JSON for Modern C++

JSON for Modern C++

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

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.


You can sponsor this library at GitHub Sponsors.

🏷️ Named Sponsors

Thanks everyone!


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)
Embedded (FetchContent)

Since CMake v3.11, FetchContent can be used to automatically download the repository as a dependency at configure type.



  GIT_REPOSITORY https://github.com/nlohmann/json.git
  GIT_TAG v3.7.3)

  add_subdirectory(${json_SOURCE_DIR} ${json_BINARY_DIR} EXCLUDE_FROM_ALL)

target_link_libraries(foo PRIVATE nlohmann_json::nlohmann_json)

Note: The repository https://github.com/nlohmann/json download size is huge. It contains all the dataset used for the benchmarks. You might want to depend on a smaller repository. For instance, you might want to replace the URL above by https://github.com/ArthurSonzogni/nlohmann_json_cmake_fetchcontent

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

🍺 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 nlohmann_json/x.y.z to your conanfile'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/[email protected]. 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.

If you are using build2, you can use the nlohmann-json package from the public repository https://cppget.org or directly from the package's sources repository. In your project's manifest file, just add depends: nlohmann-json (probably with some version constraints). If you are not familiar with using dependencies in build2, please read this introduction. Please file issues here if you experience problems with the packages.

If you are using wsjcpp, you can use the command wsjcpp install "https://github.com/nlohmann/json:develop" to get the latest version. Note you can change the branch ":develop" to an existing tag or another branch.

If you are using CPM.cmake, you can check this example. After adding CPM script to your project, implement the following snippet to your CMake:

    NAME nlohmann_json
    GITHUB_REPOSITORY nlohmann/json
    VERSION 3.9.1)


If you are using bare Makefiles, you can use pkg-config to generate the include flags that point to where the library is installed:

pkg-config nlohmann_json --cflags

Users of the Meson build system will also be able to use a system wide library, which will be found by pkg-config:

json = dependency('nlohmann_json', required: true)


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() 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 value_type is an integral type of 1, 2 or 4 bytes, which will be interpreted as UTF-8, UTF-16 and UTF-32 respectively. For instance, a std::vector<std::uint8_t>, or a std::list<std::uint16_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);

Custom data source

Since the parse function accepts arbitrary iterator ranges, you can provide your own data sources by implementing the LegacyInputIterator concept.

struct MyContainer {
  void advance();
  const char& get_current();

struct MyIterator {
    using difference_type = std::ptrdiff_t;
    using value_type = char;
    using pointer = const char*;
    using reference = const char&;
    using iterator_category = std::input_iterator_tag;

    MyIterator& operator++() {
        return *this;

    bool operator!=(const MyIterator& rhs) const {
        return rhs.target != target;

    reference operator*() const {
        return target.get_current();

    MyContainer* target = nullptr;

MyIterator begin(MyContainer& tgt) {
    return MyIterator{&tgt};

MyIterator end(const MyContainer&) {
    return {};

void foo() {
    MyContainer c;
    json j = json::parse(c);

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 a binary value is parsed; value is passed and can be safely moved away
bool binary(binary_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\", 42, true, 1.78]"_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.contains("foo")) {
  // there is an entry with key "foo"

// or via find and an iterator
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. You can switch off implicit conversions by defining JSON_USE_IMPLICIT_CONVERSIONS to 0 before including the json.hpp header. When using CMake, you can also achieve this by setting the option JSON_ImplicitConversions to OFF.

// 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.

Simplify your life with macros

If you just want to serialize/deserialize some structs, the to_json/from_json functions can be a lot of boilerplate.

There are two macros to make your life easier as long as you (1) want to use a JSON object as serialization and (2) want to use the member variable names as object keys in that object:

  • NLOHMANN_DEFINE_TYPE_NON_INTRUSIVE(name, member1, member2, ...) is to be defined inside of the namespace of the class/struct to create code for.
  • NLOHMANN_DEFINE_TYPE_INTRUSIVE(name, member1, member2, ...) is to be defined inside of the class/struct to create code for. This macro can also access private members.

In both macros, the first parameter is the name of the class/struct, and all remaining parameters name the members.


The to_json/from_json functions for the person struct above can be created with:

namespace ns {
    NLOHMANN_DEFINE_TYPE_NON_INTRUSIVE(person, name, address, age)

Here is an example with private members, where NLOHMANN_DEFINE_TYPE_INTRUSIVE is needed:

namespace ns {
    class address {
        std::string street;
        int housenumber;
        int postcode;

        NLOHMANN_DEFINE_TYPE_INTRUSIVE(address, street, housenumber, postcode)

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);

The library also supports binary types from BSON, CBOR (byte strings), and MessagePack (bin, ext, fixext). They are stored by default as std::vector<std::uint8_t> to be processed outside of the library.

