nlohmannjson/README.md

• Design goals
• Integration
• Examples
  • JSON as first-class data type
  • Serialization / Deserialization
  • STL-like access
  • Conversion from STL containers
  • JSON Pointer and JSON Patch
  • Implicit conversions
  • Conversions to/from arbitrary types
  • Binary formats (CBOR and MessagePack)
• Supported compilers
• License
• Thanks
• Used third-party tools
• Projects using JSON for Modern C++
• Notes
• Execute unit tests

Design goals

There are myriads of JSON libraries out there, and each may even have its reason to exist. Our class had these design goals:

• Intuitive syntax. In languages such as Python, JSON feels like a first class data type. We used all the operator magic of modern C++ to achieve the same feeling in your code. Check out the examples below and you'll know what I mean.

• Trivial integration. Our whole code consists of a single header file json.hpp. That's it. No library, no subproject, no dependencies, no complex build system. The class is written in vanilla C++11. All in all, everything should require no adjustment of your compiler flags or project settings.

• Serious testing. Our class is heavily unit-tested and covers 100% of the code, including all exceptional behavior. Furthermore, we checked with Valgrind that there are no memory leaks. To maintain high quality, the project is following the Core Infrastructure Initiative (CII) best practices.

Other aspects were not so important to us:

• Memory efficiency. Each JSON object has an overhead of one pointer (the maximal size of a union) and one enumeration element (1 byte). The default generalization uses the following C++ data types: std::string for strings, int64_t, uint64_t or double for numbers, std::map for objects, std::vector for arrays, and bool for Booleans. However, you can template the generalized class basic_json to your needs.

• Speed. There are certainly faster JSON libraries out there. However, if your goal is to speed up your development by adding JSON support with a single header, then this library is the way to go. If you know how to use a std::vector or std::map, you are already set.

See the contribution guidelines for more information.

Integration

The single required source, file json.hpp is in the src directory or released here. All you need to do is add

#include "json.hpp"

// for convenience
using json = nlohmann::json;

to the files you want to use JSON objects. That's it. Do not forget to set the necessary switches to enable C++11 (e.g., -std=c++11 for GCC and Clang).

🍺 If you are using OS X and Homebrew, just type brew tap nlohmann/json and brew install nlohmann_json and you're set. If you want the bleeding edge rather than the latest release, use brew install nlohmann_json --HEAD.

⚠️ Version 3.0.0 is currently under development. Branch develop is used for the ongoing work and is probably unstable. Please use the master branch for the last stable version 2.1.1.

Examples

Beside the examples below, you may want to check the documentation where each function contains a separate code example (e.g., check out emplace()). All example files can be compiled and executed on their own (e.g., file emplace.cpp).

JSON as first-class data type

Here are some examples to give you an idea how to use the class.

Assume you want to create the JSON object

{
  "pi": 3.141,
  "happy": true,
  "name": "Niels",
  "nothing": null,
  "answer": {
    "everything": 42
  },
  "list": [1, 0, 2],
  "object": {
    "currency": "USD",
    "value": 42.99
  }
}

With the JSON class, you could write:

// create an empty structure (null)
json j;

// add a number that is stored as double (note the implicit conversion of j to an object)
j["pi"] = 3.141;

// add a Boolean that is stored as bool
j["happy"] = true;

// add a string that is stored as std::string
j["name"] = "Niels";

// add another null object by passing nullptr
j["nothing"] = nullptr;

// add an object inside the object
j["answer"]["everything"] = 42;

// add an array that is stored as std::vector (using an initializer list)
j["list"] = { 1, 0, 2 };

// add another object (using an initializer list of pairs)
j["object"] = { {"currency", "USD"}, {"value", 42.99} };

// instead, you could also write (which looks very similar to the JSON above)
json j2 = {
  {"pi", 3.141},
  {"happy", true},
  {"name", "Niels"},
  {"nothing", nullptr},
  {"answer", {
    {"everything", 42}
  }},
  {"list", {1, 0, 2}},
  {"object", {
    {"currency", "USD"},
    {"value", 42.99}
  }}
};

Note that in all these cases, you never need to "tell" the compiler which JSON value you want to use. If you want to be explicit or express some edge cases, the functions json::array and json::object will help:

