JSONCONS

jsoncons is a C++, header-only library for constructing JSON and JSON-like data formats such as CBOR. For each supported data format, it enables you to work with the data in a number of ways:

• As a variant-like data structure, basic_json

• As a strongly typed C++ data structure that implements json_type_traits

• With cursor-level access to a stream of parse events, somewhat analogous to StAX pull parsing and push serializing in the XML world.

Compared to other JSON libraries, jsoncons has been designed to handle very large JSON texts. At its heart are SAX-style parsers and serializers. It supports reading an entire JSON text in memory in a variant-like structure. But it also supports efficient access to the underlying data using StAX-style pull parsing and push serializing. And it supports incremental parsing into a user's preferred form, using information about user types provided by specializations of json_type_traits.

The jsoncons data model supports the familiar JSON types - nulls, booleans, numbers, strings, arrays, objects - plus byte strings. In addition, jsoncons supports semantic tagging of datetimes, epoch times, big integers, big decimals, big floats and binary encodings. This allows it to preserve these type semantics when parsing JSON-like data formats such as CBOR that have them.

jsoncons is distributed under the Boost Software License.

jsoncons is free but welcomes support to sustain its development. If you find this library helpful, please consider making a one time donation or becoming a ❤️ sponsor.

Extensions

• bson implements decode from and encode to the Binary JSON data format.
• cbor implements decode from and encode to the IETF standard Concise Binary Object Representation data format. In addition it supports tags for stringref and tags for typed arrays.
• csv implements decode from and encode to CSV files.
• jmespath implements JMESPath, a query language for transforming JSON documents into other JSON documents.
• jsonpatch implements the IETF standard JavaScript Object Notation (JSON) Patch
• json_merge_patch implements the IETF standard JSON Merge Patch
• jsonpath implements Stefan Goessner's JSONPath. It also supports search and replace using JSONPath expressions.
• jsonpointer implements the IETF standard JavaScript Object Notation (JSON) Pointer
• jsonschema implements the JSON Schema Draft 7 specification for validating input JSON. (since 0.160.0)
• msgpack implements decode from and encode to the MessagePack data format.
• ubjson implements decode from and encode to the Universal Binary JSON Specification data format.

What users say

"I am so happy I have come across your json c++ library!"

"I finally decided on jsoncons after evaluating quite a few json libraries out there."

"I’m using your library for an external interface to pass data, as well as using the conversions from csv to json, which are really helpful for converting data for use in javacode"

"Verified that, for my needs in JSON and CBOR, it is working perfectly"

"the JSONPath feature of this library, it's great"

"We use JMESPath implementation quite extensively"

"the serialization lib of choice with its beautiful mappings and ease of use"

"really good" "awesome project" "very solid and very dependable" "my team loves it" "Your repo rocks!!!!!"

Get jsoncons

You can use the vcpkg platform library manager to install the jsoncons package.

Or, download the latest release and unpack the zip file. Copy the directory include/jsoncons to your include directory. If you wish to use extensions, copy include/jsoncons_ext as well.

Or, download the latest code on master.

How to use it

• Quick guide
• Examples
• Reference
• Ask questions and suggest ideas for new features

The library requires a C++ Compiler with C++11 support. In addition the library defines jsoncons::endian, jsoncons::basic_string_view, jsoncons::optional, and jsoncons::span, which will be typedefed to their standard library equivalents if detected. Otherwise they will be typedefed to internal, C++11 compatible, implementations.

The library uses exceptions and in some cases std::error_code's to report errors. Apart from jsoncons::assertion_error, all jsoncons exception classes implement the jsoncons::json_error interface. If exceptions are disabled or if the compile time macro JSONCONS_NO_EXCEPTIONS is defined, throws become calls to std::terminate.

Benchmarks

json_benchmarks provides some measurements about how jsoncons compares to other json libraries.