// CBOR byte string with payload 0xCAFE
std::vector<std::uint8_t> v = {0x42, 0xCA, 0xFE};

// read value
json j = json::from_cbor(v);

// the JSON value has type binary
j.is_binary(); // true

// get reference to stored binary value
auto& binary = j.get_binary();

// the binary value has no subtype (CBOR has no binary subtypes)
binary.has_subtype(); // false

// access std::vector<std::uint8_t> member functions
binary.size(); // 2
binary[0]; // 0xCA
binary[1]; // 0xFE

// set subtype to 0x10

// serialize to MessagePack
auto cbor = json::to_msgpack(j); // 0xD5 (fixext2), 0x10, 0xCA, 0xFE

Supported compilers

Though it's 2021 already, the support for C++11 is still a bit sparse. Currently, the following compilers are known to work:

  • GCC 4.8 - 11.0 (and possibly later)
  • Clang 3.4 - 11.0 (and possibly later)
  • Apple Clang 9.1 - 12.3 (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 or strtof) 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, GitHub Actions, and CircleCI:

Compiler Operating System CI Provider
Apple Clang 10.0.1 (clang-1001.0.46.4); Xcode 10.2.1 macOS 10.14.4 Travis
Apple Clang 11.0.0 (clang-1100.0.33.12); Xcode 11.2.1 macOS 10.14.6 Travis
Apple Clang 11.0.3 (clang-1103.0.32.59); Xcode 11.4.1 macOS 10.15.4 GitHub Actions
Apple Clang 12.0.0 (clang-1200.0.22.7); Xcode 11.4.1 macOS 10.15.5 Travis
Clang 3.5.0 (3.5.0-4ubuntu2~trusty2) Ubuntu 14.04.5 LTS Travis
Clang 3.6.2 (3.6.2-svn240577-1~exp1) Ubuntu 14.04.5 LTS Travis
Clang 3.7.1 (3.7.1-svn253571-1~exp1) Ubuntu 14.04.5 LTS Travis
Clang 3.8.0 (3.8.0-2ubuntu3~trusty5) Ubuntu 14.04.5 LTS Travis
Clang 3.9.1 (3.9.1-4ubuntu3~14.04.3) Ubuntu 14.04.5 LTS Travis
Clang 4.0.1 (4.0.1-svn305264-1~exp1) Ubuntu 14.04.5 LTS Travis
Clang 5.0.2 (version 5.0.2-svn328729-1~exp1~20180509123505.100) Ubuntu 14.04.5 LTS Travis
Clang 6.0.1 (6.0.1-svn334776-1~exp1~20190309042707.121) Ubuntu 14.04.5 LTS Travis
Clang 7.1.0 (7.1.0-svn353565-1~exp1~20190419134007.64) Ubuntu 14.04.5 LTS Travis
Clang 7.5.0 (Ubuntu 7.5.0-3ubuntu1~18.04) Ubuntu 18.04.4 LTS Travis
Clang 9.0.0 (x86_64-pc-windows-msvc) Windows-10.0.17763 GitHub Actions
Clang 10.0.0 (x86_64-pc-windows-msvc) Windows-10.0.17763 GitHub Actions
GCC 4.8.5 (Ubuntu 4.8.5-4ubuntu8~14.04.2) Ubuntu 14.04.5 LTS Travis
GCC 4.9.4 (Ubuntu 4.9.4-2ubuntu1~14.04.1) Ubuntu 14.04.5 LTS Travis
GCC 5.5.0 (Ubuntu 5.5.0-12ubuntu1~14.04) Ubuntu 14.04.5 LTS Travis
GCC 6.3.0 (Debian 6.3.0-18+deb9u1) Debian 9 Circle CI
GCC 6.5.0 (Ubuntu 6.5.0-2ubuntu1~14.04.1) Ubuntu 14.04.5 LTS Travis
GCC 7.3.0 (x86_64-posix-seh-rev0, Built by MinGW-W64 project) Windows-6.3.9600 AppVeyor
GCC 7.5.0 (Ubuntu 7.5.0-3ubuntu1~14.04.1) Ubuntu 14.04.5 LTS Travis
GCC 7.5.0 (Ubuntu 7.5.0-3ubuntu1~18.04) Ubuntu 18.04.4 LTS GitHub Actions
GCC 8.4.0 (Ubuntu 8.4.0-1ubuntu1~14.04) Ubuntu 14.04.5 LTS Travis
GCC 9.3.0 (Ubuntu 9.3.0-11ubuntu0~14.04) Ubuntu 14.04.5 LTS Travis
GCC 10.1.0 (Arch Linux latest) Arch Linux Circle CI
MSVC 19.0.24241.7 (Build Engine version 14.0.25420.1) Windows-6.3.9600 AppVeyor
MSVC 19.16.27035.0 (15.9.21+g9802d43bc3 for .NET Framework) Windows-10.0.14393 AppVeyor
MSVC 19.25.28614.0 (Build Engine version 16.5.0+d4cbfca49 for .NET Framework) Windows-10.0.17763 AppVeyor
MSVC 19.25.28614.0 (Build Engine version 16.5.0+d4cbfca49 for .NET Framework) Windows-10.0.17763 GitHub Actions
MSVC 19.25.28614.0 (Build Engine version 16.5.0+d4cbfca49 for .NET Framework) with ClangCL 10.0.0 Windows-10.0.17763 GitHub Actions


