to error

Added error_code to warning, error and fatal_error methods of error_handler

Release 0.30

json_in_stream renamed to json_listener

json_out_stream renamed to json_writer

Added buffer accessor method to parsing_context

Release 0.20

Added parsing_context class for providing information about the
element being parsed.

error_handler methods take message and parsing_context parameters

json_in_stream handlers take parsing_context parameter

Release 0.19

Added error_handler class for json_reader

Made json_exception a base class for all json exceptions

Added root() method to json_deserializer to get a reference to the json value

Removed swap_root() method from json_deserializer

Release 0.18

Renamed serialize() class method to to_stream() in json

Custom data serialization supported through template function specialization of serialize
(reverses change in 0.17)

Release 0.17

Added is_custom() method to json and proxy

get_custom() method renamed to custom_data() in json and proxy

Added clear() method to json and proxy

set_member() method renamed to set()

set_custom() method renamed to set_custom_data()

push_back() method renamed to add() in json and proxy

Added add_custom_data method() in json and proxy

Custom data serialization supported through template class specialization of custom_serialization
(replaces template function specialization of serialize)

Release 0.16

Change to json_out_stream and json_serializer:

void value(const custom_data& value)

removed.

Free function serialize replaces free function to_stream for
serializing custom data.

pretty print tidied up for nested arrays

Release 0.15

Made eof() method on json_reader public, to support reading
multiple JSON objects from a stream.

Release 0.14

Added pretty_print class

Renamed json_stream_writer to json_serializer,
implements pure virtual class json_out_stream

Renamed json_stream_listener to json_deserializer
implements pure virtual class json_in_stream

Renamed json_parser to json_reader, parse to read.

Changed indenting so object and array members start on new line.

Added support for storing user data in json object, and
serializing to JSON.

Release 0.13

Replaced simple_string union member with json_string that
wraps std::basic_string

name() and value() event handler methods on
basic_json_stream_writer take const std::basic_string&
rather than const Char* and length.

Release 0.12

Implemented operator<< for json::proxy

Added to_stream methods to json::proxy

Release 0.11

Added members to json_parser to access and modify the buffer capacity

Added checks when parsing integer values to determine overflow for
long long and unsigned long long, and if overflow, parse them as
doubles.