JSON for Modern C++

Build Status Build Status Build Status Coverage Status Coverity Scan Build Status Codacy Badge Language grade: C/C++ Fuzzing Status Try online Documentation GitHub license GitHub Releases GitHub Issues Average time to resolve an issue CII Best Practices

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 and the Clang Sanitizers that there are no memory leaks. Google OSS-Fuzz additionally runs fuzz tests against all parsers 24/7, effectively executing billions of tests so far. 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.


json.hpp is the single required file in single_include/nlohmann or released here. You need to add

#include <nlohmann/json.hpp>

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

to the files you want to process JSON and set the necessary switches to enable C++11 (e.g., -std=c++11 for GCC and Clang).

You can further use file include/nlohmann/json_fwd.hpp for forward-declarations. The installation of json_fwd.hpp (as part of cmake's install step), can be achieved by setting -DJSON_MultipleHeaders=ON.


You can also use the nlohmann_json::nlohmann_json interface target in CMake. This target populates the appropriate usage requirements for INTERFACE_INCLUDE_DIRECTORIES to point to the appropriate include directories and INTERFACE_COMPILE_FEATURES for the necessary C++11 flags.


To use this library from a CMake project, you can locate it directly with find_package() and use the namespaced imported target from the generated package configuration:

# CMakeLists.txt
find_package(nlohmann_json 3.2.0 REQUIRED)
add_library(foo ...)
target_link_libraries(foo PRIVATE nlohmann_json::nlohmann_json)

The package configuration file, nlohmann_jsonConfig.cmake, can be used either from an install tree or directly out of the build tree.


To embed the library directly into an existing CMake project, place the entire source tree in a subdirectory and call add_subdirectory() in your CMakeLists.txt file:

# Typically you don't care so much for a third party library's tests to be
# run from your own project's code.

# If you only include this third party in PRIVATE source files, you do not
# need to install it when your main project gets installed.

# Don't use include(nlohmann_json/CMakeLists.txt) since that carries with it
# unintended consequences that will break the build.  It's generally
# discouraged (although not necessarily well documented as such) to use
# include(...) for pulling in other CMake projects anyways.
add_library(foo ...)
target_link_libraries(foo PRIVATE nlohmann_json::nlohmann_json)

Supporting Both

To allow your project to support either an externally supplied or an embedded JSON library, you can use a pattern akin to the following:

# Top level CMakeLists.txt
option(FOO_USE_EXTERNAL_JSON "Use an external JSON library" OFF)
add_library(foo ...)
# Note that the namespaced target will always be available regardless of the
# import method
target_link_libraries(foo PRIVATE nlohmann_json::nlohmann_json)
# thirdparty/CMakeLists.txt
  find_package(nlohmann_json 3.2.0 REQUIRED)
  set(JSON_BuildTests OFF CACHE INTERNAL "")

thirdparty/nlohmann_json is then a complete copy of this source tree.

Package Managers

:beer: 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.

If you are using the Meson Build System, add this source tree as a meson subproject. You may also use the include.zip published in this project's Releases to reduce the size of the vendored source tree. Alternatively, you can get a wrap file by downloading it from Meson WrapDB, or simply use meson wrap install nlohmann_json. Please see the meson project for any issues regarding the packaging.

The provided meson.build can also be used as an alternative to cmake for installing nlohmann_json system-wide in which case a pkg-config file is installed. To use it, simply have your build system require the nlohmann_json pkg-config dependency. In Meson, it is preferred to use the dependency() object with a subproject fallback, rather than using the subproject directly.

If you are using Conan to manage your dependencies, merely add jsonformoderncpp/x.y.z@vthiery/stable to your conanfile.py's requires, where x.y.z is the release version you want to use. Please file issues here if you experience problems with the packages.

If you are using Spack to manage your dependencies, you can use the nlohmann-json package. Please see the spack project for any issues regarding the packaging.

If you are using hunter on your project for external dependencies, then you can use the nlohmann_json package. Please see the hunter project for any issues regarding the packaging.

If you are using Buckaroo, you can install this library's module with buckaroo add github.com/buckaroo-pm/nlohmann-json. Please file issues here. There is a demo repo here.

If you are using vcpkg on your project for external dependencies, then you can use the nlohmann-json package. Please see the vcpkg project for any issues regarding the packaging.