The class is licensed under the MIT License:

Copyright © 2013-2021 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 [email protected]

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 and adjusted MSVC calls in the CI.
  • 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.
  • Yuriy Vountesmery noted a subtle bug in a preprocessor check.
  • Chen fixed numerous issues in the library.
  • Antony Kellermann added a CI step for GCC 10.1.
  • Alex fixed an MSVC warning.
  • Rainer proposed an improvement in the floating-point serialization in CBOR.
  • Francois Chabot made performance improvements in the input adapters.
  • Arthur Sonzogni documented how the library can be included via FetchContent.
  • Rimas Misevičius fixed an error message.
  • Alexander Myasnikov fixed some examples and a link in the README.
  • Hubert Chathi made CMake's version config file architecture-independent.
  • OmnipotentEntity implemented the binary values for CBOR, MessagePack, BSON, and UBJSON.
  • ArtemSarmini fixed a compilation issue with GCC 10 and fixed a leak.
  • Evgenii Sopov integrated the library to the wsjcpp package manager.
  • Sergey Linev fixed a compiler warning.
  • Miguel Magalhães fixed the year in the copyright.
  • Gareth Sylvester-Bradley fixed a compilation issue with MSVC.
  • Alexander “weej” Jones fixed an example in the README.
  • Antoine Cœur fixed some typos in the documentation.
  • jothepro updated links to the Hunter package.
  • Dave Lee fixed link in the README.
  • Joël Lamotte added instruction for using Build2's package manager.
  • Paul Jurczak fixed an example in the README.
  • Sonu Lohani fixed a warning.
  • Carlos Gomes Martinho updated the Conan package source.
  • Konstantin Podsvirov fixed the MSYS2 package documentation.
  • Tridacnid improved the CMake tests.
  • Michael fixed MSVC warnings.
  • Quentin Barbarat fixed an example in the documentation.
  • XyFreak fixed a compiler warning.
  • TotalCaesar659 fixed links in the README.
  • Tanuj Garg improved the fuzzer coverage for UBSAN input.
  • AODQ fixed a compiler warning.
  • jwittbrodt made NLOHMANN_DEFINE_TYPE_NON_INTRUSIVE inline.
  • pfeatherstone improved the upper bound of arguments of the NLOHMANN_DEFINE_TYPE_NON_INTRUSIVE/NLOHMANN_DEFINE_TYPE_INTRUSIVE macros.
  • Jan Procházka fixed a bug in the CBOR parser for binary and string values.
  • T0b1-iOS fixed a bug in the new hash implementation.
  • Matthew Bauer adjusted the CBOR writer to create tags for binary subtypes.
  • gatopeich implemented an ordered map container for nlohmann::ordered_json.
  • Érico Nogueira Rolim added support for pkg-config.
  • KonanM proposed an implementation for the NLOHMANN_DEFINE_TYPE_NON_INTRUSIVE/NLOHMANN_DEFINE_TYPE_INTRUSIVE macros.
  • Guillaume Racicot implemented string_view support and allowed C++20 support.
  • Alex Reinking improved CMake support for FetchContent.
  • Hannes Domani provided a GDB pretty printer.

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 by default. 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.

However, you can pass set parameter ignore_comments to true in the parse function to ignore // or /* */ comments. Comments will then be treated as whitespace.

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 try the type nlohmann::ordered_json. Alternatively, you can use a more sophisticated ordered map like tsl::ordered_map (integration) or nlohmann::fifo_map (integration).

Memory Release

We checked with Valgrind and the Address Sanitizer (ASAN) that there are no memory leaks.

If you find that a parsing program with this library does not release memory, please consider the following case and it maybe unrelated to this library.

Your program is compiled with glibc. There is a tunable threshold that glibc uses to decide whether to actually return memory to the system or whether to cache it for later reuse. If in your program you make lots of small allocations and those small allocations are not a contiguous block and are presumably below the threshold, then they will not get returned to the OS. Here is a related issue #1924.

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. Furthermore, you can define JSON_ASSERT(x) to replace calls to assert(x).
  • 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 .. -DJSON_BuildTests=On
$ cmake --build .
$ ctest --output-on-failure

Note that during the ctest stage, several JSON test files are downloaded from an external repository. If policies forbid downloading artifacts during testing, you can download the files yourself and pass the directory with the test files via -DJSON_TestDataDirectory=path to CMake. Then, no Internet connectivity is required. See issue #2189 for more information.

In case you have downloaded the library rather than checked out the code via Git, test cmake_fetch_content_configure. Please execute ctest -LE git_required to skip these tests. See issue #2189 for more information.

Some tests change the installed files and hence make the whole process not reproducible. Please execute ctest -LE not_reproducible to skip these tests. See issue #2324 for more information.

Note you need to call cmake -LE "not_reproducible|git_required" to exclude both labels. See issue #2596 for more information.

As Intel compilers use unsafe floating point optimization by default, the unit tests may fail. Use flag /fp:precise then.

