json_dto

A small header-only library for converting data between json representation and c++ structs

Description

json_dto library is a small header-only helper for converting data between json representation and c++ structs. DTO here stands for data transfer object.

Programming language: C++

License: BSD 3-clause "New" or "Revised" License

Tags: JSON     Serialization     Header Only     C++14    

README

Table of Contents

• Table of Contents
• What Is json_dto?
• What's new?
  • v.0.2.13
  • v.0.2.12
  • v.0.2.11
  • v.0.2.10
  • v.0.2.9
  • v.0.2.8
  • v.0.2.7
  • v.0.2.6.2
  • v.0.2.6.1
  • v.0.2.6
  • v.0.2.5
  • v.0.2.4
  • v.0.2.3
  • v.0.2.2
  • v.0.2.1
  • v.0.2.0
• Obtain and build
  • Prerequisites
  • Obtaining
    • Cloning of Git Repository
    • MxxRu::externals recipe
  • Build
• How to use it?
  • Getting started
  • Non intrusive json_io()
  • Supported field types
  • Mandatory and optional fields
    • Mandatory fields
    • Optional fields
      • Optional fields and std::optional
  • Array support
    • Array fields
    • Arrays and to_json and from_json
  • Other types of containers
    • Multimaps and multisets
  • Nullable fields
  • Complex types
  • Inheritance
  • Validators
    • Standard validators
  • User defined IO
    • Overloading of read_json_value and write_json_value
    • Usage of Reader_Writer
      • Custom Reader_Writer with containers and nullable_t, and std::optional
    • Overloading of read_json_value/write_json_value for const_map_key_t/mutable_map_key_t
    • Custom Reader_Writer and mutable/const_map_key_t
• License

Created by gh-md-toc

What Is json_dto?

json_dto library is a small header-only helper for converting data between json representation and c++ structs. DTO here stands for data transfer object. It was made and used as a part of a larger project inside StiffStream. And since Fall 2016 is ready for public. We are still using it for working with JSON in various projects.

What's new?

v.0.2.13

A new overload to_json added that receives an instance of pretty_writer_params_t with parameters for RapidJSON's PrettyWriter:

my_data my_obj;
...
// Make default serialization, without pretty-writer:
std::string my_obj_image = json_dto::to_json(my_obj);

// Make serialization with pretty-writer:
std::string my_obj_pretty_image = json_dto::to_json(my_obj,
        json_dto::pretty_writer_params_t{}
                .indent_char(' ')
                .indent_char_count(3u)
                .format_options(rapidjson::kFormatSingleLineArray));

v.0.2.12

Functions mandatory, optional, optional_no_default, and optional_null now accepts const- and rvalue references. That can be useful for types that have to be only serializable. For example:

struct demo {
    const int priority_;

    std::vector<int> version() const { return { 1, 2, 3 }; }

    const std::string payload_;

    demo(int priority, std::string payload)
        :   priority_{priority}, payload_{std::move(payload)}
    {}

    template<typename Json_Io>
    void json_io(Json_Io & io) {
        io & json_dto::mandatory("priority", priority_)
            & json_dto::mandatory("version", version())
            & json_dto::mandatory("payload", payload_)
            ;
    }
};

Please note that this code will lead to a compilation error in an attempt to deserialize an instance of demo type.

The class template json_dto::binder_t was refactored and now it uses several new customization points in the implementation: binder_data_holder_t, binder_read_from_implementation_t and binder_write_to_implementation_t. Those customization points allow to add a new functionality without modifying the json-dto source code.

For example, a user now can do something like:

namespace tricky_stuff {

template<typename F> struct serialize_only_proxy {...};

template<typename F> serialize_only_proxy<F> serialize_only(const F & f) {...}

template<typename F> struct deserialize_only_proxy {...};

template<typename F> deserialize_only_proxy<F> deserialize_only(F & f) {...}

} // namespace tricky_stuff

namespace json_dto {

... // Several partial specializations of binder_data_holder_t,
    // binder_read_from_implementation_t and binder_write_to_implementation_t
    // for tricky_stuff::serialize_only_proxy and tricky_stuff::deserialize_only_proxy.

} // namespace json_dto


struct demo {
    const int priority_;

    std::vector<int> version() const { return { 1, 2, 3 }; }

    const std::string payload_;

    std::vector<std::string> obsolete_properties_;

    demo(int priority, std::string payload)
        :   priority_{priority}, payload_{std::move(payload)}
    {}

    template<typename Json_Io>
    void json_io(Json_Io & io) {
        io & json_dto::mandatory("priority",
                    tricky_stuff::serialize_only(priority_))
            & json_dto::mandatory("version",
                    tricky_stuff::serialize_only(version()))
            & json_dto::mandatory("payload", payload_)
            & json_dto::optional_no_default("properties",
                    tricky_stuff::deserialize_only(obsolete_properties_)
            ;
    }
};

Several examples of how stuff like that can be implemented are shown in json_dto's samples folder: [serialize_only implementation](./dev/sample/tutorial20.1/main.cpp), [deserialize_only implementation](./dev/sample/tutorial20.2/main.cpp), [ignore_after_deserialization implementation](./dev/sample/tutorial20.3/main.cpp).