// a way to express the empty array []
json empty_array_explicit = json::array();

// ways to express the empty object {}
json empty_object_implicit = json({});
json empty_object_explicit = json::object();

// a way to express an _array_ of key/value pairs [["currency", "USD"], ["value", 42.99]]
json array_not_object = { json::array({"currency", "USD"}), json::array({"value", 42.99}) };

Serialization / Deserialization

To/from strings

You can create an object (deserialization) by appending _json to a string literal:

// create object from string literal
json j = "{ \"happy\": true, \"pi\": 3.141 }"_json;

// or even nicer with a raw string literal
auto j2 = R"(
  {
    "happy": true,
    "pi": 3.141
  }
)"_json;

Note that without appending the _json suffix, the passed string literal is not parsed, but just used as JSON string value. That is, json j = "{ \"happy\": true, \"pi\": 3.141 }" would just store the string "{ "happy": true, "pi": 3.141 }" rather than parsing the actual object.

The above example can also be expressed explicitly using json::parse():

// parse explicitly
auto j3 = json::parse("{ \"happy\": true, \"pi\": 3.141 }");

You can also get a string representation (serialize):

// explicit conversion to string
std::string s = j.dump();    // {\"happy\":true,\"pi\":3.141}

// serialization with pretty printing
// pass in the amount of spaces to indent
std::cout << j.dump(4) << std::endl;
// {
//     "happy": true,
//     "pi": 3.141
// }

To/from streams (e.g. files, string streams)

You can also use streams to serialize and deserialize:

// deserialize from standard input
json j;
std::cin >> j;

// serialize to standard output
std::cout << j;

// the setw manipulator was overloaded to set the indentation for pretty printing
std::cout << std::setw(4) << j << std::endl;

These operators work for any subclasses of std::istream or std::ostream. Here is the same example with files:

// read a JSON file
std::ifstream i("file.json");
json j;
i >> j;

// write prettified JSON to another file
std::ofstream o("pretty.json");
o << std::setw(4) << j << std::endl;

Please note that setting the exception bit for failbit is inappropriate for this use case. It will result in program termination due to the noexcept specifier in use.

Read from iterator range

You can also read JSON from an iterator range; that is, from any container accessible by iterators whose content is stored as contiguous byte sequence, for instance a std::vector<std::uint8_t>:

std::vector<std::uint8_t> v = {'t', 'r', 'u', 'e'};
json j = json::parse(v.begin(), v.end());

You may leave the iterators for the range [begin, end):

std::vector<std::uint8_t> v = {'t', 'r', 'u', 'e'};
json j = json::parse(v);

STL-like access

We designed the JSON class to behave just like an STL container. In fact, it satisfies the ReversibleContainer requirement.

// create an array using push_back
json j;
j.push_back("foo");
j.push_back(1);
j.push_back(true);

// also use emplace_back
j.emplace_back(1.78);

// iterate the array
for (json::iterator it = j.begin(); it != j.end(); ++it) {
  std::cout << *it << '\n';
}

// range-based for
for (auto& element : j) {
  std::cout << element << '\n';
}

// getter/setter
const std::string tmp = j[0];
j[1] = 42;
bool foo = j.at(2);

// comparison
j == "[\"foo\", 1, true]"_json;  // true

// other stuff
j.size();     // 3 entries
j.empty();    // false
j.type();     // json::value_t::array
j.clear();    // the array is empty again

// convenience type checkers
j.is_null();
j.is_boolean();
j.is_number();
j.is_object();
j.is_array();
j.is_string();

// create an object
json o;
o["foo"] = 23;
o["bar"] = false;
o["baz"] = 3.141;

// also use emplace
o.emplace("weather", "sunny");

// special iterator member functions for objects
for (json::iterator it = o.begin(); it != o.end(); ++it) {
  std::cout << it.key() << " : " << it.value() << "\n";
}

// find an entry
if (o.find("foo") != o.end()) {
  // there is an entry with key "foo"
}

// or simpler using count()
int foo_present = o.count("foo"); // 1
int fob_present = o.count("fob"); // 0

// delete an entry
o.erase("foo");