If you are using cget, you can install the latest development version with cget install nlohmann/json. A specific version can be installed with cget install nlohmann/json@v3.1.0. Also, the multiple header version can be installed by adding the -DJSON_MultipleHeaders=ON flag (i.e., cget install nlohmann/json -DJSON_MultipleHeaders=ON).

If you are using CocoaPods, you can use the library by adding pod "nlohmann_json", '~>3.1.2' to your podfile (see an example). Please file issues here.

If you are using NuGet, you can use the package nlohmann.json. Please check this extensive description on how to use the package. Please files issues here.

If you are using conda, you can use the package nlohmann_json from conda-forge executing conda install -c conda-forge nlohmann_json. Please file issues here.

If you are using MSYS2, your can use the mingw-w64-nlohmann_json package, just type pacman -S mingw-w64-i686-nlohmann_json or pacman -S mingw-w64-x86_64-nlohmann_json for installation. Please file issues here if you experience problems with the packages.


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 this library, 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 type 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"}, {"value", 42.99} });

Serialization / Deserialization

To/from strings

You can create a JSON value (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

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 of a JSON value (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
// }

Note the difference between serialization and assignment:

// store a string in a JSON value
json j_string = "this is a string";

// retrieve the string value
auto cpp_string = j_string.get<std::string>();
// retrieve the string value (alternative when an variable already exists)
std::string cpp_string2;

// retrieve the serialized value (explicit JSON serialization)
std::string serialized_string = j_string.dump();

// output of original string
std::cout << cpp_string << " == " << cpp_string2 << " == " << j_string.get<std::string>() << '\n';
// output of serialized value
std::cout << j_string << " == " << serialized_string << std::endl;

.dump() always returns the serialized value, and .get<std::string>() returns the originally stored string value.

Note the library only supports UTF-8. When you store strings with different encodings in the library, calling dump() may throw an exception unless json::error_handler_t::replace or json::error_handler_t::ignore are used as error handlers.

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 parse 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);

SAX interface

The library uses a SAX-like interface with the following functions:

// called when null is parsed
bool null();

// called when a boolean is parsed; value is passed
bool boolean(bool val);

// called when a signed or unsigned integer number is parsed; value is passed
bool number_integer(number_integer_t val);
bool number_unsigned(number_unsigned_t val);

// called when a floating-point number is parsed; value and original string is passed
bool number_float(number_float_t val, const string_t& s);

// called when a string is parsed; value is passed and can be safely moved away
bool string(string_t& val);

// called when an object or array begins or ends, resp. The number of elements is passed (or -1 if not known)
bool start_object(std::size_t elements);
bool end_object();
bool start_array(std::size_t elements);
bool end_array();
// called when an object key is parsed; value is passed and can be safely moved away
bool key(string_t& val);

// called when a parse error occurs; byte position, the last token, and an exception is passed
bool parse_error(std::size_t position, const std::string& last_token, const detail::exception& ex);

The return value of each function determines whether parsing should proceed.

To implement your own SAX handler, proceed as follows:

  1. Implement the SAX interface in a class. You can use class nlohmann::json_sax<json> as base class, but you can also use any class where the functions described above are implemented and public.
  2. Create an object of your SAX interface class, e.g. my_sax.
  3. Call bool json::sax_parse(input, &my_sax); where the first parameter can be any input like a string or an input stream and the second parameter is a pointer to your SAX interface.

Note the sax_parse function only returns a bool indicating the result of the last executed SAX event. It does not return a json value - it is up to you to decide what to do with the SAX events. Furthermore, no exceptions are thrown in case of a parse error - it is up to you what to do with the exception object passed to your parse_error implementation. Internally, the SAX interface is used for the DOM parser (class json_sax_dom_parser) as well as the acceptor (json_sax_acceptor), see file json_sax.hpp.

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;

// also use emplace_back

// 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 auto tmp = j[0].get<std::string>();
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

// 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";

// the same code as range for
for (auto& el : o.items()) {
  std::cout << el.key() << " : " << el.value() << "\n";

// even easier with structured bindings (C++17)
for (auto& [key, value] : o.items()) {
  std::cout << key << " : " << 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

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 values (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 on 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 values (see examples above) can be used 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"

// access members with a JSON pointer (RFC 6901)
// "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"}

// 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" }
// ]

JSON Merge Patch

The library supports JSON Merge Patch (RFC 7386) as a patch format. Instead of using JSON Pointer (see above) to specify values to be manipulated, it describes the changes using a syntax that closely mimics the document being modified.