Such functions like serialize_only and deserialize_only can be useful in data-transformation code. For example, when we have to read some old data in JSON format, modify the data read and write it in a slightly different JSON.

v.0.2.11

New types mutable_map_key_t<T> and const_map_key_t<T> are now used for (de)serializing keys of map-like structures. See the description below.

A new Reader_Writer proxy apply_to_content_t added to address an issue of using custom Reader_Writers to the content of containers, nullable_t and std::optional. See the description below.

v.0.2.10

Another way of custom read/write operations added. It's based on specifying an instance of some user-supplied Reader_Writer type in the description of a field. See Usage of Reader_Writer for more details.

For support of that feature new overloads of json_dto::mandatory, json_dto::optional, and json_dto::optional_no_default have been added.

There is also a new json_dto::write_json_value overload:

void write_json_value(
    const rapidjson::Value::StringRefType & s,
    rapidjson::Value & object,
    rapidjson::MemoryPoolAllocator<> & allocator );

Please note that json_dto::string_ref_t is just an alias for rapidjson::Value::StringRefType.

v.0.2.9

New overloads for from_json function:

// Parses null-terminated string and returns a new object.
template<typename Type, unsigned Rapidjson_Parseflags = rapidjson::kParseDefaultFlags>
Type from_json( const char * json );

// Parses null-terminated string into alredy existed object.
template<typename Type, unsigned Rapidjson_Parseflags = rapidjson::kParseDefaultFlags>
void from_json( const char * json, Type & o );

// Parses a string-view and returns a new object.
// NOTE. string_ref_t is just an alias for RapidJSON's StringRefType.
template<typename Type, unsigned Rapidjson_Parseflags = rapidjson::kParseDefaultFlags>
Type from_json( const string_ref_t & json );

// Parses a string-view into alredy existed object.
template<typename Type, unsigned Rapidjson_Parseflags = rapidjson::kParseDefaultFlags>
void from_json( const string_ref_t & json, Type & o );

Versions with string_ref_t arguments are intended to be used in cases where a part of existing buffer should be parsed. For example:

std::vector<char> pdu = extract_data();

// string_view from C++17 is used just for a demonstration.
string_view headers;
string_view payload;
std::tie(headers, payload) = split_pdu_to_headers_and_payload(&pdu.front(), pdu.size());

auto parsed_payload = json_dto::from_json<PayloadType>(
        json_dto::make_string_ref(payload.data(), payload.size());

Versions with const char * are added to resolve ambious overloads with from_string(const std::string &) and from_string(const string_ref_t &) in the following cases:

auto payload = json_dto::from_json<PayloadType>(R"JSON({"id":10})JSON");

Please note that string_ref_t is not std::string_view. string_ref_t is just an alias for RapidJSON's StringRefType. And type StringRefType has a constructor is the form:

StringRefType(const char * ch, SizeType length);

But RapidJSON's SizeType is not std::size_t. So if someone writes:

std::vector<char> payload{...};
auto data = json_dto::from_json<PayloadType>(
        json_dto::string_ref_t{&payload.front(), payload.size()} );

there could be a warning from the compiler about narrowing of std::size_t to SizeType.

A set on make_string_ref functions is added to json_dto to avoid such warnings:

string_ref_t make_string_ref(const char * v);
string_ref_t make_string_ref(const char * v, std::size_t length);
string_ref_t make_string_ref(const std::string & v);

Use those functions to avoid warnings from the compiler:

std::vector<char> payload{...};
auto data = json_dto::from_json<PayloadType>(
        json_dto::make_string_ref(&payload.front(), payload.size()) );

v.0.2.8

Support for STL containers like std::deque, std::list, std::forward_list, std::set, std::unordered_set, std::map and std::unordered_map is implemented. These types can be used as types of fields in a serialized type, for example:

#include <json_dto/pub.hpp>

#include <deque>
#include <set>
#include <map>

struct my_message {
    std::deque<int> ids_;
    std::set<std::string> tags_;
    std::map<std::string, some_another_type> props_;
    ...
    template<typename Json_Io>
    void json_io(Json_Io & io) {
        io & json_dto::mandatory("ids", ids_)
            & json_dto::mandatory("tags", tags_)
            & json_dto::mandatory("properties", props_)
            ...
            ;
    }
};

These types can also be used with json_dto::from_json() and json_dto::to_json() functions:

auto messages = json_dto::from_json< std::forward_list<my_message> >(...);
...
auto json = json_dto::to_json(messages);

A new example tutorial17 added. This example shows the usage of new features.