Conversion from STL containers

Any sequence container (std::array, std::vector, std::deque, std::forward_list, std::list) whose values can be used to construct JSON types (e.g., integers, floating point numbers, Booleans, string types, or again STL containers described in this section) can be used to create a JSON array. The same holds for similar associative containers (std::set, std::multiset, std::unordered_set, std::unordered_multiset), but in these cases the order of the elements of the array depends how the elements are ordered in the respective STL container.

std::vector<int> c_vector {1, 2, 3, 4};
json j_vec(c_vector);
// [1, 2, 3, 4]

std::deque<double> c_deque {1.2, 2.3, 3.4, 5.6};
json j_deque(c_deque);
// [1.2, 2.3, 3.4, 5.6]

std::list<bool> c_list {true, true, false, true};
json j_list(c_list);
// [true, true, false, true]

std::forward_list<int64_t> c_flist {12345678909876, 23456789098765, 34567890987654, 45678909876543};
json j_flist(c_flist);
// [12345678909876, 23456789098765, 34567890987654, 45678909876543]

std::array<unsigned long, 4> c_array {{1, 2, 3, 4}};
json j_array(c_array);
// [1, 2, 3, 4]

std::set<std::string> c_set {"one", "two", "three", "four", "one"};
json j_set(c_set); // only one entry for "one" is used
// ["four", "one", "three", "two"]

std::unordered_set<std::string> c_uset {"one", "two", "three", "four", "one"};
json j_uset(c_uset); // only one entry for "one" is used
// maybe ["two", "three", "four", "one"]

std::multiset<std::string> c_mset {"one", "two", "one", "four"};
json j_mset(c_mset); // both entries for "one" are used
// maybe ["one", "two", "one", "four"]

std::unordered_multiset<std::string> c_umset {"one", "two", "one", "four"};
json j_umset(c_umset); // both entries for "one" are used
// maybe ["one", "two", "one", "four"]

Likewise, any associative key-value containers (std::map, std::multimap, std::unordered_map, std::unordered_multimap) whose keys can construct an std::string and whose values can be used to construct JSON types (see examples above) can be used to to create a JSON object. Note that in case of multimaps only one key is used in the JSON object and the value depends on the internal order of the STL container.

std::map<std::string, int> c_map { {"one", 1}, {"two", 2}, {"three", 3} };
json j_map(c_map);
// {"one": 1, "three": 3, "two": 2 }

std::unordered_map<const char*, double> c_umap { {"one", 1.2}, {"two", 2.3}, {"three", 3.4} };
json j_umap(c_umap);
// {"one": 1.2, "two": 2.3, "three": 3.4}

std::multimap<std::string, bool> c_mmap { {"one", true}, {"two", true}, {"three", false}, {"three", true} };
json j_mmap(c_mmap); // only one entry for key "three" is used
// maybe {"one": true, "two": true, "three": true}

std::unordered_multimap<std::string, bool> c_ummap { {"one", true}, {"two", true}, {"three", false}, {"three", true} };
json j_ummap(c_ummap); // only one entry for key "three" is used
// maybe {"one": true, "two": true, "three": true}

JSON Pointer and JSON Patch

The library supports JSON Pointer (RFC 6901) as alternative means to address structured values. On top of this, JSON Patch (RFC 6902) allows to describe differences between two JSON values - effectively allowing patch and diff operations known from Unix.

// a JSON value
json j_original = R"({
  "baz": ["one", "two", "three"],
  "foo": "bar"
})"_json;

// access members with a JSON pointer (RFC 6901)
j_original["/baz/1"_json_pointer];
// "two"

// a JSON patch (RFC 6902)
json j_patch = R"([
  { "op": "replace", "path": "/baz", "value": "boo" },
  { "op": "add", "path": "/hello", "value": ["world"] },
  { "op": "remove", "path": "/foo"}
])"_json;

// apply the patch
json j_result = j_original.patch(j_patch);
// {
//    "baz": "boo",
//    "hello": ["world"]
// }

// calculate a JSON patch from two JSON values
json::diff(j_result, j_original);
// [
//   { "op":" replace", "path": "/baz", "value": ["one", "two", "three"] },
//   { "op": "remove","path": "/hello" },
//   { "op": "add", "path": "/foo", "value": "bar" }
// ]

Implicit conversions

The type of the JSON object is determined automatically by the expression to store. Likewise, the stored value is implicitly converted.

