wqking/jsonpp

C++ JSON parser and stringify

jsonpp -- easy to use and reusable C++ JSON library

• Features
• Why jsonpp when there are hundreds of JSON libraries?
• Basic information
  • License
  • Version 1.0.0-RC
  • Source code
  • Supported compilers
• Quick start
  • Namespace
• Example code
  • Parse JSON document
  • Parse JSON document as specified type
  • Dump JSON object to text (stringify)
  • Dump/parse class object
• Performance
  • Parse JSON file
  • Stringify JSON document
• Documentations
• Known compiler related quirks in MSVC

jsonpp is a cross platform C++ JSON library, focusing on easy to use and reusable technology.
jsonpp can parse and stringify JSON document.

Features

• Reusable technology. The core technology is from C++ reflection library metapp developed by the same developer (wqking) of jsonpp. The technology is not unique to jsonpp, it's reusable for other purposes such as runtime reflection system, serialization, script binding, etc.

• Very easy to use. We can parse and stringify any C++ data structures, include STL containers, classes, etc. C++ native data types and STL containers are ready to use, no need extra work, because metapp already supports them. Reflecting meta data for class is very easy.

• Very easy to learn. jsonpp doesn't define any data types to represent JSON document, instead it uses native C++ data type and STL containers which you have already mastered. The only "new" data type is metapp::Variant, you only need to learn the very basic knowledge.

• Multiple parser backends. jsonpp uses existing matured and well tested JSON parser libraries as the parser backend, such as simdjson 2.2.0 and json-parser. That means you can choose the best viable backend to achieve your goals and avoid the backends that make you trouble. Also adding new backend is very easy.

• Decent performance. Performance is not jsonpp strength and it's not close to high performance JSON libraries such as simdjson. However, the performance is better than some existing popular JSON libraries which also focus on usability, thanks to the high performance simdjson and well optimized metapp.

• Support stringify and parse almost all C++ native types and STL containers, such as integers, float points, C string, std::string, std::array, std::vector, std::list, std::deque, std::map, std::unordered_map, std::tuple, std::pair, etc. New containers can be added via metapp reflection system.

• JSON object can be parsed as sequence containers such as std::vector, and vice versa on dumping.

• Support stringify and parse classes and enumerators.

• Enumerators can be stringified as string names and parsed back to enum values.

• Support stringify to different storage, such as std::string, std::vector<char>, stream, or customized writer.

• Cross platforms, cross compilers.

• Written in standard and portable C++, only require C++11, and support later C++ standard.

• Use CMake to build and install.

Why jsonpp when there are hundreds of JSON libraries?

jsonpp is very easy to use and has good performance, and more important, the technology is reusable. For example, to parse/stringify customized classes, unlike the other libraries that require you to register read/write method to the JSON libraries, in jsonpp you only need to reflect meta data using metapp, then jsonpp can read and write the type for you. The meta data is reusable, your knowledge and experience is reusable too.

Basic information

License

Apache License, Version 2.0

Version 1.0.0-RC

The project is in release candidate stage and near the first release.
I will do more tests, code review, and gather feedbacks. After there is no critical bugs or serious design flaw, I will release the first version.

You are welcome to try the project and give feedback. Your participation will help to make the development faster and better.

Source code

https://github.com/wqking/jsonpp

Supported compilers

jsonpp requires C++ compiler that supports C++11 standard.
The library is tested with MSVC 2022, 2019, MinGW (Msys) GCC 8.3 and 11.3.0, Clang (carried by MSVC).
In brief, MSVC, GCC, Clang that has well support for C++11, or released after 2019, should be able to compile the library.

Quick start

Namespace

jsonpp

You should not use any nested namespace in jsonpp.

Example code

Parse JSON document

Header for Parser

#include "jsonpp/parser.h"

Create a parser with default configuration and default backend which is simdjson.

jsonpp::Parser parser;

This is the JSON we are going to parse.

const std::string jsonText = R"(
  [ 5, "abc", true, null, 3.14, [ 1, 2, 3 ], { "one": 1, "two": 2 } ]
)";

Parse the JSON, the result is a metapp::Variant.

const metapp::Variant var = parser.parse(jsonText);

The result is an array.

ASSERT(jsonpp::getJsonType(var) == jsonpp::JsonType::jtArray);

Get the underlying array. jsonpp::JsonArray is alias of std::vector<metapp::Variant>

const jsonpp::JsonArray & array = var.get<const jsonpp::JsonArray &>();

Now verify the elements. Don't feel strange on the types like JsonInt or JsonString, they are just aliases of C++ types. And we can parse JSON document as specified type, see next example.