An important note about support for std::multiset, std::unordered_multiset, std::multimap and std::unordered_multimap: those containers are also supported. But json_dto doesn't do any checks for duplicate keys. In that aspect, json_dto relies on RapidJSON behavior. For example, if an instance of std::multimap contains several values for some key all those values will be serialized. What happens to those values is dependent on RapidJSON.

v.0.2.7

Two new forms of from_json added. It is possible now to deserialize a DTO from already parsed document. For example:

struct update_period {
    ...
    template<typename Json_Io> void json_io(Json_Io & io) {...}
};
struct read_sensor {
    ...
    template<typename Json_Io> void json_io(Json_Io & io) {...}
};
...
void parse_and_handle_message( const std::string & raw_msg )
{
    rapidjson::Document whole_msg;
    whole_msg.Parse< rapidjson::kParseDefaultFlags >( raw_msg );
    if( whole_msg.HasParseError() )
        throw std::runtime_error(
                std::string{ "unable to parse message: " } +
                rapidjson::GetParseError_En( whole_msg.GetParseError() ) );

    const std::string msg_type = whole_msg[ "message_type" ].GetString();
    const auto & payload = whole_msg[ "payload" ];
    if( "Update-Period" == msg_type )
    {
        auto dto = json_dto::from_json< update_period >( payload );
        ...
    }
    else if( "Read-Sensor" == msg_type )
    {
        auto dto = json_dto::from_json< read_sensor >( payload );
        ...
    }
    else
        ...
}

Fix: compilation problems on FreeBSD 12 with clang-6.0.1.

v.0.2.6.2

Fix: add check for reading fields of DTO to ensure that a source JSON value is of type Object.

v.0.2.6.1

Improve std::optional availability check.

v.0.2.6

Support for std::vector for json_dto::to_json and json_dto::from_json functions.

v.0.2.5

Modify cmake-scripts for vcpkg port (target name json-dto::json-dto).

v.0.2.4

Add cmake support.

Make string value setter independent to RAPIDJSON_HAS_STDSTRING.

v.0.2.3

Bug fix in support of std::vector<bool>.

v.0.2.2

Bug fix in implementation of std::optional support.

New example tutorial6.1 added.

v.0.2.1

Some code style changes to meet expectations of some users.

v.0.2.0

New format of read_json_value function. NOTE: this is a breaking change!

Support for std::optional (and std::experimental::optional) added. Note: this may require to specify C++17 standard in compiler params (like /std:c++17 for MSVC or -std=c++17 for GCC).

Obtain and build

Prerequisites

To use json_dto it is necessary to have:

• C++14 compiler (VC++15.0, GCC 5.4 or above, clang 4.0 or above)
• rapidjson

And for building with mxxru:

• rapidjson_mxxru (v.1.0.0 or above)
• Mxx_ru 1.6.13 or above

And for running test:

• CATCH2 2.5.0 or above

Obtaining

Assuming that Git and Mxx_ru are already installed.

Cloning of Git Repository

git clone https://github.com/Stiffstream/json_dto.git

And then:

cd json_dto-0.2
mxxruexternals

to download and extract json_dto's dependencies.

MxxRu::externals recipe

For json_dto itself:

MxxRu::arch_externals :json_dto do |e|
  e.url 'https://github.com/Stiffstream/json_dto/archive/v.0.2.8.1.tar.gz'

  e.map_dir 'dev/json_dto' => 'dev'
end

For rapidjson and rapidjson_mxxru dependencies:

MxxRu::arch_externals :rapidjson do |e|
  e.url 'https://github.com/miloyip/rapidjson/archive/v1.1.0.zip'

  e.map_dir 'include/rapidjson' => 'dev/rapidjson/include'
end

MxxRu::arch_externals :rapidjson_mxxru do |e|
  e.url 'https://github.com/Stiffstream/rapidjson_mxxru/archive/v.1.0.1.tar.gz'

  e.map_dir 'dev/rapidjson_mxxru' => 'dev'
end

Build

While json_dto is header-only library test and samples require a build.

Compiling with Mxx_ru:

git clone https://github.com/Stiffstream/json_dto
cd json_dto
mxxruexternals
cd dev
ruby build.rb

NOTE. It might be necessary to set up MXX_RU_CPP_TOOLSET environment variable, see Mxx_ru documentation for further details.

How to use it?

An important notice: if you do not use Mxx_ru for building your project then add the following defines for your project:

RAPIDJSON_HAS_STDSTRING
RAPIDJSON_HAS_CXX11_RVALUE_REFS

If you use Mxx_ru and rapidjson_mxxru/prj.rb then these definitions will be added automatically.

Getting started

To start using json_dto simply include <json_dto/pub.hpp> header.

The usage principle of json_dto is borrowed from Boost serialization where rapidjson::Value plays the role of archive.

Let's assume we have a c++ structure that must be serialized to JSON and deserialized from JSON:

struct message_t
{
    std::string m_from;
    std::int64_t m_when;
    std::string m_text;
};

For integrating this struct with json_dto facilities the struct must be modified as follows:

struct message_t
{
    std::string m_from;
    std::int64_t m_when;
    std::string m_text;

    // Entry point for json_dto.
    template<typename Json_Io>
    void json_io(Json_Io & io)
    {
        io & json_dto::mandatory("from", m_from)
            & json_dto::mandatory("when", m_when)
            & json_dto::mandatory("text", m_text);
    }
};

Here json_io() function is an entry point for json_dto library. It describes how to read the data from rapidjson::Value (that is usualy parsed from string) and how to set the data in rapidjson::Value. json_io() is a template function. It allows to have a single description for read and write operations. The template is instantiated with Json_Io=json_dto::json_input_t for reading dto from JSON-value and Json_Io=json_dto::json_output_t for writing dto to JSON-value. Both json_dto::json_input_t and json_dto::json_output_t override operator& for splitting io functionality.