// a JSON value
json j_document = R"({
  "a": "b",
  "c": {
    "d": "e",
    "f": "g"

// a patch
json j_patch = R"({
  "c": {
    "f": null

// apply the patch
// {
//  "a": "z",
//  "c": {
//    "d": "e"
//  }
// }

Implicit conversions

Supported types can be implicitly converted to JSON values.

It is recommended to NOT USE implicit conversions FROM a JSON value. You can find more details about this recommendation here.

// strings
std::string s1 = "Hello, world!";
json js = s1;
auto s2 = js.get<std::string>();
std::string s3 = js;
std::string s4;
s4 = js;

// Booleans
bool b1 = true;
json jb = b1;
auto b2 = jb.get<bool>();
bool b3 = jb;
bool b4;
b4 = jb;

// numbers
int i = 42;
json jn = i;
auto f = jn.get<double>();
double f2 = jb;
double f3;
f3 = jb;

// etc.

Note that char types are not automatically converted to JSON strings, but to integer numbers. A conversion to a string must be specified explicitly:

char ch = 'A';                       // ASCII value 65
json j_default = ch;                 // stores integer number 65
json j_string = std::string(1, ch);  // stores string "A"

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 {

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
auto p2 = j.get<ns::person>();

// 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) {
} // 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>() or get_to(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).
  • Those methods MUST be available (e.g., proper headers must be included) everywhere you use these conversions. Look at issue 1108 for errors that may occur otherwise.
  • 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 exist, at throws an exception that you can handle, whereas operator[] exhibits undefined behavior.
  • You do not need to add serializers or deserializers for STL types like std::vector: the library already implements these.

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!

Specializing enum conversion

By default, enum values are serialized to JSON as integers. In some cases this could result in undesired behavior. If an enum is modified or re-ordered after data has been serialized to JSON, the later de-serialized JSON data may be undefined or a different enum value than was originally intended.

It is possible to more precisely specify how a given enum is mapped to and from JSON as shown below:

// example enum type declaration
enum TaskState {

// map TaskState values to JSON as strings
    {TS_INVALID, nullptr},
    {TS_STOPPED, "stopped"},
    {TS_RUNNING, "running"},
    {TS_COMPLETED, "completed"},

The NLOHMANN_JSON_SERIALIZE_ENUM() macro declares a set of to_json() / from_json() functions for type TaskState while avoiding repetition and boilerplate serialization code.


// enum to JSON as string
json j = TS_STOPPED;
assert(j == "stopped");

// json string to enum
json j3 = "running";
assert(j3.get<TaskState>() == TS_RUNNING);

// undefined json value to enum (where the first map entry above is the default)
json jPi = 3.14;
assert(jPi.get<TaskState>() == TS_INVALID );

Just as in Arbitrary Type Conversions above,

  • NLOHMANN_JSON_SERIALIZE_ENUM() MUST be declared in your enum type's namespace (which can be the global namespace), or the library will not be able to locate it and it will default to integer serialization.
  • It MUST be available (e.g., proper headers must be included) everywhere you use the conversions.

Other Important points:

  • When using get<ENUM_TYPE>(), undefined JSON values will default to the first pair specified in your map. Select this default pair carefully.
  • If an enum or JSON value is specified more than once in your map, the first matching occurrence from the top of the map will be returned when converting to or from JSON.

Binary formats (BSON, CBOR, MessagePack, and UBJSON)

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 BSON (Binary JSON), CBOR (Concise Binary Object Representation), MessagePack, and UBJSON (Universal Binary JSON Specification) 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 BSON
std::vector<std::uint8_t> v_bson = json::to_bson(j);

// 0x1B, 0x00, 0x00, 0x00, 0x08, 0x63, 0x6F, 0x6D, 0x70, 0x61, 0x63, 0x74, 0x00, 0x01, 0x10, 0x73, 0x63, 0x68, 0x65, 0x6D, 0x61, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00

// roundtrip
json j_from_bson = json::from_bson(v_bson);