• JSONTestSuite and JSON_checker test suites

• Performance benchmarks with text and integers

• Performance benchmarks with text and doubles

JSONPath Comparison shows how jsoncons JsonPath compares with other implementations

Examples

Working with JSON data

Working with CBOR data

Working with JSON data

For the examples below you need to include some header files and initialize a string of JSON data:

#include <jsoncons/json.hpp>
#include <jsoncons_ext/jsonpath/jsonpath.hpp>
#include <iostream>

using namespace jsoncons; // for convenience

std::string data = R"(
    {
       "application": "hiking",
       "reputons": [
       {
           "rater": "HikingAsylum",
           "assertion": "advanced",
           "rated": "Marilyn C",
           "rating": 0.90,
           "generated": 1514862245
         }
       ]
    }
)";

jsoncons allows you to work with the data in a number of ways:

• As a variant-like data structure, basic_json

• As a strongly typed C++ data structure that implements json_type_traits

• With cursor-level access to a stream of parse events

As a variant-like data structure

int main()
{
    // Parse the string of data into a json value
    json j = json::parse(data);

    // Does object member reputons exist?
    std::cout << "(1) " << std::boolalpha << j.contains("reputons") << "\n\n";

    // Get a reference to reputons array 
    const json& v = j["reputons"]; 

    // Iterate over reputons array 
    std::cout << "(2)\n";
    for (const auto& item : v.array_range())
    {
        // Access rated as string and rating as double
        std::cout << item["rated"].as<std::string>() << ", " << item["rating"].as<double>() << "\n";
    }
    std::cout << "\n";

    // Select all "rated" with JSONPath
    std::cout << "(3)\n";
    json result = jsonpath::json_query(j,"$..rated");
    std::cout << pretty_print(result) << "\n\n";

    // Serialize back to JSON
    std::cout << "(4)\n" << pretty_print(j) << "\n\n";
}

Output:

(1) true

(2)
Marilyn C, 0.9

(3)
[
    "Marilyn C"
]

(4)
{
    "application": "hiking",
    "reputons": [
        {
            "assertion": "advanced",
            "generated": 1514862245,
            "rated": "Marilyn C",
            "rater": "HikingAsylum",
            "rating": 0.9
        }
    ]
}

As a strongly typed C++ data structure

jsoncons supports transforming JSON texts into C++ data structures. The functions decode_json and encode_json convert strings or streams of JSON data to C++ data structures and back. Decode and encode work for all C++ classes that have json_type_traits defined. jsoncons already supports many types in the standard library, and your own types will be supported too if you specialize json_type_traits in the jsoncons namespace.

namespace ns {
    enum class hiking_experience {beginner,intermediate,advanced};

    class hiking_reputon
    {
        std::string rater_;
        hiking_experience assertion_;
        std::string rated_;
        double rating_;
        std::optional<std::chrono::seconds> generated_; // assumes C++17, if not use jsoncons::optional
        std::optional<std::chrono::seconds> expires_;
    public:
        hiking_reputon(const std::string& rater,
                       hiking_experience assertion,
                       const std::string& rated,
                       double rating,
                       const std::optional<std::chrono::seconds>& generated = 
                           std::optional<std::chrono::seconds>(),
                       const std::optional<std::chrono::seconds>& expires = 
                           std::optional<std::chrono::seconds>())
            : rater_(rater), assertion_(assertion), rated_(rated), rating_(rating),
              generated_(generated), expires_(expires)
        {
        }

        const std::string& rater() const {return rater_;}
        hiking_experience assertion() const {return assertion_;}
        const std::string& rated() const {return rated_;}
        double rating() const {return rating_;}
        std::optional<std::chrono::seconds> generated() const {return generated_;}
        std::optional<std::chrono::seconds> expires() const {return expires_;}

        friend bool operator==(const hiking_reputon& lhs, const hiking_reputon& rhs)
        {
            return lhs.rater_ == rhs.rater_ && lhs.assertion_ == rhs.assertion_ && 
                   lhs.rated_ == rhs.rated_ && lhs.rating_ == rhs.rating_ &&
                   lhs.confidence_ == rhs.confidence_ && lhs.expires_ == rhs.expires_;
        }

        friend bool operator!=(const hiking_reputon& lhs, const hiking_reputon& rhs)
        {
            return !(lhs == rhs);
        };
    };

    class hiking_reputation
    {
        std::string application_;
        std::vector<hiking_reputon> reputons_;
    public:
        hiking_reputation(const std::string& application, 
                          const std::vector<hiking_reputon>& reputons)
            : application_(application), 
              reputons_(reputons)
        {}

        const std::string& application() const { return application_;}
        const std::vector<hiking_reputon>& reputons() const { return reputons_;}
    };

} // namespace ns

// Declare the traits. Specify which data members need to be serialized.