// strings
std::string s1 = "Hello, world!";
json js = s1;
std::string s2 = js;

// Booleans
bool b1 = true;
json jb = b1;
bool b2 = jb;

// numbers
int i = 42;
json jn = i;
double f = jn;

// etc.

You can also explicitly ask for the value:

std::string vs = js.get<std::string>();
bool vb = jb.get<bool>();
int vi = jn.get<int>();

// etc.

Arbitrary types conversions

Every type can be serialized in JSON, not just STL-containers and scalar types. Usually, you would do something along those lines:

namespace ns {
    // a simple struct to model a person
    struct person {
        std::string name;
        std::string address;
        int age;
    };
}

ns::person p = {"Ned Flanders", "744 Evergreen Terrace", 60};

// convert to JSON: copy each value into the JSON object
json j;
j["name"] = p.name;
j["address"] = p.address;
j["age"] = p.age;

// ...

// convert from JSON: copy each value from the JSON object
ns::person p {
    j["name"].get<std::string>(),
    j["address"].get<std::string>(),
    j["age"].get<int>()
};

It works, but that's quite a lot of boilerplate... Fortunately, there's a better way:

// create a person
ns::person p {"Ned Flanders", "744 Evergreen Terrace", 60};

// conversion: person -> json
json j = p;

std::cout << j << std::endl;
// {"address":"744 Evergreen Terrace","age":60,"name":"Ned Flanders"}

// conversion: json -> person
ns::person p2 = j;

// that's it
assert(p == p2);

Basic usage

To make this work with one of your types, you only need to provide two functions:

using nlohmann::json;

namespace ns {
    void to_json(json& j, const person& p) {
        j = json{{"name", p.name}, {"address", p.address}, {"age", p.age}};
    }

    void from_json(const json& j, person& p) {
        p.name = j.at("name").get<std::string>();
        p.address = j.at("address").get<std::string>();
        p.age = j.at("age").get<int>();
    }
} // namespace ns

That's all! When calling the json constructor with your type, your custom to_json method will be automatically called. Likewise, when calling get<your_type>(), the from_json method will be called.

Some important things:

• Those methods MUST be in your type's namespace (which can be the global namespace), or the library will not be able to locate them (in this example, they are in namespace ns, where person is defined).
• When using get<your_type>(), your_type MUST be DefaultConstructible. (There is a way to bypass this requirement described later.)
• In function from_json, use function at() to access the object values rather than operator[]. In case a key does not exists, at throws an exception that you can handle, whereas operator[] exhibits undefined behavior.

How do I convert third-party types?

This requires a bit more advanced technique. But first, let's see how this conversion mechanism works:

The library uses JSON Serializers to convert types to json. The default serializer for nlohmann::json is nlohmann::adl_serializer (ADL means Argument-Dependent Lookup).

It is implemented like this (simplified):

template <typename T>
struct adl_serializer {
    static void to_json(json& j, const T& value) {
        // calls the "to_json" method in T's namespace
    }

    static void from_json(const json& j, T& value) {
        // same thing, but with the "from_json" method
    }
};

This serializer works fine when you have control over the type's namespace. However, what about boost::optional, or std::filesystem::path (C++17)? Hijacking the boost namespace is pretty bad, and it's illegal to add something other than template specializations to std...

To solve this, you need to add a specialization of adl_serializer to the nlohmann namespace, here's an example:

// partial specialization (full specialization works too)
namespace nlohmann {
    template <typename T>
    struct adl_serializer<boost::optional<T>> {
        static void to_json(json& j, const boost::optional<T>& opt) {
            if (opt == boost::none) {
                j = nullptr;
            } else {
              j = *opt; // this will call adl_serializer<T>::to_json which will
                        // find the free function to_json in T's namespace!
            }
        }

        static void from_json(const json& j, boost::optional<T>& opt) {
            if (j.is_null()) {
                opt = boost::none;
            } else {
                opt = j.get<T>(); // same as above, but with 
                                  // adl_serializer<T>::from_json
            }
        }
    };
}

How can I use get() for non-default constructible/non-copyable types?