Niels Lohmann
You may know me from my JSON library for C++.
Niels Lohmann
  • How can I use std::string_view as the json_key to

    How can I use std::string_view as the json_key to "operator []" ?

    //as follow, this will fail as the key_type pof map is std::string_view std::map<string_view, bool> mv{{"111",true}, {"222", false}, {"333", true}}; nlohmann::json j4{mv};

    //another: fail again, the std::string_view is the json_key nlohmann::json j5; std::string_view key = "my_key"sv; j5[key] = 100;

  • Add a SAX parser

    Add a SAX parser

    The library currently only supports DOM-like parsing. This does not scale when the input files are enormous (#927). I would like to discuss how a SAX-like parser could look like.

    My proposal (heavily motivated by RapidJSON) is as follows:

    struct SAX
        // a null value was read
        bool null();
        // a boolean value was read
        bool boolean(bool);
        // an integer number was read
        bool number_integer(number_integer_t);
        // an unsigned integer number was read
        bool number_unsigned(number_unsigned_t);
        // a floating-point number was read
        // the string parameter contains the raw number value
        bool number_float(number_float_t, const std::string&);
        // a string value was read
        bool string(const std::string&);
        // the beginning of an object was read
        // binary formats may report the number of elements
        bool start_object(std::size_t elements);
        // an object key was read
        bool key(const std::string&);
        // the end of an object was read
        bool end_object();
        // the beginning of an array was read
        // binary formats may report the number of elements
        bool start_array(std::size_t elements);
        // the end of an array was read
        bool end_array();
        // a binary value was read
        // examples are CBOR type 2 strings, MessagePack bin, and maybe UBJSON array<uint8t>
        bool binary(const std::vector<uint8_t>& vec);
        // a parse error occurred
        // the byte position and the last token are reported
        bool parse_error(int position, const std::string& last_token);

    Some remarks:

    • All functions return a bool: true if the parser should continue or false if the parser should stop processing the input.
    • The proposal covers parsing of JSON, but also of CBOR, MessagePack, and UBJSON. Therefore, it contains extensions like array or object sizes as well as a binary type which do not occur when parsing JSON.
    • The idea is that the user would implement the above struct (we need to discuss whether we make all functions virtual, define a default implementation, etc.) and pass it to new parse functions, e.g. void parse_json(SAX &sax); or void parse_ubjson(SAX &sax);.

    What do you think?

  • json.hpp:5746:32: error: 'to_string' is not a member of 'std'

    json.hpp:5746:32: error: 'to_string' is not a member of 'std'

    Hello. On Android i have this messages

    • json.hpp:5746:32: error: 'to_string' is not a member of 'std'
    • json.hpp:6911:36: error: 'strtold' is not a member of 'std'

    android ndk have no to_strign realization

  • operator T() considered harmful

    operator T() considered harmful

    operator T() considered harmful

    There's two subjects of importance that I'd like to tackle. This is the first one, I'll open another issue quickly (hopefully).

    The problem

    There's been several issues in the past related to the implicit conversion operator that the library provides. The current Readme demonstrates the following use-case:

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

    This will call json::operator T() with T = ns::person, so no problems here.

    You'd expect the following code to always work too:

    ns::person p2;
    p2 = j;

    And it works! Sometimes.

    If ns::person has a user-defined operator=, it will break with the following message:

    t.cpp: In function ‘int main(int, const char**)’:
    t.cpp:14:7: error: ambiguous overload for ‘operator=’ (operand types are ‘ns::person’ and ‘nlohmann::json {aka nlohmann::basic_json<>}’)
       p2 = j;
    t.cpp:6:6: note: candidate: ns::person& ns::person::operator=(int)
       ns::person& operator=(int) { return *this; }
    t.cpp:3:8: note: candidate: constexpr ns::person& ns::person::operator=(const ns::person&)
     struct ns::person
    t.cpp:3:8: note: candidate: constexpr ns::person& ns::person::operator=(ns::person&&)

    Hard to understand that error, and it's not something that can be fixed by the library. It's triggered by the compiler before any template machinery on our side.

    Now with such code:

    void need_precise_measurement(Milliseconds);
    // later on
    json j = json::parse(config);

    It can fail in two cases:

    1. Someone adds an overload for need_precise_measurement. Compiler error.
    2. Someone changes the type of the argument to Nanoseconds. Runtime error at best.

    Implicit conversions

    Implicit conversions make the library pleasant to use. Especially when implicitely converting to JSON. That's because we have a templated non-explicit constructor with lots of constraints to correctly handle user types. There's also a single type to convert to: nlohmann::json, so everything works.

    However, when converting implicitely from json, it's the complete opposite. If the compiler cannot decide which type it should implicit convert the nlohmann::json value to, you get a compiler error.

    And it could happen in a codebase months after writing the code, (e.g. when adding an operator= to a type).

    In the end the only consistent way to convert from a json value is to call json.get<T>.

    Proposed change

    I propose to simply and completely remove operator T().

    It will obviously break code, so it can only be done in a major version.