ASSERT(array[0].get<jsonpp::JsonInt>() == 5);
ASSERT(array[1].get<const jsonpp::JsonString &>() == "abc");
ASSERT(array[2].get<jsonpp::JsonBool>());
ASSERT(array[3].get<jsonpp::JsonNull>() == nullptr);
ASSERT(array[4].get<jsonpp::JsonReal>() == 3.14);
const jsonpp::JsonArray & nestedArray = array[5].get<const jsonpp::JsonArray &>();
ASSERT(nestedArray[0].get<jsonpp::JsonInt>() == 1);
ASSERT(nestedArray[1].get<jsonpp::JsonInt>() == 2);
ASSERT(nestedArray[2].get<jsonpp::JsonInt>() == 3);
jsonpp::JsonObject & nestedObject = array[6].get<jsonpp::JsonObject &>();
ASSERT(nestedObject["one"].get<jsonpp::JsonInt>() == 1);
ASSERT(nestedObject["two"].get<jsonpp::JsonInt>() == 2);

Parse JSON document as specified type

Create a parser.

jsonpp::Parser parser;

This is the JSON we are going to parse. It's an array of integer, so we can use std::vector<int> to represent it. Note the float point 567.1 will be converted to int automatically.

const std::string jsonText = R"(
  [ 1, -2, 3, 567.1 ]
)";

Parse the JSON as std::vector.

const std::vector<int> array = parser.parse<std::vector<int> >(jsonText);

Verify the result.

REQUIRE(array[0] == 1);
REQUIRE(array[1] == -2);
REQUIRE(array[2] == 3);
REQUIRE(array[3] == 567);

Dump JSON object to text (stringify)

Header for Dumper

#include "jsonpp/dumper.h"

std::string text;

Create a dumper with default configuration.

jsonpp::Dumper dumper;

Dump a simple integer.

text = dumper.dump(5);
ASSERT(text == "5");

Dump a boolean.

text = dumper.dump(true);
ASSERT(text == "true");

Dump complicated data struct.

text = dumper.dump(jsonpp::JsonObject {
  { "first", "hello" },
  { "second", nullptr },
  { "third", std::vector<int> { 5, 6, 7 } },
  { "fourth", jsonpp::JsonArray { "abc", 9.1 } },
});
// JsonObject is alias of std::map<std::string, metapp::Variant>, so the keys are sorted alphabetically.
ASSERT(text == R"({"first":"hello","fourth":["abc",9.1],"second":null,"third":[5,6,7]})");

Dump/parse class object

Now let's dump and parse customized class objects. First let's define the enum and classes that we will use later.

enum class Gender
{
  female,
  male
};

struct Skill
{
  std::string name;
  int level;
};

struct Person
{
  std::string name;
  Gender gender;
  int age;
  std::vector<Skill> skills;
};

Now make the enum and class information available to metapp. jsonpp uses the reflection information from metapp. The information is not special to jsonpp, it's general reflection and can be used for other purposes such as serialization, script binding, etc.

template <>
struct metapp::DeclareMetaType <Gender> : metapp::DeclareMetaTypeBase <Gender>
{
  static const metapp::MetaEnum * getMetaEnum() {
    static const metapp::MetaEnum metaEnum([](metapp::MetaEnum & me) {
        me.registerValue("female", Gender::female);
        me.registerValue("male", Gender::male);
      }
    );
    return &metaEnum;
  }
};

template <>
struct metapp::DeclareMetaType <Skill> : metapp::DeclareMetaTypeBase <Skill>
{
  static const metapp::MetaClass * getMetaClass() {
    static const metapp::MetaClass metaClass(
      metapp::getMetaType<Skill>(),
      [](metapp::MetaClass & mc) {
        mc.registerAccessible("name", &Skill::name);
        mc.registerAccessible("level", &Skill::level);
      }
    );
    return &metaClass;
  }
};

// I don't encourage to use macros and I don't provide macros in metapp library.
// But for jsonpp users that don't want to dig into metapp and only want to use jsonpp features,
// jsonpp provides macros to ease the meta type declaration.
// Note: the macros are not required by jsonpp. The code can be rewritten without macros,
// same as how Skill is declared above.
JSONPP_BEGIN_DECLARE_CLASS(Person)
  JSONPP_REGISTER_CLASS_FIELD(name)
  JSONPP_REGISTER_CLASS_FIELD(gender)
  JSONPP_REGISTER_CLASS_FIELD(age)
  JSONPP_REGISTER_CLASS_FIELD(skills)
JSONPP_END_DECLARE_CLASS()

Now let's dump person to text, then parse the text back to Person object. enableNamedEnum(true) will use the name such as "female" for the Gender enum, instead of numbers such as 0. This allows the enum value change without breaking the dumped object.