JSONCONS_ENUM_TRAITS(ns::hiking_experience, beginner, intermediate, advanced)
// First four members listed are mandatory, generated and expires are optional
JSONCONS_N_CTOR_GETTER_TRAITS(ns::hiking_reputon, 4, rater, assertion, rated, rating, 
                              generated, expires)

// All members are mandatory
JSONCONS_ALL_CTOR_GETTER_TRAITS(ns::hiking_reputation, application, reputons)

int main()
{
    // Decode the string of data into a c++ structure
    ns::hiking_reputation v = decode_json<ns::hiking_reputation>(data);

    // Iterate over reputons array value
    std::cout << "(1)\n";
    for (const auto& item : v.reputons())
    {
        std::cout << item.rated() << ", " << item.rating();
        if (item.generated())
        {
            std::cout << ", " << (*item.generated()).count();
        }
        std::cout << "\n";
    }

    // Encode the c++ structure into a string
    std::string s;
    encode_json_pretty(v, s);
    std::cout << "(2)\n";
    std::cout << s << "\n";
}

Output:

(1)
Marilyn C, 0.9, 1514862245
(2)
{
    "application": "hiking",
    "reputons": [
        {
            "assertion": "advanced",
            "generated": 1514862245,
            "rated": "Marilyn C",
            "rater": "HikingAsylum",
            "rating": 0.9
        }
    ]
}

This example makes use of the convenience macros JSONCONS_ENUM_TRAITS, JSONCONS_N_CTOR_GETTER_TRAITS, and JSONCONS_ALL_CTOR_GETTER_TRAITS to specialize the json_type_traits for the enum type ns::hiking_experience, the class ns::hiking_reputon (with some non-mandatory members), and the class ns::hiking_reputation (with all mandatory members.) The macro JSONCONS_ENUM_TRAITS generates the code from the enum identifiers, and the macros JSONCONS_N_CTOR_GETTER_TRAITS and JSONCONS_ALL_CTOR_GETTER_TRAITS generate the code from the get functions and a constructor. These macro declarations must be placed outside any namespace blocks.

See examples for other ways of specializing json_type_traits.

With cursor-level access

A typical pull parsing application will repeatedly process the current() event and call next() to advance to the next event, until done() returns true.

int main()
{
    json_string_cursor cursor(data);
    for (; !cursor.done(); cursor.next())
    {
        const auto& event = cursor.current();
        switch (event.event_type())
        {
            case staj_event_type::begin_array:
                std::cout << event.event_type() << " " << "\n";
                break;
            case staj_event_type::end_array:
                std::cout << event.event_type() << " " << "\n";
                break;
            case staj_event_type::begin_object:
                std::cout << event.event_type() << " " << "\n";
                break;
            case staj_event_type::end_object:
                std::cout << event.event_type() << " " << "\n";
                break;
            case staj_event_type::key:
                // Or std::string_view, if supported
                std::cout << event.event_type() << ": " << event.get<jsoncons::string_view>() << "\n";
                break;
            case staj_event_type::string_value:
                // Or std::string_view, if supported
                std::cout << event.event_type() << ": " << event.get<jsoncons::string_view>() << "\n";
                break;
            case staj_event_type::null_value:
                std::cout << event.event_type() << "\n";
                break;
            case staj_event_type::bool_value:
                std::cout << event.event_type() << ": " << std::boolalpha << event.get<bool>() << "\n";
                break;
            case staj_event_type::int64_value:
                std::cout << event.event_type() << ": " << event.get<int64_t>() << "\n";
                break;
            case staj_event_type::uint64_value:
                std::cout << event.event_type() << ": " << event.get<uint64_t>() << "\n";
                break;
            case staj_event_type::double_value:
                std::cout << event.event_type() << ": " << event.get<double>() << "\n";
                break;
            default:
                std::cout << "Unhandled event type: " << event.event_type() << " " << "\n";
                break;
        }
    }    
}