There are also iostream-like overrides for operator<< and operator>>:

template<typename Dto>
json_input_t &
operator>>(json_input_t & i, Dto & v);

template<typename Dto>
inline json_output_t &
operator<<(json_output_t & o, const Dto & v);

But they are only helpful for top level read/write operations.

In general json_dto gets data from rapidjson::Value and puts the data into rapidjson::Value. So read/write operations look like this:

// Read
rapidjson::Document document;

// ...

json_dto::json_input_t jin{ document };

message_t msg;
jin >> msg;

// If no exceptions were thrown DTO contains data received from JSON.

// Write
rapidjson::Document document;

// ...

json_dto::json_output_t jout{ document, document.GetAllocator() };

const message_t msg = get_message();
jout << msg;

// If no exceptions were thrown document contains data received from DTO.

But usually it is enough to work with std::string objects, so json_dto comes with handy to/from string helpers:

template<typename Dto>
std::string
to_json(const Dto & dto);

template<typename Type>
Type
from_json(const std::string & json);

[See full example](./dev/sample/tutorial1/main.cpp).

[See full example without to/from string helpers](./dev/sample/tutorial1.1/main.cpp).

Non intrusive json_io()

When it is unwanted to add an extra function to C++ structure it is possible to use a non intrusive json_io() version. In previous example dto part will look like this:

struct message_t
{
    std::string m_from;
    std::int64_t m_when;
    std::string m_text;
};

namespace json_dto
{

template<typename Json_Io>
void json_io(Json_Io & io, message_t & msg)
{
    io & json_dto::mandatory("from", msg.m_from)
        & json_dto::mandatory("when", msg.m_when)
        & json_dto::mandatory("text", msg.m_text);
}

} /* namespace json_dto */

[See full example](./dev/sample/tutorial2/main.cpp).

Note that it is necessary to define json_io() in namespace json_dto.

Supported field types

Out of the box json_dto lib supports following types:

• Bool: bool;
• Numeric: std::int16_t, std::uint16_t, std::int32_t, std::uint32_t, std::int64_t, std::uint64_t, double;
• Strings: std::string
• C++17 specific: std::optional (or std::experimental::optional)

Example:

struct supported_types_t
{
    bool m_bool{ false };

    std::int16_t m_int16{};
    std::uint16_t m_uint16{};

    std::int32_t m_int32{};
    std::uint32_t m_uint32{};

    std::int64_t m_int64{};
    std::uint64_t m_uint64{};
    double m_double{};

    std::string m_string{};
};

namespace json_dto
{

template<typename Json_Io>
void json_io(Json_Io & io, supported_types_t & obj)
{
    io & json_dto::mandatory("bool", obj.m_bool)
        & json_dto::mandatory("int16", obj.m_int16)
        & json_dto::mandatory("uint16", obj.m_uint16)
        & json_dto::mandatory("int32", obj.m_int32)
        & json_dto::mandatory("uint32", obj.m_uint32)
        & json_dto::mandatory("int64", obj.m_int64)
        & json_dto::mandatory("uint64", obj.m_uint64)
        & json_dto::mandatory("double", obj.m_double)
        & json_dto::mandatory("string", obj.m_string);
}

} /* namespace json_dto */

[See full example](./dev/sample/tutorial3/main.cpp)

Mandatory and optional fields

Each data member (at least those of them which are considered to be present in JSON) in C++ struct binds to JSON field. Bind can be mandatory or optional. Optional bind is extended with default value, but it is also possible to set optional fields without defaults. Also it is possible to add a value validator to the bind.

Binds are created by mandatory(), optional() and optional_no_default() functions. These functions returns a field binder. Binder is an instantiation of binder_t template class which carries a part of internal logic capable for handling field input/output operations. With the help of binders Json_Io object understands how read, write and validate the underlying field.

Mandatory fields

Binders for mandatory fields are created via mandatory() function:

template<
        typename Field_Type,
        typename Validator = empty_validator_t>
auto mandatory(
    string_ref_t field_name,
    Field_Type & field,
    Validator validator = Validator{});

// Since v.0.2.10
template<
        typename Reader_Writer,
        typename Field_Type,
        typename Validator = empty_validator_t>
auto mandatory(
    Reader_Writer reader_writer,
    string_ref_t field_name,
    Field_Type & field,
    Validator validator = Validator{});

The parameter field_name is of type string_ref_t which is an alias for rapidjson::Value::StringRefType. Typically it is enough to pass std::string or char * args (see rapidjson documentation for further details). The parameter field is a reference to the instance of the field value. The parameter validator is optional and it sets validator on fields value. Validators will be described later. By default empty_validator_t is used, and as it says it does nothing.