There is a way, if your type is MoveConstructible. You will need to specialize the adl_serializer as well, but with a special from_json overload:

struct move_only_type {
    move_only_type() = delete;
    move_only_type(int ii): i(ii) {}
    move_only_type(const move_only_type&) = delete;
    move_only_type(move_only_type&&) = default;
    
    int i;
};

namespace nlohmann {
    template <>
    struct adl_serializer<move_only_type> {
        // note: the return type is no longer 'void', and the method only takes
        // one argument
        static move_only_type from_json(const json& j) {
            return {j.get<int>()};
        }
        
        // Here's the catch! You must provide a to_json method! Otherwise you
        // will not be able to convert move_only_type to json, since you fully
        // specialized adl_serializer on that type
        static void to_json(json& j, move_only_type t) {
            j = t.i;
        }
    };
}

Can I write my own serializer? (Advanced use)

Yes. You might want to take a look at unit-udt.cpp in the test suite, to see a few examples.

If you write your own serializer, you'll need to do a few things:

• use a different basic_json alias than nlohmann::json (the last template parameter of basic_json is the JSONSerializer)
• use your basic_json alias (or a template parameter) in all your to_json/from_json methods
• use nlohmann::to_json and nlohmann::from_json when you need ADL

Here is an example, without simplifications, that only accepts types with a size <= 32, and uses ADL.

// You should use void as a second template argument
// if you don't need compile-time checks on T
template<typename T, typename SFINAE = typename std::enable_if<sizeof(T) <= 32>::type>
struct less_than_32_serializer {
    template <typename BasicJsonType>
    static void to_json(BasicJsonType& j, T value) {
        // we want to use ADL, and call the correct to_json overload
        using nlohmann::to_json; // this method is called by adl_serializer,
                                 // this is where the magic happens
        to_json(j, value);
    }
    
    template <typename BasicJsonType>
    static void from_json(const BasicJsonType& j, T& value) {
        // same thing here
        using nlohmann::from_json;
        from_json(j, value);
    }
};

Be very careful when reimplementing your serializer, you can stack overflow if you don't pay attention:

template <typename T, void>
struct bad_serializer
{
    template <typename BasicJsonType>
    static void to_json(BasicJsonType& j, const T& value) {
      // this calls BasicJsonType::json_serializer<T>::to_json(j, value);
      // if BasicJsonType::json_serializer == bad_serializer ... oops!
      j = value;
    }
    
    template <typename BasicJsonType>
    static void to_json(const BasicJsonType& j, T& value) {
      // this calls BasicJsonType::json_serializer<T>::from_json(j, value);
      // if BasicJsonType::json_serializer == bad_serializer ... oops!
      value = j.template get<T>(); // oops!
    }
};

Binary formats (CBOR and MessagePack)

Though JSON is a ubiquitous data format, it is not a very compact format suitable for data exchange, for instance over a network. Hence, the library supports CBOR (Concise Binary Object Representation) and MessagePack to efficiently encode JSON values to byte vectors and to decode such vectors.

// create a JSON value
json j = R"({"compact": true, "schema": 0})"_json;

// serialize to CBOR
std::vector<std::uint8_t> v_cbor = json::to_cbor(j);

// 0xa2, 0x67, 0x63, 0x6f, 0x6d, 0x70, 0x61, 0x63, 0x74, 0xf5, 0x66, 0x73, 0x63, 0x68, 0x65, 0x6d, 0x61, 0x00

// roundtrip
json j_from_cbor = json::from_cbor(v_cbor);

// serialize to MessagePack
std::vector<std::uint8_t> v_msgpack = json::to_msgpack(j);

// 0x82, 0xa7, 0x63, 0x6f, 0x6d, 0x70, 0x61, 0x63, 0x74, 0xc3, 0xa6, 0x73, 0x63, 0x68, 0x65, 0x6d, 0x61, 0x00

// roundtrip
json j_from_msgpack = json::from_msgpack(v_msgpack);

Supported compilers

Though it's 2016 already, the support for C++11 is still a bit sparse. Currently, the following compilers are known to work:

• GCC 4.9 - 6.0 (and possibly later)
• Clang 3.4 - 3.9 (and possibly later)
• Microsoft Visual C++ 2015 / Build Tools 14.0.25123.0 (and possibly later)
• Microsoft Visual C++ 2017 / Build Tools 15.1.548.43366 (and possibly later)

I would be happy to learn about other compilers/versions.