Output:

begin_object
key: application
string_value: hiking
key: reputons
begin_array
begin_object
key: rater
string_value: HikingAsylum
key: assertion
string_value: advanced
key: rated
string_value: Marilyn C
key: rating
double_value: 0.9
key: generated
uint64_value: 1514862245
end_object
end_array
end_object

You can apply a filter to a cursor using the pipe syntax (e.g., cursor | filter1 | filter2 | ...)

int main()
{
    std::string name;
    auto filter = [&](const staj_event& ev, const ser_context&) -> bool
    {
        if (ev.event_type() == staj_event_type::key)
        {
            name = ev.get<std::string>();
            return false;
        }
        if (name == "rated")
        {
            name.clear();
            return true;
        }
        return false;
    };

    json_string_cursor cursor(data);
    auto filtered_c = cursor | filter;

    for (; !filtered_c.done(); filtered_c.next())
    {
        const auto& event = filtered_c.current();
        switch (event.event_type())
        {
            case staj_event_type::string_value:
                // Or std::string_view, if C++17
                std::cout << event.event_type() << ": " << event.get<jsoncons::string_view>() << "\n";
                break;
            default:
                std::cout << "Unhandled event type\n";
                break;
        }
    }
}    

Output:

Marilyn C

Working with CBOR data

For the examples below you need to include some header files and initialize a buffer of CBOR data:

#include <iomanip>
#include <iostream>
#include <jsoncons/json.hpp>
#include <jsoncons_ext/cbor/cbor.hpp>
#include <jsoncons_ext/jsonpath/jsonpath.hpp>

using namespace jsoncons; // for convenience

const std::vector<uint8_t> data = {
    0x9f, // Start indefinte length array
      0x83, // Array of length 3
        0x63, // String value of length 3
          0x66,0x6f,0x6f, // "foo" 
        0x44, // Byte string value of length 4
          0x50,0x75,0x73,0x73, // 'P''u''s''s'
        0xc5, // Tag 5 (bigfloat)
          0x82, // Array of length 2
            0x20, // -1
            0x03, // 3   
      0x83, // Another array of length 3
        0x63, // String value of length 3
          0x62,0x61,0x72, // "bar"
        0xd6, // Expected conversion to base64
        0x44, // Byte string value of length 4
          0x50,0x75,0x73,0x73, // 'P''u''s''s'
        0xc4, // Tag 4 (decimal fraction)
          0x82, // Array of length 2
            0x38, // Negative integer of length 1
              0x1c, // -29
            0xc2, // Tag 2 (positive bignum)
              0x4d, // Byte string value of length 13
                0x01,0x8e,0xe9,0x0f,0xf6,0xc3,0x73,0xe0,0xee,0x4e,0x3f,0x0a,0xd2,
    0xff // "break"
};

jsoncons allows you to work with the CBOR data similarly to JSON data:

• As a variant-like data structure, basic_json

• As a strongly typed C++ data structure that implements json_type_traits

• With cursor-level access to a stream of parse events

As a variant-like data structure

int main()
{
    // Parse the CBOR data into a json value
    json j = cbor::decode_cbor<json>(data);

    // Pretty print
    std::cout << "(1)\n" << pretty_print(j) << "\n\n";

    // Iterate over rows
    std::cout << "(2)\n";
    for (const auto& row : j.array_range())
    {
        std::cout << row[1].as<jsoncons::byte_string>() << " (" << row[1].tag() << ")\n";
    }
    std::cout << "\n";

    // Select the third column with JSONPath
    std::cout << "(3)\n";
    json result = jsonpath::json_query(j,"$[*][2]");
    std::cout << pretty_print(result) << "\n\n";

    // Serialize back to CBOR
    std::vector<uint8_t> buffer;
    cbor::encode_cbor(j, buffer);
    std::cout << "(4)\n" << byte_string_view(buffer) << "\n\n";
}

Output:

(1)
[
    ["foo", "UHVzcw", "0x3p-1"],
    ["bar", "UHVzcw==", "1.23456789012345678901234567890"]
]

(2)
50,75,73,73 (n/a)
50,75,73,73 (base64)

(3)
[
    "0x3p-1",
    "1.23456789012345678901234567890"
]