// 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);

// serialize to UBJSON
std::vector<std::uint8_t> v_ubjson = json::to_ubjson(j);

// 0x7B, 0x69, 0x07, 0x63, 0x6F, 0x6D, 0x70, 0x61, 0x63, 0x74, 0x54, 0x69, 0x06, 0x73, 0x63, 0x68, 0x65, 0x6D, 0x61, 0x69, 0x00, 0x7D

// roundtrip
json j_from_ubjson = json::from_ubjson(v_ubjson);

Supported compilers

Though it's 2019 already, the support for C++11 is still a bit sparse. Currently, the following compilers are known to work:

  • GCC 4.8 - 9.2 (and possibly later)
  • Clang 3.4 - 9.0 (and possibly later)
  • Intel C++ Compiler 17.0.2 (and possibly later)
  • Microsoft Visual C++ 2015 / Build Tools 14.0.25123.0 (and possibly later)
  • Microsoft Visual C++ 2017 / Build Tools (and possibly later)
  • Microsoft Visual C++ 2019 / Build Tools 16.3.1+1def00d3d (and possibly later)

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

Please note:

  • GCC 4.8 has a bug 57824): multiline raw strings cannot be the arguments to macros. Don't use multiline raw strings directly in macros with this compiler.

  • 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
    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.

  • Unsupported versions of GCC and Clang are rejected by #error directives. This can be switched off by defining JSON_SKIP_UNSUPPORTED_COMPILER_CHECK. Note that you can expect no support in this case.

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

CompilerOperating SystemVersion String
GCC 4.8.5Ubuntu 14.04.5 LTSg++-4.8 (Ubuntu 4.8.5-2ubuntu1~14.04.2) 4.8.5
GCC 4.8.5CentOS Release-7-6.1810.2.el7.centos.x86_64g++ (GCC) 4.8.5 20150623 (Red Hat 4.8.5-36)
GCC 4.9.2 (armv7l)Raspbian GNU/Linux 8 (jessie)g++ (Raspbian 4.9.2-10+deb8u2) 4.9.2
GCC 4.9.4Ubuntu 14.04.1 LTSg++-4.9 (Ubuntu 4.9.4-2ubuntu1~14.04.1) 4.9.4
GCC 5.3.1 (armv7l) Ubuntu 16.04 LTSg++ (Ubuntu/Linaro 5.3.1-14ubuntu2) 5.3.1 20160413
GCC 5.5.0Ubuntu 14.04.1 LTSg++-5 (Ubuntu 5.5.0-12ubuntu1~14.04) 5.5.0 20171010
GCC 6.3.0Debian 9 (stretch)g++ (Debian 6.3.0-18+deb9u1) 6.3.0 20170516
GCC 6.3.1Fedora release 24 (Twenty Four)g++ (GCC) 6.3.1 20161221 (Red Hat 6.3.1-1)
GCC 6.4.0Ubuntu 14.04.1 LTSg++-6 (Ubuntu 6.4.0-17ubuntu1~14.04) 6.4.0 20180424
GCC 7.3.0Ubuntu 14.04.1 LTSg++-7 (Ubuntu 7.3.0-21ubuntu1~14.04) 7.3.0
GCC 7.3.0Windows Server 2012 R2 (x64)g++ (x86_64-posix-seh-rev0, Built by MinGW-W64 project) 7.3.0
GCC 8.1.0Ubuntu 14.04.1 LTSg++-8 (Ubuntu 8.1.0-5ubuntu1~14.04) 8.1.0
 GCC 9.2.1Ubuntu 14.05.1 LTS g++-9 (Ubuntu 9.2.1-16ubuntu1~14.04.1) 9.2.1 20191030
Clang 3.5.0Ubuntu 14.04.1 LTSclang version 3.5.0-4ubuntu2~trusty2 (tags/RELEASE_350/final) (based on LLVM 3.5.0)
Clang 3.6.2Ubuntu 14.04.1 LTSclang version 3.6.2-svn240577-1~exp1 (branches/release_36) (based on LLVM 3.6.2)
Clang 3.7.1Ubuntu 14.04.1 LTSclang version 3.7.1-svn253571-1~exp1 (branches/release_37) (based on LLVM 3.7.1)
Clang 3.8.0Ubuntu 14.04.1 LTSclang version 3.8.0-2ubuntu3~trusty5 (tags/RELEASE_380/final)
Clang 3.9.1Ubuntu 14.04.1 LTSclang version 3.9.1-4ubuntu3~14.04.3 (tags/RELEASE_391/rc2)
Clang 4.0.1Ubuntu 14.04.1 LTSclang version 4.0.1-svn305264-1~exp1 (branches/release_40)
Clang 5.0.2Ubuntu 14.04.1 LTSclang version 5.0.2-svn328729-1~exp1~20180509123505.100 (branches/release_50)
Clang 6.0.1Ubuntu 14.04.1 LTSclang version 6.0.1-svn334776-1~exp1~20180726133705.85 (branches/release_60)
Clang 7.0.1Ubuntu 14.04.1 LTSclang version 7.0.1-svn348686-1~exp1~20181213084532.54 (branches/release_70)
Clang Xcode 8.3OSX 10.11.6Apple LLVM version 8.1.0 (clang-802.0.38)
Clang Xcode 9.0OSX 10.12.6Apple LLVM version 9.0.0 (clang-900.0.37)
Clang Xcode 9.1OSX 10.12.6Apple LLVM version 9.0.0 (clang-900.0.38)
Clang Xcode 9.2OSX 10.13.3Apple LLVM version 9.1.0 (clang-902.0.39.1)
Clang Xcode 9.3OSX 10.13.3Apple LLVM version 9.1.0 (clang-902.0.39.2)
Clang Xcode 10.0OSX 10.13.3Apple LLVM version 10.0.0 (clang-1000.11.45.2)
Clang Xcode 10.1OSX 10.13.3Apple LLVM version 10.0.0 (clang-1000.11.45.5)
Clang Xcode 10.2OSX 10.14.4Apple LLVM version 10.0.1 (clang-1001.0.46.4)
Visual Studio 14 2015Windows Server 2012 R2 (x64)Microsoft (R) Build Engine version 14.0.25420.1, MSVC 19.0.24215.1
Visual Studio 15 2017Windows Server 2012 R2 (x64)Microsoft (R) Build Engine version 15.9.21+g9802d43bc3, MSVC 19.16.27032.1
Visual Studio 16 2019Windows Server 2012 R2 (x64)Microsoft (R) Build Engine version 16.3.1+1def00d3d, MSVC 19.23.28106.4