    Of course this change cannot be adopted right away, so I propose to add a deprecation warning inside operator T(), as we did for basic_json::basic_json(std::istream&) overloads.

    Then it would be reasonable to remove it in a future major version.

  • Soften the landing when dumping non-UTF8 strings (type_error.316 exception)

    Soften the landing when dumping non-UTF8 strings (type_error.316 exception)

    When dumping a json object that contains an std:string we could be thrown the type_error.316 exception if the string contains invalid UTF8 sequences. While this sounds like a reasonable thing to do, it actually causes more problems than it fixes. In all the cases where user-entered (unsafe) strings end up stored in nlohmann::json objects the developer now has to "do something" before assigning a string value to some nlohmann::json variable. This effectively renders the whole string assignment functionality unsafe and defeats its purpose.

    Below is the wrapper code I had to write in order to investigate the random crashes my application went through due to the 316 exception.

    // Declaration:
    std::string dump_json(const nlohmann::json &json, const int indent =-1, const char* file = __builtin_FILE(), int line = __builtin_LINE()) const;
    // Definition
    std::string MANAGER::dump_json(const nlohmann::json &json, const int indent, const char* file, int line) const {
        std::string result;
        try {
            result = json.dump(indent);
        catch (nlohmann::json::type_error& e) {
            bug("%s: %s (%s:%d)", __FUNCTION__, e.what(), file, line);
        return result;

    This led me to the discovery of the code in my application that was sending json formatted log events to the log server. The log event has to store user entered data and I would expect the dump function to deal with invalid UTF8 sequences.

    If I have to use my json dump wrapper everywhere in my application code then of what use is the original dump function to begin with? What is more, I'd actually have to enhance the wrapper function to iterate over all strings stored in the json object and do something about the possible invalid UTF8 sequences. Not very convenient.

    Therefore, I would propose the default behavior of the dump function to simply ignore (skip or replace) invalid UTF8 sequences and NOT throw. Or perhaps add a nothrow parameter to the string assignment = operator.

    If left like that, I probably won't be the last developer who assumed that the dump method is safe to use. After all, if the lib allows me to assign a value to the json object then how can it be that it lets me assign values that later invalidate the whole json? This is illogical.

    // nothrow assignment would look like this
    nlohmann::json j_nothrow = (std::nothrow) "\xF0\xA4\xAD\xC0"; // Does not throw.
    // I am not 100% sure if this can be done, but since the `new` operator can have such
    // a parameter I would suppose the assignment operator = could also be customized
    // like that.
    std::string dumped = j_nothrow.dump(); // Never throws, just works.
    nlohmann::json j_throw;
    try {
        j_throw = "\xF0\xA4\xAD\xC0"; // Throws immediately.
    catch (nlohmann::json::type_error& e) {
        // handle the invalid UTF8 string

    One more thing. Why not include the file and line from where the exception originated in the exception description? You can see in my above wrapper function how I am using __builtin_LINE() and __builtin_FILE() to store the file and line information from where the wrapper call was made. It is super useful when debugging and exception description is all about debugging.

  • Custom type registration : instrusive API

    Custom type registration : instrusive API

    Currently, the API for registering custom types is as follows:

    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) {

    It would be great if there was a MACRO-style registration a bit like what msgpack-c uses:

    struct person 
       std::string name;
       std::string address;
       int age;
       MSGPACK_DEFINE_MAP(name, address, age);

    or yas:

    struct   person
       std::string name;
       std::string address;
       int age;
        YAS_DEFINE_STRUCT_SERIALIZE("person", name, address, age);


    struct   person
       std::string name;
       std::string address;
       int age;
    YAS_DEFINE_INTRUSIVE_SERIALIZE("person", name, address, age);
  • Refactor/split it

    Refactor/split it

    Here it is.


    This PR is the follow-up of #643, and the second phase of my planned refactoring (there will likely be one more, but we shall discuss it in a separate thread, since it requires naming/structural changes).

    I've added a tool to amalgamate the library into a single header (like Catch does), @nlohmann pointed out this was necessary.

    There is a new rule in the Makefile: single_include, that builds, and executes the tool.

    The output is located in the single_include folder, at the root of the repo.

    The only requirement is that every included file that is part of the library must be with the following style:

    #include <will_not_be_inlined.hpp>
    #include "will_be_inlined.hpp"

    Each include must be done with a path relative to the source folder:

    // detail/conversions/from_json.hpp
    #include "detail/value_t.hpp" // good
    #include "../value_t.hpp" // NOPE, not refactor-proof

    I've also added one test for each source file (except json.hpp), to check that we didn't miss any include. This will be helpful for future refactoring.

  • How to make sure a string or string literal is a valid JSON?

    How to make sure a string or string literal is a valid JSON?

    I'm gonna be sending strings to the parser that might or might not be valid JSONs and I need a way to check they are. I'm programming for a Nintendo 3DS, but using try to catch an exception does not work and the application hangs anyway if it finds an error while parsing. What can I do?