(4)
82,83,63,66,6f,6f,44,50,75,73,73,c5,82,20,03,83,63,62,61,72,d6,44,50,75,73,73,c4,82,38,1c,c2,4d,01,8e,e9,0f,f6,c3,73,e0,ee,4e,3f,0a,d2

As a strongly typed C++ data structure

int main()
{
    // Parse the string of data into a std::vector<std::tuple<std::string,jsoncons::byte_string,std::string>> value
    auto val = cbor::decode_cbor<std::vector<std::tuple<std::string,jsoncons::byte_string,std::string>>>(data);

    std::cout << "(1)\n";
    for (const auto& row : val)
    {
        std::cout << std::get<0>(row) << ", " << std::get<1>(row) << ", " << std::get<2>(row) << "\n";
    }
    std::cout << "\n";

    // Serialize back to CBOR
    std::vector<uint8_t> buffer;
    cbor::encode_cbor(val, buffer);
    std::cout << "(2)\n" << byte_string_view(buffer) << "\n\n";
}

Output:

(1)
foo, 50,75,73,73, 0x3p-1
bar, 50,75,73,73, 1.23456789012345678901234567890

(2)
82,9f,63,66,6f,6f,44,50,75,73,73,66,30,78,33,70,2d,31,ff,9f,63,62,61,72,44,50,75,73,73,78,1f,31,2e,32,33,34,35,36,37,38,39,30,31,32,33,34,35,36,37,38,39,30,31,32,33,34,35,36,37,38,39,30,ff

Note that when decoding the bigfloat and decimal fraction into a std::string, we lose the semantic information that the variant like data structure preserved with a tag, so serializing back to CBOR produces a text string.

With cursor-level access

A typical pull parsing application will repeatedly process the current() event and call next() to advance to the next event, until done() returns true.

int main()
{
    cbor::cbor_bytes_cursor cursor(data);
    for (; !cursor.done(); cursor.next())
    {
        const auto& event = cursor.current();
        switch (event.event_type())
        {
            case staj_event_type::begin_array:
                std::cout << event.event_type() << " " << "(" << event.tag() << ")\n";
                break;
            case staj_event_type::end_array:
                std::cout << event.event_type() << " " << "(" << event.tag() << ")\n";
                break;
            case staj_event_type::begin_object:
                std::cout << event.event_type() << " " << "(" << event.tag() << ")\n";
                break;
            case staj_event_type::end_object:
                std::cout << event.event_type() << " " << "(" << event.tag() << ")\n";
                break;
            case staj_event_type::key:
                // Or std::string_view, if supported
                std::cout << event.event_type() << ": " << event.get<jsoncons::string_view>() << " " << "(" << event.tag() << ")\n";
                break;
            case staj_event_type::string_value:
                // Or std::string_view, if supported
                std::cout << event.event_type() << ": " << event.get<jsoncons::string_view>() << " " << "(" << event.tag() << ")\n";
                break;
            case staj_event_type::byte_string_value:
                std::cout << event.event_type() << ": " << event.get<jsoncons::span<const uint8_t>>() << " " << "(" << event.tag() << ")\n";
                break;
            case staj_event_type::null_value:
                std::cout << event.event_type() << " " << "(" << event.tag() << ")\n";
                break;
            case staj_event_type::bool_value:
                std::cout << event.event_type() << ": " << std::boolalpha << event.get<bool>() << " " << "(" << event.tag() << ")\n";
                break;
            case staj_event_type::int64_value:
                std::cout << event.event_type() << ": " << event.get<int64_t>() << " " << "(" << event.tag() << ")\n";
                break;
            case staj_event_type::uint64_value:
                std::cout << event.event_type() << ": " << event.get<uint64_t>() << " " << "(" << event.tag() << ")\n";
                break;
            case staj_event_type::half_value:
            case staj_event_type::double_value:
                std::cout << event.event_type() << ": "  << event.get<double>() << " " << "(" << event.tag() << ")\n";
                break;
            default:
                std::cout << "Unhandled event type " << event.event_type() << " " << "(" << event.tag() << ")\n";
                break;
        }
    }
}