The class is licensed under the MIT License:

Copyright © 2013-2019 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 class contains the UTF-8 Decoder from Bjoern Hoehrmann which is licensed under the MIT License (see above). Copyright © 2008-2009 Björn Hoehrmann bjoern@hoehrmann.de

The class contains a slightly modified version of the Grisu2 algorithm from Florian Loitsch which is licensed under the MIT License (see above). Copyright © 2009 Florian Loitsch

The class contains a copy of Hedley from Evan Nemerson which is licensed as CC0-1.0.


If you have questions regarding the library, I would like to invite you to open an issue at GitHub. Please describe your request, problem, or question as detailed as possible, and also mention the version of the library you are using as well as the version of your compiler and operating system. Opening an issue at GitHub allows other users and contributors to this library to collaborate. For instance, I have little experience with MSVC, and most issues in this regard have been solved by a growing community. If you have a look at the closed issues, you will see that we react quite timely in most cases.

Only if your request would contain confidential information, please send me an email. For encrypted messages, please use this key.


Commits by Niels Lohmann and releases are signed with this PGP Key.


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 and split the single header file into smaller chunks.
  • 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.
  • Fytch found a bug in the documentation.
  • Jay Sistar implemented a Meson build description.
  • Henry Lee fixed a warning in ICC and improved the iterator implementation.
  • Vincent Thiery maintains a package for the Conan package manager.
  • Steffen fixed a potential issue with MSVC and std::min.
  • Mike Tzou fixed some typos.
  • amrcode noted a misleading documentation about comparison of floats.
  • Oleg Endo reduced the memory consumption by replacing <iostream> with <iosfwd>.
  • dan-42 cleaned up the CMake files to simplify including/reusing of the library.
  • Nikita Ofitserov allowed for moving values from initializer lists.
  • Greg Hurrell fixed a typo.
  • Dmitry Kukovinets fixed a typo.
  • kbthomp1 fixed an issue related to the Intel OSX compiler.
  • Markus Werle fixed a typo.
  • WebProdPP fixed a subtle error in a precondition check.
  • Alex noted an error in a code sample.
  • Tom de Geus reported some warnings with ICC and helped fixing them.
  • Perry Kundert simplified reading from input streams.
  • Sonu Lohani fixed a small compilation error.
  • Jamie Seward fixed all MSVC warnings.
  • Nate Vargas added a Doxygen tag file.
  • pvleuven helped fixing a warning in ICC.
  • Pavel helped fixing some warnings in MSVC.
  • Jamie Seward avoided unnecessary string copies in find() and count().
  • Mitja fixed some typos.
  • Jorrit Wronski updated the Hunter package links.
  • Matthias Möller added a .natvis for the MSVC debug view.
  • bogemic fixed some C++17 deprecation warnings.
  • Eren Okka fixed some MSVC warnings.
  • abolz integrated the Grisu2 algorithm for proper floating-point formatting, allowing more roundtrip checks to succeed.
  • Vadim Evard fixed a Markdown issue in the README.
  • zerodefect fixed a compiler warning.
  • Kert allowed to template the string type in the serialization and added the possibility to override the exceptional behavior.
  • mark-99 helped fixing an ICC error.
  • Patrik Huber fixed links in the README file.
  • johnfb found a bug in the implementation of CBOR's indefinite length strings.
  • Paul Fultz II added a note on the cget package manager.
  • Wilson Lin made the integration section of the README more concise.
  • RalfBielig detected and fixed a memory leak in the parser callback.
  • agrianius allowed to dump JSON to an alternative string type.
  • Kevin Tonon overworked the C++11 compiler checks in CMake.
  • Axel Huebl simplified a CMake check and added support for the Spack package manager.
  • Carlos O'Ryan fixed a typo.
  • James Upjohn fixed a version number in the compilers section.
  • Chuck Atkins adjusted the CMake files to the CMake packaging guidelines and provided documentation for the CMake integration.