Optional fields

Binders for optional fields are created via optional() and optional_no_default() functions:

template<
        typename Field_Type,
        typename Field_Default_Value_Type,
        typename Validator = empty_validator_t>
auto optional(
    string_ref_t field_name,
    Field_Type & field,
    Field_Default_Value_Type default_value,
    Validator validator = Validator{});

template<
        typename Field_Type,
        typename Validator = empty_validator_t>
auto optional_no_default(
    string_ref_t field_name,
    Field_Type & field,
    Validator validator = Validator{});

// Since v.0.2.10
template<
        typename Reader_Writer,
        typename Field_Type,
        typename Field_Default_Value_Type,
        typename Validator = empty_validator_t>
auto optional(
    Reader_Writer reader_writer,
    string_ref_t field_name,
    Field_Type & field,
    Field_Default_Value_Type default_value,
    Validator validator = Validator{});

// Since v.0.2.10
template<
        typename Reader_Writer,
        typename Field_Type,
        typename Validator = empty_validator_t>
auto optional_no_default(
    Reader_Writer reader_writer,
    string_ref_t field_name,
    Field_Type & field,
    Validator validator = Validator{});

Parameters for functions are pretty much the same as for mandatory() functon.

The only difference is the third parameter for optional() function, it defines default value for a field if it is not defined in JSON.

In case of reading DTO, if optional field has default value and JSON object doesn't define this field then default value is used. In case of writing DTO, if value equals to default then this field wouldn't be included in JSON.

For optional() there is a partial specification that accepts nullptr argument as default_value parameter, it is usefull for nullable_t<T> fields.

Example of using optional fields:

struct message_t
{
    std::string m_from;
    std::int64_t m_when;
    std::string m_text;
    std::string m_text_format;
    bool m_is_private{ false };
};

namespace json_dto
{

template<typename Json_Io>
void json_io(Json_Io & io, message_t & msg)
{
    io & json_dto::mandatory("from", msg.m_from)
        & json_dto::mandatory("when", msg.m_when)
        & json_dto::mandatory("text", msg.m_text)
        & json_dto::optional("text_format", msg.m_text_format, "text/plain")
        & json_dto::optional_no_default("is_private", msg.m_is_private);
}

} /* namespace json_dto */

[See full example](./dev/sample/tutorial4/main.cpp)

Optional fields and std::optional

Since v.0.2 it is possible to use C++17's std::optional template as a type for field. In this case std::nullopt can be passed as third argument to json_dto::optional() function:

struct email_data_t
{
    std::string m_from;
    std::string m_to;
    std::string m_subject;
    std::optional<std::vector<std::string>> m_cc;
    std::optional<std::vector<std::string>> m_bcc;
    ...
    template<typename Json_Io>
    void json_io(Json_Io & io)
    {
        io & json_dto::mandatory("from", m_from)
            & json_dto::mandatory("to", m_to)
            & json_dto::mandatory("subject", m_subject)
            & json_dto::optional("cc", m_cc, std::nullopt)
            & json_dto::optional("bcc", m_bcc, std::nullopt)
            ...
    }
};

Note. If a compiler doesn't have std::optional but have std::experimental::optional then std::experimental::optional and std::experimental::nullopt can be used.

Array support

Array fields

JSON arrays are supported by json_dto, but there is one very important limitation: all elements of the array must have the same type. To set up an array simply use std::vector<T>. If DTO member is of std::vector<T> type, then corresponding JSON field is considered to be an array. While for output the elements of the array-field will be automatically of the same type, for successful input it is mandatory that all elements of the array are convertible to vector value type.

Example for array-fields:

struct vector_types_t
{
    std::vector<bool> m_bool{};

    std::vector<std::int16_t> m_int16{};
    std::vector<std::uint16_t> m_uint16{};

    std::vector<std::int32_t> m_int32{};
    std::vector<std::uint32_t> m_uint32{};

    std::vector<std::int64_t> m_int64{};
    std::vector<std::uint64_t> m_uint64{};
    std::vector<double> m_double{};

    std::vector<std::string> m_string{};
};

namespace json_dto
{

template<typename Json_Io>
void json_io(Json_Io & io, vector_types_t & obj)
{
    io & json_dto::mandatory("bool", obj.m_bool)
        & json_dto::mandatory("int16", obj.m_int16)
        & json_dto::mandatory("uint16", obj.m_uint16)
        & json_dto::mandatory("int32", obj.m_int32)
        & json_dto::mandatory("uint32", obj.m_uint32)
        & json_dto::mandatory("int64", obj.m_int64)
        & json_dto::mandatory("uint64", obj.m_uint64)
        & json_dto::mandatory("double", obj.m_double)
        & json_dto::mandatory("string", obj.m_string);
}

} /* namespace json_dto */

[See full example](./dev/sample/tutorial5/main.cpp)

Arrays and to_json and from_json

Since v.0.2.6 it is possible to serialize array of objects into JSON by json_dto::to_json function. It is also possible to deserialize JSON with array of objects into std::vector by json_dto::from_json function. For example:

#include <json_dto/pub.hpp>

#include <iostream>
#include <algorithm>

struct data_t {
    std::string m_key;
    int m_value;

    template<typename Json_Io>
    void json_io(Json_Io & io) {
        io & json_dto::mandatory("key", m_key)
            & json_dto::mandatory("value", m_value);
    }
};

int main() {
    const std::string json_data{
        R"JSON(
            [{"key":"first", "value":32},
             {"key":"second", "value":15},
             {"key":"third", "value":80}]
        )JSON"
    };

    auto data = json_dto::from_json< std::vector<data_t> >(json_data);
    std::sort(data.begin(), data.end(),
        [](const auto & a, const auto & b) { return a.m_value < b.m_value; });

    std::cout << "Sorted data: " << json_dto::to_json(data) << std::endl;
}

Other types of containers

Since v.0.2.8 there is a support for STL containers like std::deque, std::list, std::forward_list, std::set, std::unordered_set, std::map and std::unordered_map. Those types can be used as types of fields of serialized struct/classes:

#include <json_dto/pub.hpp>

#include <deque>
#include <set>
#include <map>

struct my_message {
    std::deque<int> ids_;
    std::set<std::string> tags_;
    std::map<std::string, some_another_type> props_;
    ...
    template<typename Json_Io>
    void json_io(Json_Io & io) {
        io & json_dto::mandatory("ids", ids_)
            & json_dto::mandatory("tags", tags_)
            & json_dto::mandatory("properties", props_)
            ...
            ;
    }
};

Also STL containers are supported by json_dto::from_json() and json_dto::to_json() functions:

auto messages = json_dto::from_json< std::forward_list<my_message> >(...);
...
auto json = json_dto::to_json(messages);

[See a special example with usage of STL containers](./dev/sample/tutorial17/main.cpp)

Note that support for those STL-containers is not hardcoded in json_dto. Instead, json_dto tries to detect a type of a container by inspecting the presence of types like value_type, key_type, mapped_type and methods like begin()/end(), emplace(), emplace_back() and so on. It means that json_dto may work not only with STL-containers but with other containers those mimics like STL-containers.

Note. Type std::array is not supported now. If you have to deal with std::array and want to have a support of it in json_dto please open an issue and we'll discuss some corner cases related to std::array.

Multimaps and multisets

An important note about support for std::multiset, std::unordered_multiset, std::multimap and std::unordered_multimap: those containers are also supported. But json_dto doesn't do any checks for duplicate keys. In that aspect, json_dto relies on RapidJSON behavior. For example, if an instance of std::multimap contains several values for some key all those values will be serialized. What happens to those values is dependent on RapidJSON.

Nullable fields

To support JSON null values, json_dto introduces nullable_t<T>. It is required that nullable field is explicitly defined as data member of type nullable_t<T>.

Interface of nullable_t<T> tries to mimic std::optional interface.

Example for nullable_t<T> field:

struct message_t
{
    message_t() {}

    message_t(
        std::string from,
        std::int64_t when,
        std::string text)
        :   m_from{ std::move(from) }
        ,   m_when{ when }
        ,   m_text{ std::move(text) }
    {}

    std::string m_from;
    std::int64_t m_when;
    std::string m_text;

    // Log level.
    // By default is constructed with null value.
    json_dto::nullable_t<std::int32_t> m_log_level{};
};

namespace json_dto
{

template<typename Json_Io>
void json_io(Json_Io & io, message_t & msg)
{
    io & json_dto::mandatory("from", msg.m_from)
        & json_dto::mandatory("when", msg.m_when)
        & json_dto::mandatory("text", msg.m_text)
        & json_dto::optional("log_level", msg.m_log_level, nullptr);
}

} /* namespace json_dto */

void
some_function( ... )
{
    // ...
    auto msg = json_dto::from_json<message_t>(json_data);

    // ...

    // If field is defined then its value can be obtained and used.
    if( msg.m_log_level )
        use_value(*msg.m_log_level);

    // ...

    msg.m_log_level = 1; // Set new value.

    // ...

    // equivalent to msg.m_log_level.reset();
    msg.m_log_level = nullptr; // Reset value.

    // ...
}

[See full example](./dev/sample/tutorial6/main.cpp)

Here default value for optional nullble field is nullptr. And it means that absence of value is a default state for a field. So when converting to JSON no-value nullable field wouldn't be included in JSON as "field":null piece.