Person person { "Mary", Gender::female, 26, { { "Writing", 8 }, { "Cooking", 6 } } };
jsonpp::Dumper dumper(jsonpp::DumperConfig().enableBeautify(true).enableNamedEnum(true));
// We don't user `person` any more, so we can move it to `dump` to avoid copying.
const std::string jsonText = dumper.dump(std::move(person));

The jsonText looks like,

{
    "name": "Mary",
    "gender": "female",
    "age": 26,
    "skills": [
        {
            "name": "Writing",
            "level": 8
        },
        {
            "name": "Cooking",
            "level": 6
        }
    ]
}

Now let's parse the JSON text back to Person object, and verify the values.

jsonpp::Parser parser;
const Person parsedPerson = parser.parse<Person>(jsonText);
ASSERT(parsedPerson.name == "Mary");
ASSERT(parsedPerson.gender == Gender::female);
ASSERT(parsedPerson.age == 26);
ASSERT(parsedPerson.skills[0].name == "Writing");
ASSERT(parsedPerson.skills[0].level == 8);
ASSERT(parsedPerson.skills[1].name == "Cooking");
ASSERT(parsedPerson.skills[1].level == 6);

We can not only dump/parse a single object, but also any STL containers with the objects.

Person personAlice { "Alice", Gender::female, 28, { { "Excel", 7 }, { "Word", 8 } } };
Person personTom { "Tom", Gender::male, 29, { { "C++", 9 }, { "Python", 10 }, { "PHP", 7 } } };

jsonpp::Dumper dumper(jsonpp::DumperConfig().enableBeautify(true).enableNamedEnum(true));
const std::string jsonText = dumper.dump(std::vector<Person> { personAlice, personTom });

jsonpp::Parser parser;
const std::vector<Person> parsedPersons = parser.parse<std::vector<Person> >(jsonText);
ASSERT(parsedPersons[0] == personAlice);
ASSERT(parsedPersons[1] == personTom);

Performance

Hardware: HP laptop, Intel(R) Core(TM) i5-8300H CPU @ 2.30GHz, 16 GB RAM.
Software: Windows 10 Pro, 21H2. MinGW gcc version 11.3.0, optimization level is -O3.

There are two parts in each benchmark data. The first part is the time to parse/dump one file, in milliseconds. The second part is the per second throughput.

Absolute data doesn't make sense, so the other library "JSON for Modern C++" (nlohmann) is tested for comparison.

Parse JSON file

File name	File size	jsonpp (simdjson)	jsonpp (json-parser)	nlohmann
canada.json	2,198 KB	18.7 ms, 114 MB/s	41.2 ms, 52 MB/s	108.8 ms, 19 MB/s
citm_catalog.json	1,686 KB	7.02 ms, 234 MB/s	18.84 ms, 87 MB/s	13.16 ms, 125 MB/s
twitter.json	616 KB	3.93 ms, 153 MB/s	8.81 ms, 68 MB/s	6 ms, 100 MB/s
airlines.json	4,848 KB	32.7 ms, 144 MB/s	73.5 ms, 64 MB/s	53.7 ms, 88 MB/s
tiny.json	348 B	0.0026 ms, 127 MB/s	0.0058 ms, 57 MB/s	0.007 ms, 47 MB/s
Zurich_Building.json	278 MB	4187 ms, 66 MB/s	9076 ms, 30 MB/s	10388 ms, 26 MB/s

Stringify JSON document

File name	jsonpp (beautify)	jsonpp (minify)	nlohmann (beautify)	nlohmann (minify)
canada.json	14.2 ms, 544 MB/s	10.1 ms, 197 MB/s	20.4 ms, 379 MB/s	14.9 ms, 133 MB/s
citm_catalog.json	4.11 ms, 461 MB/s	2.65 ms, 180 MB/s	4.86 ms, 338 MB/s	4.12 ms, 115 MB/s
twitter.json	2.05 ms, 364 MB/s	1.52 ms, 292 MB/s	4.08 ms, 179 MB/s	3.22 ms, 138 MB/s
airlines.json	21.1 ms, 275 MB/s	17.9 ms, 187 MB/s	23.8 ms, 244 MB/s	21.8 ms, 154 MB/s
tiny.json	0.0034 ms, 130 MB/s	0.003 ms, 80 MB/s	0.0017 ms, 251 MB/s	0.0016 ms, 152 MB/s
Zurich_Building.json	3724 ms, 345 MB/s	2649 ms, 105 MB/s	2665 ms, 482 MB/s	1943 ms, 143 MB/s

Documentations

Below are tutorials and documents. Don't feel upset if you find issues or missing stuff in the documents, I'm not native English speaker and it's not that exciting to write document. Any way, the code quality is always much better than the document, for ever.
If you want to contribute to the documents, be sure to read How to generate documentations.

• Build and install the library
• Use class Parser to read and parse JSON document
• Use class Dumper to dump and stringify JSON data
• Common and default data types
• Declare meta data, use classes and enumerators, use metapp

Known compiler related quirks in MSVC

MSVC 2022 and 2019, can build the CMake generated test projects and the tests run correctly in Debug and RelWithDebugInfo configurations. But some tests fail in Release mode when incremental linking is disabled.
Those failed tests should not fail, because they work correct in MSVC debug mode and in GCC/Clang. Adding /Debug option to linking which generates debug information makes the tests success.
Without /Debug option, but enabling incremental linking, will cause the tests success too.
So if jsonpp shows weird behavior in MSVC, try to enable incremental linking.