  • Please add a Pretty-Print option for arrays to stay always in one line

    Please add a Pretty-Print option for arrays to stay always in one line

    Please add a Pretty-Printing option for arrays to stay always in one line (don't add lines) if dump() parameter > -1 or std::setw()) is set. So that only other JSON types than arrays are expanded.


    json j;
    j["person"]["name"] = "Glory Rainbow";
    j["person"]["age"] = 61;
    j["person"]["mind"] = "Easy going";
    j["color"]["month"] = { 1, 3, 5, 7, 8, 10, 12};
    j["color"]["order"] = {"red", "orange", "yellow", "green", "blue", "indigo", "violet"};
    std::cout << j.dump(4);

    Wanted output:

        "color": {
            "month":  [1, 3, 5, 7, 8, 10, 12],
            "order": ["red", "orange", "yellow", "green", "blue", "indigo", "violet"]
        } ,
        "person": {
            "age": 61,
            "name": "Glory Rainbow",
            "mind": "Easy going"

    Thank you!

  • Fail to build when including json.hpp as a system include

    Fail to build when including json.hpp as a system include

    • What is the issue you have?

    When including json.hpp if the include directory is marked as a system include (-isystem rather than -I) it will fail to build

    I have to turn on CMAKE_NO_SYSTEM_FROM_IMPORTED and force the include path to be used as -I instead of -isystem. -isystem is used whenever nlohmann is provided as part of an imported CMake target. This happened to me due to using nlohmann from hunter with LLVM for Windows.

    • Please describe the steps to reproduce the issue. Can you provide a small but working code example?

    Sample CMake script

    cmake_minimum_required(VERSION 3.14)
    add_executable(test main.cpp)
    target_include_directories(test SYSTEM PRIVATE "<path/to/nlohmann/json>")

    Simple C++

    #include <nlohmann/json.hpp>
    int main()
    // Don't even have to call anything
    return 0;
    • What is the expected behavior?

    Program compiles successfully

    • And what is the actual behavior instead?

    A ton of errors along the line of error: expected '>'enable_if_t<is_detected<mapped_type_t, ConstructibleObjectType>::value and

    Clang 9.0.0 but the same behavior occurs with 8.0.0/8.0.1

    This is with LLVM for Windows and happens with the ClangCL toolset for VS2019 and the Ninja generator in CMake 3.15+

    • Did you use a released version of the library or the version from the develop branch?

    Version 3.6.1 and 3.7.0 release versions have the same problem


  • CBOR byte string support

    CBOR byte string support

    Currently, the CBOR feature of the library only supports writing strings out as UTF-8 and doesn't support byte arrays as described in the CBOR standard. I'd like to implement COSE on top of this library, but in order to do that, byte string support is needed.

    I'm happy to submit a pull request for this, but I wanted to check in advance if this is something you'd be open to

  • Add CIFuzz CI GitHub action

    Add CIFuzz CI GitHub action

    Add CIFuzz workflow action to have fuzzers build and run on each PR.

    This is a service offered by OSS-Fuzz where json already runs. CIFuzz can help detect regressions and catch fuzzing build issues early, and has a variety of features (see the URL above). In the current PR the fuzzers gets build on a pull request and will run for 300 seconds.

  • Occur error when parse character '\'

    Occur error when parse character '\'


    When I run code as follows:

        std::string sData("{\"a\":\"b\\a\",\"b\":2}");
        nlohmann::json obj = nlohmann::json::parse(sData);

    It both throw error in version 3.7.3 and 3.11.2.

    After delete '' as follows:

        std::string sData("{\"a\":\"ba\",\"b\":2}");
        nlohmann::json obj = nlohmann::json::parse(sData);

    It will run successfuly.

    Reproduction steps

    Run code as follows:

        std::string sData("{\"a\":\"b\\a\",\"b\":2}");
        nlohmann::json obj = nlohmann::json::parse(sData);

    It both throw error in version 3.7.3 and 3.11.2.

    Expected vs. actual results

    expect: json::parse run success and no error occur.

    actually: error occur

    Minimal code example

    No response

    Error messages

    No response

    Compiler and operating system

    Microsoft Visual C++ 2019 / Build Tools 16.3.1+1def00d3d (and possibly later)

    Library version

    3.11.2 or 3.7.3


  • Added support for pretty printing predicate

    Added support for pretty printing predicate

    I have a few json formats that contain a lot of primitive data arrays, and the current pretty printing features would either add too many or too few newlines to make them human-readable. To take the example from the unit test, I want this:

      "foo": [1,2,3]

    instead of the 'too compressed' {"foo":[1,2,3]} or the 'too spacious'

      "foo": [

    This PR proposes injecting a custom predicate into dump to let the user decide when to pretty print with indentation and when not to. This way, the user can define predicates based on the contents (e.g. arrays with all primitive types, or objects with just one key or anything beyond indentation level 6). I'm pretty happy with the changes to serializer, except for the name dump_configured, but I'm not too happy with the quasi duplication in basic_json, but I did not want to resort to SFINEA to select the correct overload. Should the old overload instead call the new one?

    Pull request checklist

    Read the Contribution Guidelines for detailed information.