Please note:

• GCC 4.8 does not work because of two bugs (55817 and 57824) in the C++11 support. Note there is a pull request to fix some of the issues.

• Android defaults to using very old compilers and C++ libraries. To fix this, add the following to your Application.mk. This will switch to the LLVM C++ library, the Clang compiler, and enable C++11 and other features disabled by default.

  APP_STL := c++_shared
  NDK_TOOLCHAIN_VERSION := clang3.6
  APP_CPPFLAGS += -frtti -fexceptions
  

  The code compiles successfully with Android NDK, Revision 9 - 11 (and possibly later) and CrystaX's Android NDK version 10.

• For GCC running on MinGW or Android SDK, the error 'to_string' is not a member of 'std' (or similarly, for strtod) may occur. Note this is not an issue with the code, but rather with the compiler itself. On Android, see above to build with a newer environment. For MinGW, please refer to this site and this discussion for information on how to fix this bug. For Android NDK using APP_STL := gnustl_static, please refer to this discussion.

The following compilers are currently used in continuous integration at Travis and AppVeyor:

Compiler	Operating System	Version String
GCC 4.9.3	Ubuntu 14.04.4 LTS	g++-4.9 (Ubuntu 4.9.3-8ubuntu2~14.04) 4.9.3
GCC 5.3.0	Ubuntu 14.04.4 LTS	g++-5 (Ubuntu 5.3.0-3ubuntu1~14.04) 5.3.0 20151204
GCC 6.1.1	Ubuntu 14.04.4 LTS	g++-6 (Ubuntu 6.1.1-3ubuntu11~14.04.1) 6.1.1 20160511
Clang 3.6.0	Ubuntu 14.04.4 LTS	clang version 3.6.0 (tags/RELEASE_360/final)
Clang 3.6.1	Ubuntu 14.04.4 LTS	clang version 3.6.1 (tags/RELEASE_361/final)
Clang 3.6.2	Ubuntu 14.04.4 LTS	clang version 3.6.2 (tags/RELEASE_362/final)
Clang 3.7.0	Ubuntu 14.04.4 LTS	clang version 3.7.0 (tags/RELEASE_370/final)
Clang 3.7.1	Ubuntu 14.04.4 LTS	clang version 3.7.1 (tags/RELEASE_371/final)
Clang 3.8.0	Ubuntu 14.04.4 LTS	clang version 3.8.0 (tags/RELEASE_380/final)
Clang 3.8.1	Ubuntu 14.04.4 LTS	clang version 3.8.1 (tags/RELEASE_381/final)
Clang Xcode 6.4	Darwin Kernel Version 14.3.0 (OSX 10.10.3)	Apple LLVM version 6.1.0 (clang-602.0.53) (based on LLVM 3.6.0svn)
Clang Xcode 7.3	Darwin Kernel Version 15.0.0 (OSX 10.10.5)	Apple LLVM version 7.3.0 (clang-703.0.29)
Clang Xcode 8.0	Darwin Kernel Version 15.6.0	Apple LLVM version 8.0.0 (clang-800.0.38)
Clang Xcode 8.1	Darwin Kernel Version 16.1.0 (macOS 10.12.1)	Apple LLVM version 8.0.0 (clang-800.0.42.1)
Clang Xcode 8.2	Darwin Kernel Version 16.1.0 (macOS 10.12.1)	Apple LLVM version 8.0.0 (clang-800.0.42.1)
Clang Xcode 8.3	Darwin Kernel Version 16.5.0 (macOS 10.12.4)	Apple LLVM version 8.1.0 (clang-802.0.38)
Visual Studio 14 2015	Windows Server 2012 R2 (x64)	Microsoft (R) Build Engine version 14.0.25123.0
Visual Studio 2017	Windows Server 2016	Microsoft (R) Build Engine version 15.1.548.43366

License

The class is licensed under the MIT License:

Copyright © 2013-2017 Niels Lohmann

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Thanks

I deeply appreciate the help of the following people.

