JSON for Modern C++ 3.11.0
Merge branch 'release/3.11.0'
tree: e76a616b0b88ee97dc9a6ce1f5d74ca16afb5076
  1. .github/
  2. .reuse/
  3. cmake/
  4. docs/
  5. include/
  7. single_include/
  8. tests/
  9. tools/
  10. .clang-format
  11. .clang-tidy
  12. .gitignore
  13. ChangeLog.md
  14. CITATION.cff
  15. CMakeLists.txt
  17. Makefile
  18. meson.build
  19. nlohmann_json.natvis
  20. README.md
  21. wsjcpp.yml

JSON for Modern C++

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

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 code 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.

:office: Corporate Sponsor

:label: Named Sponsors

Thanks everyone!


:question: If you have a question, please check if it is already answered in the FAQ or the Q&A section. If not, please ask a new question there.

:books: If you want to learn more about how to use the library, check out the rest of the README, have a look at code examples, or browse through the help pages.

:construction: If you want to understand the API better, check out the API Reference.

:bug: If you found a bug, please check the FAQ if it is a known issue or the result of a design decision. Please also have a look at the issue list before you create a new issue. Please provide as much information as possible to help us understand and reproduce your issue.

There is also a docset for the documentation browsers Dash, Velocity, and Zeal that contains the full documentation as offline resource.


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

Beside the examples below, you may want to:

→ Check the documentation
→ Browse the standalone example files

Every API function (documented in the API Documentation) has a corresponding standalone example file. For example, the emplace() function has a matching emplace.cpp example file.

Read JSON from a file

The json class provides an API for manipulating a JSON value. To create a json object by reading a JSON file:

#include <fstream>
#include <nlohmann/json.hpp>
using json = nlohmann::json;

// ...

std::ifstream f("example.json");
json data = json::parse(f);

Creating json objects from JSON literals

Assume you want to create hard-code this literal JSON value in a file, as a json object:

  "pi": 3.141,
  "happy": true

There are various options:

// Using (raw) string literals and json::parse
json ex1 = json::parse(R"(
    "pi": 3.141,
    "happy": true

// Using user-defined (raw) string literals
using namespace nlohmann::literals;
json ex2 = R"(
    "pi": 3.141,
    "happy": true

// Using initializer lists
json ex3 = {
  {"happy", true},
  {"pi", 3.141},

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 string literal should be brought into scope with with using namespace nlohmann::literals; (see json::parse()).

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

// parse explicitly
auto j3 = json::parse(R"({"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 a 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 == R"(["foo", 1, true, 1.78])"_json;  // true

// other stuff
j.size();     // 4 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 describing 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 json = 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 the namespace of the class/struct to create code for.
  • NLOHMANN_DEFINE_TYPE_INTRUSIVE(name, member1, member2, ...) is to be defined inside 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, UBJSON, and BJData)

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, UBJSON (Universal Binary JSON Specification) and BJData (Binary JData) 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 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 2022 already, the support for C++11 is still a bit sparse. Currently, the following compilers are known to work:

  • GCC 4.8 - 12.0 (and possibly later)
  • Clang 3.4 - 15.0 (and possibly later)
  • Apple Clang 9.1 - 13.1 (and possibly later)
  • Intel C++ Compiler 17.0.2 (and possibly later)
  • Nvidia CUDA Compiler 11.0.221 (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)
  • Microsoft Visual C++ 2022 / Build Tools 19.30.30709.0 (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 AppVeyor, Drone CI, and GitHub Actions:

CompilerOperating SystemCI Provider
Apple Clang 10.0.1 (clang-1001.0.46.4); Xcode 10.3macOS 10.15.7GitHub Actions
Apple Clang 11.0.0 (clang-1100.0.33.12); Xcode 11.2.1macOS 10.15.7GitHub Actions
Apple Clang 11.0.0 (clang-1100.0.33.17); Xcode 11.3.1macOS 10.15.7GitHub Actions
Apple Clang 11.0.3 (clang-1103.0.32.59); Xcode 11.4.1macOS 10.15.7GitHub Actions
Apple Clang 11.0.3 (clang-1103.0.32.62); Xcode 11.5macOS 10.15.7GitHub Actions
Apple Clang 11.0.3 (clang-1103.0.32.62); Xcode 11.6macOS 10.15.7GitHub Actions
Apple Clang 11.0.3 (clang-1103.0.32.62); Xcode 11.7macOS 10.15.7GitHub Actions
Apple Clang 12.0.0 (clang-1200.0.32.2); Xcode 12macOS 10.15.7GitHub Actions
Apple Clang 12.0.0 (clang-1200.0.32.21); Xcode 12.1macOS 10.15.7GitHub Actions
Apple Clang 12.0.0 (clang-1200.0.32.21); Xcode 12.1.1macOS 10.15.7GitHub Actions
Apple Clang 12.0.0 (clang-1200.0.32.27); Xcode 12.2macOS 10.15.7GitHub Actions
Apple Clang 12.0.0 (clang-1200.0.32.28); Xcode 12.3macOS 10.15.7GitHub Actions
Apple Clang 12.0.0 (clang-1200.0.32.29); Xcode 12.4macOS 10.15.7GitHub Actions
Apple Clang 13.0.0 (clang-1300.0.29.3); Xcode 13.1macOS 12.3.1GitHub Actions
Apple Clang 13.0.0 (clang-1300.0.29.30); Xcode 13.2macOS 12.3.1GitHub Actions
Apple Clang 13.0.0 (clang-1300.0.29.30); Xcode 13.2.1macOS 12.3.1GitHub Actions
Apple Clang 13.1.6 (clang-1316.0.21.2); Xcode 13.3macOS 12.3.1GitHub Actions
Apple Clang 13.1.6 (clang-1316.; Xcode 13.3.1macOS 12.3.1GitHub Actions
Clang 3.5.2 (3.5.2-3ubuntu1)Ubuntu 20.04.3 LTSGitHub Actions
Clang 3.6.2 (3.6.2-3ubuntu2)Ubuntu 20.04.3 LTSGitHub Actions
Clang 3.7.1 (3.7.1-2ubuntu2)Ubuntu 20.04.3 LTSGitHub Actions
Clang 3.8.0 (3.8.0-2ubuntu4)Ubuntu 20.04.3 LTSGitHub Actions
Clang 3.9.1 (3.9.1-4ubuntu3~16.04.2)Ubuntu 20.04.3 LTSGitHub Actions
Clang 4.0.0 (4.0.0-1ubuntu1~16.04.2)Ubuntu 20.04.3 LTSGitHub Actions
Clang 5.0.0 (5.0.0-3~16.04.1)Ubuntu 20.04.3 LTSGitHub Actions
Clang 6.0.1 (6.0.1-14)Ubuntu 20.04.3 LTSGitHub Actions
Clang 7.0.1 (7.0.1-12)Ubuntu 20.04.3 LTSGitHub Actions
Clang 8.0.1 (8.0.1-9)Ubuntu 20.04.3 LTSGitHub Actions
Clang 9.0.1 (9.0.1-12)Ubuntu 20.04.3 LTSGitHub Actions
Clang 10.0.0 (10.0.0-4ubuntu1)Ubuntu 20.04.3 LTSGitHub Actions
Clang 10.0.0 with GNU-like command-lineWindows-10.0.17763GitHub Actions
Clang 11.0.0 with GNU-like command-lineWindows-10.0.17763GitHub Actions
Clang 11.0.0 with MSVC-like command-lineWindows-10.0.17763GitHub Actions
Clang 11.0.0 (11.0.0-2~ubuntu20.04.1)Ubuntu 20.04.3 LTSGitHub Actions
Clang 12.0.0 (12.0.0-3ubuntu1~20.04.3)Ubuntu 20.04.3 LTSGitHub Actions
Clang 13.0.1 (13.0.1-++20211015123032+cf15ccdeb6d5-1exp120211015003613.5)Ubuntu 20.04.3 LTSGitHub Actions
Clang 14.0.5-++20220603124341+2f0a69c32a4c-1exp120220603124352.149Ubuntu 20.04.3 LTSGitHub Actions
Clang 15.0.0 (15.0.0-++20220530052901+b7d2b160c3ba-1exp120220530172952.268)Ubuntu 20.04.3 LTSGitHub Actions
GCC 4.8.5 (Ubuntu 4.8.5-4ubuntu2)Ubuntu 20.04.3 LTSGitHub Actions
GCC 4.9.3 (Ubuntu 4.9.3-13ubuntu2)Ubuntu 20.04.3 LTSGitHub Actions
GCC 5.4.0 (Ubuntu 5.4.0-6ubuntu1~16.04.12)Ubuntu 20.04.3 LTSGitHub Actions
GCC 6.4.0 (Ubuntu 6.4.0-17ubuntu1)Ubuntu 20.04.3 LTSGitHub Actions
GCC 7.5.0 (Ubuntu 7.5.0-6ubuntu2)Ubuntu 20.04.3 LTSGitHub Actions
GCC 8.1.0 (i686-posix-dwarf-rev0, Built by MinGW-W64 project)Windows-10.0.17763GitHub Actions
GCC 8.1.0 (x86_64-posix-seh-rev0, Built by MinGW-W64 project)Windows-10.0.17763GitHub Actions
GCC 8.4.0 (Ubuntu 8.4.0-3ubuntu2)Ubuntu 20.04.3 LTSGitHub Actions
GCC 9.3.0 (Ubuntu 9.3.0-17ubuntu1~20.04)Ubuntu 20.04.3 LTSGitHub Actions
GCC 10.2.0 (Ubuntu 10.2.0-5ubuntu1~20.04)Ubuntu 20.04.3 LTSGitHub Actions
GCC 11.1.0Ubuntu (aarch64)Drone CI
GCC 11.1.0 (Ubuntu 11.1.0-1ubuntu1~20.04)Ubuntu 20.04.3 LTSGitHub Actions
GCC 13.0.0 13.0.0 20220605 (experimental)Ubuntu 20.04.3 LTSGitHub Actions
Intel C++ Compiler 2021.5.0.20211109Ubuntu 20.04.3 LTSGitHub Actions
NVCC 11.0.221Ubuntu 20.04.3 LTSGitHub Actions
Visual Studio 14 2015 MSVC 19.0.24241.7 (Build Engine version 14.0.25420.1)Windows-6.3.9600AppVeyor
Visual Studio 15 2017 MSVC 19.16.27035.0 (Build Engine version 15.9.21+g9802d43bc3 for .NET Framework)Windows-10.0.14393AppVeyor
Visual Studio 16 2019 MSVC 19.28.29912.0 (Build Engine version 16.9.0+57a23d249 for .NET Framework)Windows-10.0.17763GitHub Actions
Visual Studio 16 2019 MSVC 19.28.29912.0 (Build Engine version 16.9.0+57a23d249 for .NET Framework)Windows-10.0.17763AppVeyor
Visual Studio 17 2022 MSVC 19.30.30709.0 (Build Engine version 17.0.31804.368 for .NET Framework)Windows-10.0.20348GitHub Actions