Output:

begin_array (n/a)
begin_array (n/a)
string_value: foo (n/a)
byte_string_value: 50,75,73,73 (n/a)
string_value: 0x3p-1 (bigfloat)
end_array (n/a)
begin_array (n/a)
string_value: bar (n/a)
byte_string_value: 50,75,73,73 (base64)
string_value: 1.23456789012345678901234567890 (bigdec)
end_array (n/a)
end_array (n/a)

You can apply a filter to a cursor using the pipe syntax,

int main()
{
    auto filter = [&](const staj_event& ev, const ser_context&) -> bool
    {
        return (ev.tag() == semantic_tag::bigdec) || (ev.tag() == semantic_tag::bigfloat);  
    };

    cbor::cbor_bytes_cursor cursor(data);
    auto filtered_c = cursor | filter;

    for (; !filtered_c.done(); filtered_c.next())
    {
        const auto& event = filtered_c.current();
        switch (event.event_type())
        {
            case staj_event_type::string_value:
                // Or std::string_view, if supported
                std::cout << event.event_type() << ": " << event.get<jsoncons::string_view>() << " " << "(" << event.tag() << ")\n";
                break;
            default:
                std::cout << "Unhandled event type " << event.event_type() << " " << "(" << event.tag() << ")\n";
                break;
        }
    }
}

Output:

string_value: 0x3p-1 (bigfloat)
string_value: 1.23456789012345678901234567890 (bigdec)

Supported compilers

jsoncons requires a compiler with C++11 support. It is tested in continuous integration on Github Actions, AppVeyor, Travis, and doozer. UndefinedBehaviorSanitizer (UBSan) diagnostics are enabled for selected gcc and clang builds. Since v0.151.0, it is integrated with Google OSS-fuzz, with coverage for all parsers and encoders.

Compiler	Version	Architecture	Operating System	Notes
Microsoft Visual Studio	vs2015 (MSVC 19.0.24241.7)	x86,x64	Windows 10
	vs2017	x86,x64	Windows 10
	vs2019	x86,x64	Windows 10
g++	4.8 and above	x64	Ubuntu	std::regex isn't fully implemented in 4.8, so jsoncons::jsonpath regular expression filters aren't supported in 4.8
	4.8.5	x64	CentOS 7.6	std::regex isn't fully implemented in 4.8, so jsoncons::jsonpath regular expression filters aren't supported in 4.8
	6.3.1 (Red Hat 6.3.1-1)	x64	Fedora release 24
	4.9.2	i386	Debian 8
clang	4.0, 5.0, 6.0, 7, 8, 9, 10	x64	Ubuntu
clang xcode	9.3, 9.4, 10, 10.1, 10.2, 11.2, 12	x64	OSX

Building the test suite and examples with CMake

CMake is a cross-platform build tool that generates makefiles and solutions for the compiler environment of your choice. On Windows you can download a Windows Installer package. On Linux it is usually available as a package, e.g., on Ubuntu,

sudo apt-get install cmake

Once cmake is installed, you can build and run the unit tests from the jsoncons directory,

On Windows:

> mkdir build
> cd build
> cmake .. -DJSONCONS_BUILD_TESTS=On
> cmake --build .
> ctest -C Debug --output-on-failure

On UNIX:

$ mkdir build
$ cd build
$ cmake .. -DJSONCONS_BUILD_TESTS=On
$ cmake --build .
$ ctest --output-on-failure

Acknowledgements

jsoncons uses the PVS-Studio static analyzer, provided free for open source projects.

A big thanks to the comp.lang.c++ community for help with implementation details.

The jsoncons platform dependent binary configuration draws on to the excellent MIT licensed tinycbor.

Thanks to Milo Yip, author of RapidJSON, for raising the quality of JSON libraries across the board, by publishing the benchmarks, and contacting this project (among others) to share the results.

The jsoncons implementation of the Grisu3 algorithm for printing floating-point numbers follows Florian Loitsch's MIT licensed grisu3_59_56 implementation, with minor modifications.

The macro JSONCONS_ALL_MEMBER_TRAITS follows the approach taken by Martin York's ThorsSerializer

The jsoncons implementations of BSON decimal128 to and from string, and ObjectId to and from string, are based on the Apache 2 licensed libbson.

Special thanks to our contributors