Nullable fields can be used with arrays:

struct message_t
{
    message_t() {}

    message_t(
        std::string from,
        std::int64_t when,
        std::string text)
        :   m_from{ std::move(from) }
        ,   m_when{ when }
        ,   m_text{ std::move(text) }
    {}

    // Who sent a message.
    std::string m_from;

    // When the message was sent (unixtime).
    std::int64_t m_when;

    // Message text.
    std::string m_text;

    // Log level.
    // By default is constructed with null value.
    json_dto::nullable_t<std::int32_t> m_log_level{};

    json_dto::nullable_t< std::vector<std::string> > m_tags{};
};

namespace json_dto
{

template<typename Json_Io>
void json_io(Json_Io & io, message_t & msg)
{
    io & json_dto::mandatory("from", msg.m_from)
        & json_dto::mandatory("when", msg.m_when)
        & json_dto::mandatory("text", msg.m_text)
        & json_dto::optional("log_level", msg.m_log_level, nullptr)
        & json_dto::optional("tags", msg.m_tags, nullptr);
}

} /* namespace json_dto */

void some_function( ... )
{
    // ...
    auto msg = json_dto::from_json<message_t>(json_data);

    // ...

    if( msg.m_tags )
        use_tags(*msg.m_tags);

    // ...
}

void some_other_function( ... )
{
    message_t msg{ ... };
    // ...

    // Add tags:
    msg.m_tags.emplace(); // equivalent to msg = std::vector<std::string>{};
    msg.m_tags->emplace_back("sample");
    msg.m_tags->emplace_back("tutorial");

    // ...
}

[See full example](./dev/sample/tutorial7/main.cpp)

Complex types

json_dto allows to construct complex types with nested objects. Using nested objects is pretty much the same as using data of a simple types. Nested objects can be optional, nullable and be elements of array-fields. However there are some constraints:

• nested type must be itself integrated with json_dto;
• type must be default-constructible (for input);
• for optional fields with default value equality operator must be defined (more precisely an equality operator between nested type and type of passed default value).

Suppose there is a type which is already integrated with json_dto:

struct message_source_t
{
    std::int32_t m_thread_id{ 0 };
    std::string m_subsystem{};

    template<typename Json_Io>
    void json_io(Json_Io & io)
    {
        io & json_dto::optional("thread_id", m_thread_id, 0)
            & json_dto::mandatory("subsystem", m_subsystem);
    }
};

Then it can be used as a nested object in other type:

struct message_t
{
    message_source_t m_from;
    std::int64_t m_when;
    std::string m_text;

    template<typename Json_Io>
    void json_io(Json_Io & io)
    {
        io & json_dto::mandatory("from", m_from) // Exactly as with simple types.
            & json_dto::mandatory("when", m_when)
            & json_dto::mandatory("text", m_text);
    }
};

[See full example](./dev/sample/tutorial8/main.cpp)

[And see full example using nested objects as nullable and arrays](./dev/sample/tutorial9/main.cpp)

Inheritance

json_dto works well with inheritance. It is possible to use base implementation of json_io() function or completely override it.

For example derived class can use base class like this:

struct derived_t : public base_t
{
    //...

    template<typename Json_Io>
    void json_io(Json_Io & io)
    {
        base_t::json_io(io); // Run io on base class.

        // Run io on extra data:
        io & json_dto::mandatory("some_field", m_some_field)
            // ...
            ;
    }
};

However for easier maintenance it is recommended to use non intrusive json_io() function. Because if base class is integrated with json_dto in non intrusive manner, then the following wouldn't work:

    template<typename Json_Io>
    void json_io(Json_Io & io)
    {
        // Base class doesn't provide such member function.
        base_t::json_io(io); // Run io on base class.
        // ...
    }

So it is preferred to put inheritance this way:

struct message_source_t
{
    std::int32_t m_thread_id{ 0 };
    std::string m_subsystem{};
};

namespace json_dto
{

template<typename Json_Io>
void json_io(Json_Io & io, message_source_t & m)
{
    io & json_dto::optional("thread_id", m.m_thread_id, 0)
        & json_dto::mandatory("subsystem", m.m_subsystem);
}

} /* namespace json_dto */

struct message_t : public message_source_t
{
    std::int64_t m_when;
    std::string m_text;

    template<typename Json_Io>
    void json_io(Json_Io & io)
    {
        json_dto::json_io(io, static_cast<message_source_t &>(*this));

        io & json_dto::mandatory("when", m_when)
            & json_dto::mandatory("text", m_text);
    }
};

[See full example](./dev/sample/tutorial10/main.cpp)

Validators

json_dto allows to set validator on each field. Validator is a function object (an object of a type supporting an operator() member function) that receives a single parameter.

When handling input json_dto calls specified validator and passes resulting field value as an argument. If validator returns without throwing exception, then field value considered to be valid, and execution continues. Otherwise exception is catched and another will be thrown: json_dto::ex_t. This exeption contains original exception description supplemented with field name information.

When handling ouput json_dto calls specified validator before trying to assign field value of JSON object. In all other respects validation is the same as for input.

A simple example of using validators:

void check_all_7bit(
    const std::string & text)
{
    const auto it = std::find_if(std::begin(text), std::end(text),
            [](char c){ return c & 0x80; });

    if( std::end(text) != it )
    {
        throw std::runtime_error{
            "non 7bit char at pos " +
            std::to_string(std::distance(std::begin(text), it)) };
    }
}

struct message_t
{
    std::string m_from;
    std::int64_t m_when;

    // Message text. Must be 7bit ascii.
    std::string m_text;

    template<typename Json_Io>
    void json_io(Json_Io & io)
    {
        io & json_dto::mandatory("from", m_from)
            & json_dto::mandatory("when", m_when)
            & json_dto::mandatory("text", m_text, check_all_7bit);
    }
};

[See full example](./dev/sample/tutorial11/main.cpp)

Standard validators

json_dto comes with some useful ready to use validators for simple types. They are defined in <json_dto/pub.hpp> header.