• Teemperor implemented CMake support and lcov integration, realized escape and Unicode handling in the string parser, and fixed the JSON serialization.
• elliotgoodrich fixed an issue with double deletion in the iterator classes.
• kirkshoop made the iterators of the class composable to other libraries.
• wancw fixed a bug that hindered the class to compile with Clang.
• Tomas Åblad found a bug in the iterator implementation.
• Joshua C. Randall fixed a bug in the floating-point serialization.
• Aaron Burghardt implemented code to parse streams incrementally. Furthermore, he greatly improved the parser class by allowing the definition of a filter function to discard undesired elements while parsing.
• Daniel Kopeček fixed a bug in the compilation with GCC 5.0.
• Florian Weber fixed a bug in and improved the performance of the comparison operators.
• Eric Cornelius pointed out a bug in the handling with NaN and infinity values. He also improved the performance of the string escaping.
• 易思龙 implemented a conversion from anonymous enums.
• kepkin patiently pushed forward the support for Microsoft Visual studio.
• gregmarr simplified the implementation of reverse iterators and helped with numerous hints and improvements. In particular, he pushed forward the implementation of user-defined types.
• Caio Luppi fixed a bug in the Unicode handling.
• dariomt fixed some typos in the examples.
• Daniel Frey cleaned up some pointers and implemented exception-safe memory allocation.
• Colin Hirsch took care of a small namespace issue.
• Huu Nguyen correct a variable name in the documentation.
• Silverweed overloaded parse() to accept an rvalue reference.
• dariomt fixed a subtlety in MSVC type support and implemented the get_ref() function to get a reference to stored values.
• ZahlGraf added a workaround that allows compilation using Android NDK.
• whackashoe replaced a function that was marked as unsafe by Visual Studio.
• 406345 fixed two small warnings.
• Glen Fernandes noted a potential portability problem in the has_mapped_type function.
• Corbin Hughes fixed some typos in the contribution guidelines.
• twelsby fixed the array subscript operator, an issue that failed the MSVC build, and floating-point parsing/dumping. He further added support for unsigned integer numbers and implemented better roundtrip support for parsed numbers.
• Volker Diels-Grabsch fixed a link in the README file.
• msm- added support for american fuzzy lop.
• Annihil fixed an example in the README file.
• Themercee noted a wrong URL in the README file.
• Lv Zheng fixed a namespace issue with int64_t and uint64_t.
• abc100m analyzed the issues with GCC 4.8 and proposed a partial solution.
• zewt added useful notes to the README file about Android.
• Róbert Márki added a fix to use move iterators and improved the integration via CMake.
• Chris Kitching cleaned up the CMake files.
• Tom Needham fixed a subtle bug with MSVC 2015 which was also proposed by Michael K..
• Mário Feroldi fixed a small typo.
• duncanwerner found a really embarrassing performance regression in the 2.0.0 release.
• Damien fixed one of the last conversion warnings.
• Thomas Braun fixed a warning in a test case.
• Théo DELRIEU patiently and constructively oversaw the long way toward iterator-range parsing. He also implemented the magic behind the serialization/deserialization of user-defined types.
• Stefan fixed a minor issue in the documentation.
• Vasil Dimov fixed the documentation regarding conversions from std::multiset.
• ChristophJud overworked the CMake files to ease project inclusion.
• Vladimir Petrigo made a SFINAE hack more readable and added Visual Studio 17 to the build matrix.
• Denis Andrejew fixed a grammar issue in the README file.
• Pierre-Antoine Lacaze found a subtle bug in the dump() function.
• TurpentineDistillery pointed to std::locale::classic() to avoid too much locale joggling, found some nice performance improvements in the parser, improved the benchmarking code, and realized locale-independent number parsing and printing.
• cgzones had an idea how to fix the Coverity scan.
• Jared Grubb silenced a nasty documentation warning.
• Yixin Zhang fixed an integer overflow check.
• Bosswestfalen merged two iterator classes into a smaller one.
• Daniel599 helped to get Travis execute the tests with Clang's sanitizers.
• Jonathan Lee fixed an example in the README file.
• gnzlbg supported the implementation of user-defined types.
• Alexej Harm helped to get the user-defined types working with Visual Studio.
• Jared Grubb supported the implementation of user-defined types.
• EnricoBilla noted a typo in an example.
• Martin Hořeňovský found a way for a 2x speedup for the compilation time of the test suite.
• ukhegg found proposed an improvement for the examples section.
• rswanson-ihi noted a typo in the README.
• Mihai Stan fixed a bug in the comparison with nullptrs.
• Tushar Maheshwari added cotire support to speed up the compilation.
• TedLyngmo noted a typo in the README, removed unnecessary bit arithmetic, and fixed some -Weffc++ warnings.
• Krzysztof Woś made exceptions more visible.
• ftillier fixed a compiler warning.
• tinloaf made sure all pushed warnings are properly popped.

