C++ JSON Serialization

There is no reflection in C++. True. But if the compiler can't provide you the metadata you need, you can provide it yourself.

Let's start by making a property struct:

template<typename Class, typename T>
struct PropertyImpl {
    constexpr PropertyImpl(T Class::*aMember, const char* aName) : member{aMember}, name{aName} {}

    using Type = T;

    T Class::*member;
    const char* name;
};

template<typename Class, typename T>
constexpr auto property(T Class::*member, const char* name) {
    return PropertyImpl<Class, T>{member, name};
}

Of course, you also can have a property that takes a setter and getter instead of a pointer to member, and maybe read only properties for calculated value you'd like to serialize. If you use C++17, you can extend it further to make a property that works with lambdas.

Ok, now we have the building block of our compile-time introspection system.

Now in your class Dog, add your metadata:

struct Dog {
    std::string barkType;
    std::string color;
    int weight = 0;
    
    bool operator==(const Dog& rhs) const {
        return std::tie(barkType, color, weight) == std::tie(rhs.barkType, rhs.color, rhs.weight);
    }
    
    constexpr static auto properties = std::make_tuple(
        property(&Dog::barkType, "barkType"),
        property(&Dog::color, "color"),
        property(&Dog::weight, "weight")
    );
};

We will need to iterate on that list. To iterate on a tuple, there are many ways, but my preferred one is this:

template <typename T, T... S, typename F>
constexpr void for_sequence(std::integer_sequence<T, S...>, F&& f) {
    using unpack_t = int[];
    (void)unpack_t{(static_cast<void>(f(std::integral_constant<T, S>{})), 0)..., 0};
}

If C++17 fold expressions are available in your compiler, then for_sequence can be simplified to:

template <typename T, T... S, typename F>
constexpr void for_sequence(std::integer_sequence<T, S...>, F&& f) {
    (static_cast<void>(f(std::integral_constant<T, S>{})), ...);
}

This will call a function for each constant in the integer sequence.

If this method don't work or gives trouble to your compiler, you can always use the array expansion trick.

Now that you have the desired metadata and tools, you can iterate through the properties to unserialize:

// unserialize function
template<typename T>
T fromJson(const Json::Value& data) {
    T object;

    // We first get the number of properties
    constexpr auto nbProperties = std::tuple_size<decltype(T::properties)>::value;
    
    // We iterate on the index sequence of size `nbProperties`
    for_sequence(std::make_index_sequence<nbProperties>{}, [&](auto i) {
        // get the property
        constexpr auto property = std::get<i>(T::properties);

        // get the type of the property
        using Type = typename decltype(property)::Type;

        // set the value to the member
        // you can also replace `asAny` by `fromJson` to recursively serialize
        object.*(property.member) = Json::asAny<Type>(data[property.name]);
    });

    return object;
}

And for serialize:

template<typename T>
Json::Value toJson(const T& object) {
    Json::Value data;

    // We first get the number of properties
    constexpr auto nbProperties = std::tuple_size<decltype(T::properties)>::value;
    
    // We iterate on the index sequence of size `nbProperties`
    for_sequence(std::make_index_sequence<nbProperties>{}, [&](auto i) {
        // get the property
        constexpr auto property = std::get<i>(T::properties);

        // set the value to the member
        data[property.name] = object.*(property.member);
    });

    return data;
}

If you want recursive serialization and unserialization, you can replace asAny by fromJson.

Now you can use your functions like this:

Dog dog;

dog.color = "green";
dog.barkType = "whaf";
dog.weight = 30;

Json::Value jsonDog = toJson(dog); // produces {"color":"green", "barkType":"whaf", "weight": 30}
auto dog2 = fromJson<Dog>(jsonDog);

std::cout << std::boolalpha << (dog == dog2) << std::endl; // pass the test, both dog are equal!

Done! No need for run-time reflection, just some C++14 goodness!

This code could benefit from some improvement, and could of course work with C++11 with some adjustments.

Note that one would need to write the asAny function. It's just a function that takes a Json::Value and call the right as... function, or another fromJson.

Here's a complete, working example made from the various code snippet of this answer. Feel free to use it.