  • Jan Schöppach fixed a typo.
  • martin-mfg fixed a typo.
  • Matthias Möller removed the dependency from std::stringstream.
  • agrianius added code to use alternative string implementations.
  • Daniel599 allowed to use more algorithms with the items() function.
  • Julius Rakow fixed the Meson include directory and fixed the links to cppreference.com.
  • Sonu Lohani fixed the compilation with MSVC 2015 in debug mode.
  • grembo fixed the test suite and re-enabled several test cases.
  • Hyeon Kim introduced the macro JSON_INTERNAL_CATCH to control the exception handling inside the library.
  • thyu fixed a compiler warning.
  • David Guthrie fixed a subtle compilation error with Clang 3.4.2.
  • Dennis Fischer allowed to call find_package without installing the library.
  • Hyeon Kim fixed an issue with a double macro definition.
  • Ben Berman made some error messages more understandable.
  • zakalibit fixed a compilation problem with the Intel C++ compiler.
  • mandreyel fixed a compilation problem.
  • Kostiantyn Ponomarenko added version and license information to the Meson build file.
  • Henry Schreiner added support for GCC 4.8.
  • knilch made sure the test suite does not stall when run in the wrong directory.
  • Antonio Borondo fixed an MSVC 2017 warning.
  • Dan Gendreau implemented the NLOHMANN_JSON_SERIALIZE_ENUM macro to quickly define a enum/JSON mapping.
  • efp added line and column information to parse errors.
  • julian-becker added BSON support.
  • Pratik Chowdhury added support for structured bindings.
  • David Avedissian added support for Clang 5.0.1 (PS4 version).
  • Jonathan Dumaresq implemented an input adapter to read from FILE*.
  • kjpus fixed a link in the documentation.
  • Manvendra Singh fixed a typo in the documentation.
  • ziggurat29 fixed an MSVC warning.
  • Sylvain Corlay added code to avoid an issue with MSVC.
  • mefyl fixed a bug when JSON was parsed from an input stream.
  • Millian Poquet allowed to install the library via Meson.
  • Michael Behrns-Miller found an issue with a missing namespace.
  • Nasztanovics Ferenc fixed a compilation issue with libc 2.12.
  • Andreas Schwab fixed the endian conversion.
  • Mark-Dunning fixed a warning in MSVC.
  • Gareth Sylvester-Bradley added operator/ for JSON Pointers.
  • John-Mark noted a missing header.
  • Vitaly Zaitsev fixed compilation with GCC 9.0.
  • Laurent Stacul fixed compilation with GCC 9.0.
  • Ivor Wanders helped reducing the CMake requirement to version 3.1.
  • njlr updated the Buckaroo instructions.
  • Lion fixed a compilation issue with GCC 7 on CentOS.
  • Isaac Nickaein improved the integer serialization performance and implemented the contains() function.
  • past-due suppressed an unfixable warning.
  • Elvis Oric improved Meson support.
  • Matěj Plch fixed an example in the README.
  • Mark Beckwith fixed a typo.
  • scinart fixed bug in the serializer.
  • Patrick Boettcher implemented push_back() and pop_back() for JSON Pointers.
  • Bruno Oliveira added support for Conda.
  • Michele Caini fixed links in the README.
  • Hani documented how to install the library with NuGet.
  • Mark Beckwith fixed a typo.
  • yann-morin-1998 helped reducing the CMake requirement to version 3.1.
  • Konstantin Podsvirov maintains a package for the MSYS2 software distro.
  • remyabel added GNUInstallDirs to the CMake files.
  • Taylor Howard fixed a unit test.
  • Gabe Ron implemented the to_string method.
  • Watal M. Iwasaki fixed a Clang warning.
  • Viktor Kirilov switched the unit tests from Catch to doctest
  • Juncheng E fixed a typo.
  • tete17 fixed a bug in the contains function.
  • Xav83 fixed some cppcheck warnings.
  • 0xflotus fixed some typos.
  • Christian Deneke added a const version of json_pointer::back.
  • Julien Hamaide made the items() function work with custom string types.
  • Evan Nemerson updated fixed a bug in Hedley and updated this library accordingly.
  • Florian Pigorsch fixed a lot of typos.
  • Camille Bégué fixed an issue in the conversion from std::pair and std::tuple to json.
  • Anthony VH fixed a compile error in an enum deserialization.