    • [x] Changes are described in the pull request, or an existing issue is referenced.
    • [x] The test suite compiles and runs without error.
    • [x] Code coverage is 100%. Test cases can be added by editing the test suite.
    • [x] The source code is amalgamated; that is, after making changes to the sources in the include/nlohmann directory, run make amalgamate to create the single-header files single_include/nlohmann/json.hpp and single_include/nlohmann/json_fwd.hpp. The whole process is described here.

    Please don't

    • The C++11 support varies between different compilers and versions. Please note the list of supported compilers. Some compilers like GCC 4.7 (and earlier), Clang 3.3 (and earlier), or Microsoft Visual Studio 13.0 and earlier are known not to work due to missing or incomplete C++11 support. Please refrain from proposing changes that work around these compiler's limitations with #ifdefs or other means.
    • Specifically, I am aware of compilation problems with Microsoft Visual Studio (there even is an issue label for this kind of bug). I understand that even in 2016, complete C++11 support isn't there yet. But please also understand that I do not want to drop features or uglify the code just to make Microsoft's sub-standard compiler happy. The past has shown that there are ways to express the functionality such that the code compiles with the most recent MSVC - unfortunately, this is not the main objective of the project.
    • Please refrain from proposing changes that would break JSON conformance. If you propose a conformant extension of JSON to be supported by the library, please motivate this extension.
    • Please do not open pull requests that address multiple issues.
  • decode(state, codep, byte) generates warnings.

    decode(state, codep, byte) generates warnings.


    Quite minor.. BUT

    json.hpp, line 18822 .. 1/ the third parameter is named byte, 'byte' is also a 'type' and may lead to confusion. It is also a public variable used in parse_error class which isn't an 8-bit value (json.hpp, line 4440)

    2/ JSON_ASSERT(byte < utf8d.size()); (json.hpp, line 18844) Since byte is declared as an uint8_t, its range is 0 < byte < 256. This means the assert on an array of size 400 will never fail. VS2019 sometimes produces a warning for this.

    3/ JSON_ASSERT(index < 400); (json.hpp, line 18852) Shouldn't this be JSON_ASSERT(index < utf8d.size());


    Reproduction steps

    "Build"->"Run Code Analyzer on Solution"

    Expected vs. actual results


    Minimal code example

    No response

    Error messages

    Warning	C28020	The expression '0<=_Param_(1)&&_Param_(1)<=400-1' is not true at this call.
    This message doesn't always show up, I ended up analyzing the code ("Build"->"Run Code Analyzer on Solution")

    Compiler and operating system

    Windows 10, Microsoft Visual Studio Professional 2019

    Library version

    version 3.10.4


  • Compile Error on g++ using get() function

    Compile Error on g++ using get() function


    When iterating over a nlohmann::json object and then casting the values using value.get<std::string>(), we observe that g++ gives a compile error. On clang and msvc everything compiles just fine.

    Replacing value.get<std::string>() with std::string(value) fixes the problem on g++. But I wonder why the get function does not work?

    Reproduction steps

    See https://godbolt.org/z/Gxv7dbsjW

    Expected vs. actual results

    I would expect the "minimal code example" to compile as is on g++. But it does not.

    Minimal code example

    // https://godbolt.org/z/Gxv7dbsjW
    #include <nlohmann/json.hpp>
    #include <iostream>
    template<class K, class V>
    void convert(const nlohmann::json& sourceData, std::map<K, V>& targetMap)
        for (const auto& [key, value] : sourceData.items())
            if constexpr(std::is_same_v<V, std::string>)
                targetMap.emplace(key, value.get<std::string>());
                //targetMap.emplace(key, std::string(value));  // Only way to fix g++ is to use this line
                throw std::runtime_error("Case not covered");
    int main()
        const nlohmann::json data = {
            {"Year2", "2022" },
            {"Year3", "2023" },
            {"Year9", "2029" }
        std::map<std::string, std::string> m;
        convert(data, m);
        for (const auto& [k, v] : m)
            std::cout << " - " << k << ": " << v << std::endl;
        return 0;

    Error messages

    <source>:10:42: warning: expected 'template' keyword before dependent template name [-Wmissing-template-keyword]
       10 |             targetMap.emplace(key, value.get<std::string>());
          |                                          ^~~
          |                                          template
    <source>:10:57: error: expected primary-expression before '>' token
       10 |             targetMap.emplace(key, value.get<std::string>());
          |                                                         ^
    <source>:10:59: error: expected primary-expression before ')' token
       10 |             targetMap.emplace(key, value.get<std::string>());
          |                                                           ^

    Compiler and operating system

    g++ (Linux)

    Library version

    3.11.1 & trunk


Parsing gigabytes of JSON per second
Parsing gigabytes of JSON per second

simdjson : Parsing gigabytes of JSON per second JSON is everywhere on the Internet. Servers spend a *lot* of time parsing it. We need a fresh approach

Nov 30, 2022
A C++11 or library for parsing and serializing JSON to and from a DOM container in memory.
A C++11 or library for parsing and serializing JSON to and from a DOM container in memory.