As mentioned in the comments, this code won't work with msvc. Please refer to this question if you want a compatible code: Pointer to member: works in GCC but not in VS2015


For that you need reflection in C/C++, which doesn't exist. You need to have some meta data describing the structure of your classes (members, inherited base classes). For the moment C/C++ compilers don't automatically provide that information in built binaries.

I had the same idea in mind, and I used GCC XML project to get this information. It outputs XML data describing class structures. I have built a project and I'm explaining some key points in this page :

Serialization is easy, but we have to deal with complex data structure implementations (std::string, std::map for example) that play with allocated buffers. Deserialization is more complex and you need to rebuild your object with all its members, plus references to vtables ... a painful implementation.

For example you can serialize like this:

    // Random class initialization
    com::class1* aObject = new com::class1();

    for (int i=0; i<10; i++){
            aObject->setData(i,i);
    }      

    aObject->pdata = new char[7];
    for (int i=0; i<7; i++){
            aObject->pdata[i] = 7-i;
    }
    // dictionary initialization
    cjson::dictionary aDict("./data/dictionary.xml");

    // json transformation
    std::string aJson = aDict.toJson<com::class1>(aObject);

    // print encoded class
    cout << aJson << std::endl ;

To deserialize data it works like this:

    // decode the object
    com::class1* aDecodedObject = aDict.fromJson<com::class1>(aJson);

    // modify data
    aDecodedObject->setData(4,22);

    // json transformation
    aJson = aDict.toJson<com::class1>(aDecodedObject);
   
    // print encoded class
    cout << aJson << std::endl ;

Ouptuts:

>:~/cjson$ ./main
{"_index":54,"_inner":  {"_ident":"test","pi":3.141593},"_name":"first","com::class0::_type":"type","com::class0::data":[0,1,2,3,4,5,6,7,8,9],"com::classb::_ref":"ref","com::classm1::_type":"typem1","com::classm1::pdata":[7,6,5,4,3,2,1]}
{"_index":54,"_inner":{"_ident":"test","pi":3.141593},"_name":"first","com::class0::_type":"type","com::class0::data":[0,1,2,3,22,5,6,7,8,9],"com::classb::_ref":"ref","com::classm1::_type":"typem1","com::classm1::pdata":[7,6,5,4,3,2,1]}
>:~/cjson$ 

Usually these implementations are compiler dependent (ABI Specification for example), and require external descriptions to work (GCCXML output), thus are not really easy to integrate to projects.


Using quicktype, you can generate C++ serializers and deserializers from JSON sample data.

For example, given the sample JSON:

{
  "breed": "Boxer",
  "age": 5,
  "tail_length": 6.5
}

quicktype generates:

#include "json.hpp"

namespace quicktype {
    using nlohmann::json;

    struct Dog {
        int64_t age;
        std::string breed;
        double tail_length;
    };


    inline json get_untyped(const json &j, const char *property) {
        if (j.find(property) != j.end()) {
            return j.at(property).get<json>();
        }
        return json();
    }
}

namespace nlohmann {

    inline void from_json(const json& _j, struct quicktype::Dog& _x) {
        _x.age = _j.at("age").get<int64_t>();
        _x.breed = _j.at("breed").get<std::string>();
        _x.tail_length = _j.at("tail_length").get<double>();
    }

    inline void to_json(json& _j, const struct quicktype::Dog& _x) {
        _j = json{{"age", _x.age}, {"breed", _x.breed}, {"tail_length", _x.tail_length}};
    }
}

To parse the Dog JSON data, include the code above, install Boost and json.hpp, then do:

Dog dog = nlohmann::json::parse(jsonString);

Does anything, easy like that, exists?? THANKS :))

C++ does not store class member names in compiled code, and there's no way to discover (at runtime) which members (variables/methods) class contains. In other words, you cannot iterate through members of a struct. Because there's no such mechanism, you won't be able to automatically create "JSONserialize" for every object.

You can, however, use any json library to serialize objects, BUT you'll have to write serialization/deserialization code yourself for every class. Either that, or you'll have to create serializeable class similar to QVariantMap that'll be used instead of structs for all serializeable objects.

In other words, if you're okay with using specific type for all serializeable objects (or writing serialization routines yourself for every class), it can be done. However, if you want to automatically serialize every possible class, you should forget about it. If this feature is important to you, try another language.

Tags:

C++

Json