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

Used third-party tools

The library itself consists 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!

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.


Character encoding

The library supports Unicode input as follows:

  • Only UTF-8 encoded input is supported which is the default encoding for JSON according to RFC 8259.
  • std::u16string and std::u32string can be parsed, assuming UTF-16 and UTF-32 encoding, respectively. These encodings are not supported when reading from files or other input containers.
  • Other encodings such as Latin-1 or ISO 8859-1 are not supported and will yield parse or serialization 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.
  • When you store strings with different encodings in the library, calling dump() may throw an exception unless json::error_handler_t::replace or json::error_handler_t::ignore are used as error handlers.

Comments in JSON

This library does not support comments. It does so for three reasons:

  1. Comments are not part of the JSON specification. You may argue that // or /* */ are allowed in JavaScript, but JSON is not JavaScript.

  2. This was not an oversight: Douglas Crockford wrote on this in May 2012:

    I removed comments from JSON because I saw people were using them to hold parsing directives, a practice which would have destroyed interoperability. I know that the lack of comments makes some people sad, but it shouldn't.

    Suppose you are using JSON to keep configuration files, which you would like to annotate. Go ahead and insert all the comments you like. Then pipe it through JSMin before handing it to your JSON parser.

  3. It is dangerous for interoperability if some libraries would add comment support while others don't. Please check The Harmful Consequences of the Robustness Principle on this.

This library will not support comments in the future. If you wish to use comments, I see three options:

  1. Strip comments before using this library.
  2. Use a different JSON library with comment support.
  3. Use a format that natively supports comments (e.g., YAML or JSON5).

Order of object keys

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 (integration) or nlohmann::fifo_map (integration).

Further 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 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 abort() calls. You can further control this behavior by defining JSON_THROW_USER´ (overriding throw), JSON_TRY_USER (overriding try), and JSON_CATCH_USER (overriding catch). Note that JSON_THROW_USER should leave the current scope (e.g., by throwing or aborting), as continuing after it may yield undefined behavior.

Execute unit tests

To compile and run the tests, you need to execute

$ mkdir build
$ cd build
$ cmake ..
$ cmake --build .
$ ctest --output-on-failure

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