Thanks a lot for helping out! Please let me know if I forgot someone.

Used third-party tools

The library itself contains of a single header file licensed under the MIT license. However, it is built, tested, documented, and whatnot using a lot of third-party tools and services. Thanks a lot!

• American fuzzy lop for fuzz testing
• AppVeyor for continuous integration on Windows
• Artistic Style for automatic source code identation
• benchpress to benchmark the code
• Catch for the unit tests
• Clang for compilation with code sanitizers
• Cmake for build automation
• Codacity for further code analysis
• cotire to speed of compilation
• Coveralls to measure code coverage
• Coverity Scan for static analysis
• cppcheck for static analysis
• cxxopts to let benchpress parse command-line parameters
• Doxygen to generate documentation
• git-update-ghpages to upload the documentation to gh-pages
• Github Changelog Generator to generate the ChangeLog
• libFuzzer to implement fuzz testing for OSS-Fuzz
• OSS-Fuzz for continuous fuzz testing of the library
• re2c to generate an automaton for the lexical analysis
• send_to_wandbox to send code examples to Wandbox
• Travis for continuous integration on Linux and macOS
• Valgrind to check for correct memory management
• Wandbox for online examples

Projects using JSON for Modern C++

The library is currently used in Apple macOS Sierra and iOS 10. I am not sure what they are using the library for, but I am happy that it runs on so many devices.

Notes

• The code contains numerous debug assertions which can be switched off by defining the preprocessor macro NDEBUG, see the documentation of assert. In particular, note operator[] implements unchecked access for const objects: If the given key is not present, the behavior is undefined (think of a dereferenced null pointer) and yields an assertion failure if assertions are switched on. If you are not sure whether an element in an object exists, use checked access with the at() function.
• As the exact type of a number is not defined in the JSON specification, this library tries to choose the best fitting C++ number type automatically. As a result, the type double may be used to store numbers which may yield floating-point exceptions in certain rare situations if floating-point exceptions have been unmasked in the calling code. These exceptions are not caused by the library and need to be fixed in the calling code, such as by re-masking the exceptions prior to calling library functions.
• The library supports Unicode input as follows:
  • Only UTF-8 encoded input is supported which is the default encoding for JSON according to RFC 7159.
  • Other encodings such as Latin-1, UTF-16, or UTF-32 are not supported and will yield parse errors.
  • Unicode noncharacters will not be replaced by the library.
  • Invalid surrogates (e.g., incomplete pairs such as \uDEAD) will yield parse errors.
  • The strings stored in the library are UTF-8 encoded. When using the default string type (std::string), note that its length/size functions return the number of stored bytes rather than the number of characters or glyphs.
• The code can be compiled without C++ runtime type identification features; that is, you can use the -fno-rtti compiler flag.
• Exceptions are used widely within the library. They can, however, be switched off with either using the compiler flag -fno-exceptions or by defining the symbol JSON_NOEXCEPTION. In this case, exceptions are replaced by an abort() call.
• By default, the library does not preserve the insertion order of object elements. This is standards-compliant, as the JSON standard defines objects as "an unordered collection of zero or more name/value pairs". If you do want to preserve the insertion order, you can specialize the object type with containers like tsl::ordered_map or nlohmann::fifo_map.

Execute unit tests

To compile and run the tests, you need to execute

$ make json_unit -Ctest
$ ./test/json_unit "*"

===============================================================================
All tests passed (11203022 assertions in 48 test cases)

Alternatively, you can use CMake and run

$ mkdir build
$ cd build
$ cmake ..
$ make
$ ctest

For more information, have a look at the file .travis.yml.