Branch master develop Azure Docs Drone Matrix Fuzzing --- Appveyor codecov.io Boost.JSON Overview Boost.JSON is a portable C++ library which provides

Dec 1, 2022
Header-only library for automatic (de)serialization of C++ types to/from JSON.

fuser 1-file header-only library for automatic (de)serialization of C++ types to/from JSON. how it works The library has a predefined set of (de)seria

Oct 20, 2022
Similar to C++ streams, but the stream elements are structured JSON data rather than characters.

JIOS : JSON Input Output Streams Similar to C++ streams, but the stream elements are structured JSON data rather than characters. Contents Features [P

Aug 16, 2019
Yet another JSON/YAML/BSON serialization library for C++.
Yet another JSON/YAML/BSON serialization library for C++.

ThorsSerializer Support for Json Yaml Bson NEW Benchmark Results Conformance mac linux Performance max linux For details see: JsonBenchmark Yet anothe

Oct 27, 2022
libcluon is a small and efficient, single-file and header-only library written in modern C++ to power microservices.

libcluon Linux & OSX Build (TravisCI) Win64 Build (AppVeyor) Test Coverage Coverity Analysis CII Best Practices libcluon is a small single-file, heade

Nov 19, 2022
A C++, header-only library for constructing JSON and JSON-like data formats, with JSON Pointer, JSON Patch, JSON Schema, JSONPath, JMESPath, CSV, MessagePack, CBOR, BSON, UBJSON

JSONCONS jsoncons is a C++, header-only library for constructing JSON and JSON-like data formats such as CBOR. For each supported data format, it enab

Nov 26, 2022
C++ framework for json-rpc (json remote procedure call)
C++ framework for json-rpc (json remote procedure call)

I am currently working on a new C++17 implementation -> json-rpc-cxx. Master Develop | libjson-rpc-cpp This framework provides cross platform JSON-RPC

Nov 7, 2022
C++ framework for json-rpc (json remote procedure call)
C++ framework for json-rpc (json remote procedure call)

I am currently working on a new C++17 implementation -> json-rpc-cxx. Master Develop | libjson-rpc-cpp This framework provides cross platform JSON-RPC

Nov 30, 2022
json-cpp is a C++11 JSON serialization library.

JSON parser and generator for C++ Version 0.1 alpha json-cpp is a C++11 JSON serialization library. Example #include <json-cpp.hpp> struct Foo {

Oct 30, 2022
This is a JSON C++ library. It can write and read JSON files with ease and speed.

Json Box JSON (JavaScript Object Notation) is a lightweight data-interchange format. Json Box is a C++ library used to read and write JSON with ease a

Nov 3, 2022
A convenience C++ wrapper library for JSON-Glib providing friendly syntactic sugar for parsing JSON

This library is a wrapper for the json-glib library that aims to provide the user with a trivial alternative API to the API provided by the base json-

Oct 19, 2022
json-build is a zero-allocation JSON serializer compatible with C89

json-build is a zero-allocation JSON serializer compatible with C89. It is inspired by jsmn, a minimalistic JSON tokenizer.

Nov 16, 2022
JSON for Modern C++
JSON for Modern C++

Design goals Sponsors Integration CMake Package Managers Pkg-config Examples JSON as first-class data type Serialization / Deserialization STL-like ac

Nov 25, 2022
A killer modern C++ library for interacting with JSON.
A killer modern C++ library for interacting with JSON.

JSON Voorhees Yet another JSON library for C++. This one touts new C++11 features for developer-friendliness, an extremely slow-speed parser and no de

Oct 26, 2022
JSON for Modern C++
JSON for Modern C++

Design goals Sponsors Integration CMake Package Managers Pkg-config Examples JSON as first-class data type Serialization / Deserialization STL-like ac

Nov 29, 2022
Simple, fully external, smart, fast, JSON-configurated, feature-rich Windows x86 DLL Memory Dumper with Code Generation. Written in Modern C++.

altdumper Simple, fully external, smart, fast, JSON-configurated, feature-rich Windows x86 DLL Memory Dumper with Code Generation. Written in Modern C

Sep 9, 2022
A modern day direct port of BOOM 2.02 for modern times. Aiming to tastefully continue the development of BOOM, in the style of TeamTNT.
A modern day direct port of BOOM 2.02 for modern times.  Aiming to tastefully continue the development of BOOM, in the style of TeamTNT.

ReBOOM ReBOOM is a continuation of the BOOM source port, version 2.02. what is it ReBOOM is a source port, directly ported from BOOM 2.02 with additio

Jul 27, 2022
Ultralightweight JSON parser in ANSI C

cJSON Ultralightweight JSON parser in ANSI C. Table of contents License Usage Welcome to cJSON Building Copying the source CMake Makefile Vcpkg Includ

Dec 3, 2022
JSON parser and generator for C/C++ with scanf/printf like interface. Targeting embedded systems.

JSON parser and emitter for C/C++ Features ISO C and ISO C++ compliant portable code Very small footprint No dependencies json_scanf() scans a string

Nov 27, 2022