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 a release as a dependency at configure time.



FetchContent_Declare(json URL https://github.com/nlohmann/json/releases/download/v3.11.0/json.tar.xz)

target_link_libraries(foo PRIVATE nlohmann_json::nlohmann_json)

Note: It is recommended to use the URL approach described above which is supported as of version 3.10.0. See https://json.nlohmann.me/integration/cmake/#fetchcontent for more information.

Supporting Both

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

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

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

Package Managers

:beer: If you are using OS X and Homebrew, just type brew install nlohmann-json and you're set. If you want the bleeding edge rather than the latest release, use brew install nlohmann-json --HEAD. See nlohmann-json for more information.

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 install the nlohmann-json package with vcpkg install nlohmann-json and follow the then displayed descriptions. Please see the vcpkg project for any issues regarding the packaging.

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

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

If you are using NuGet, you can use the package nlohmann.json. Please check this extensive description on how to use the package. Please file 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, you 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 MacPorts, execute sudo port install nlohmann-json to install the nlohmann-json package.

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)


The class is licensed under the MIT License:

Copyright © 2013-2022 Niels Lohmann

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

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


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

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

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

The class contains parts of Google Abseil which is licensed under the Apache 2.0 License.


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.

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-Monterey and iOS 10-15. 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.
  • To store wide strings (e.g., std::wstring), you need to convert them to a UTF-8 encoded std::string before, see an example.

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 may be 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 number type 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. Note the explanatory what() string of exceptions is not available for MSVC if exceptions are disabled, see #2824.

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.

If the test suite is not found, several test suites will fail like this:

TEST CASE:  check test suite is downloaded

json/tests/src/make_test_data_available.hpp:23: FATAL ERROR: REQUIRE( utils::check_testsuite_downloaded() ) is NOT correct!
  values: REQUIRE( false )
  logged: Test data not found in 'json/cmake-build-debug/json_test_data'.
          Please execute target 'download_test_data' before running this test suite.
          See <https://github.com/nlohmann/json#execute-unit-tests> for more information.


In case you have downloaded the library rather than checked out the code via Git, test cmake_fetch_content_configure will fail. 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.