Standard validators curently available:

• min_max_constraint_t<Num> - range validator, targeted for numeric types;
• one_of_validator_t<T> - validator for set of values.

Standard validators are template classes with overloaded operator(). And as they are template classes so for convenience for each validator there is an auxiliary function that helps deduce type of template instance from arguments:

template<typename Number>
auto min_max_constraint(Number min_value, Number max_value);

template<typename Field_Type>
auto one_of_constraint(std::initializer_list<Field_Type> values);

[See full example with standard validators](./dev/sample/tutorial12/main.cpp)

User defined IO

Overloading of read_json_value and write_json_value

It is possible to define custom IO logic for a specific type. It might be useful for types when using object is an overkill, for example time point that can be stored in format of 'YYYY.MM.DD hh:mm:ss' or some token composed of several small items like '--'. But introducing custom IO logic for some type requires to work with rapidjson API directly.

There are two way to introduce custom IO logic.

The first way uses C++'s Argument Dependent Lookup feature: an user should define read_json_value and write_json_value in the same namespace where types are defined. The right implementations of read_json_value and write_json_value will be found by C++ compiler automatically. For example:

namespace importance_levels
{

enum class level_t
    {
        low,
        normal,
        high
    };

// read_json_value and write_json_value for level_t are
// defined in importance_levels namespace.
// They will be found by argument dependent lookup.
void read_json_value(
    level_t & value,
    const rapidjson::Value & from)
{...}

void write_json_value(
    const level_t & value,
    rapidjson::Value & object,
    rapidjson::MemoryPoolAllocator<> & allocator)
{...}

} /* namespace importance_levels */

This approach also allows to define read_json_value and write_json_value for user's template type. For example:

namespace demo
{

template<typename T>
class some_template
{...}

template<typename T>
void read_json_value(
    some_template<T> & value,
    const rapidjson::Value & from)
{...}

template<typename T>
void write_json_value(
    const some_template<T> & value,
    rapidjson::Value & object,
    rapidjson::MemoryPoolAllocator<> & allocator)
{...}

} /* namespace demo */

struct my_data_t
{
    demo::some_template<int> m_first;
    demo::some_template<double> m_second;
    ...
    template<typename Json_Io>
    void json_io(Json_Io & io)
    {
        io & json_dto::mandatory("first", m_first)
            & json_dto::mandatory("second", m_second)
            ...
    }
};

[See full example with custom IO and ADL](./dev/sample/tutorial15/main.cpp)

The second way uses explicit template specialization for 2 functons inside json_dto namespace:

namespace json_dto
{

template<>
void read_json_value(
    Custom_Type & v,
    const rapidjson::Value & object)
{
    // ...
}

template<>
void write_json_value(
    const Custom_Type & v,
    rapidjson::Value & object,
    rapidjson::MemoryPoolAllocator<> & allocator)
{
    // ...
}

} /* namespace json_dto */

json_dto will consider these specializations for using with specified Custom_Type. This way can be used when it is impossible to place read_json_value and write_json_value into the namespace where the type if defined (for example if it is standard type like std::filesystem::path).

[See full example with custom IO](./dev/sample/tutorial14/main.cpp)

Usage of Reader_Writer

Suppose we have an enumeration log_level defined such way:

enum class log_level { low, normal, high };

And we have two structs that use that log_level enumeration:

struct log_message
{
    log_level level_;
    std::string msg_;
};

struct log_config
{
    std::string path_;
    log_level level_;
};

Serialization of log_level to JSON should use numeric values of log levels, e.g.: {"level":0, "msg":"..."}, but the serialization of log_config should use textual names instead of numeric values, e.g.: {"path":"/var/log/demo", "level":"low"}.

Such a task can't be implemented by writing overloads of read_json_value and write_json_value functions. Custom Reader_Writers should be used in that case:

struct numeric_log_level
{
    void read( log_level & v, const rapidjson::Value & from ) const
    {
        using json_dto::read_json_value;

        int actual;
        read_json_value( actual, from );

        v = static_cast<log_level>(actual);
    }

    void
    write(
        const log_level & v,
        rapidjson::Value & to,
        rapidjson::MemoryPoolAllocator<> & allocator ) const
    {
        using json_dto::write_json_value;

        const int actual = static_cast<int>(v);
        write_json_value( actual, to, allocator );
    }
};

struct log_message
{
    log_level level_;
    std::string msg_;

    template< typename Json_Io >
    void json_io( Json_Io & io )
    {
        io & json_dto::mandatory( numeric_log_level{}, "level", level_ )
            & json_dto::mandatory( "msg", msg_ );
    }
};

struct textual_log_level
{
    void read( log_level & v, const rapidjson::Value & from ) const
    {
        using json_dto::read_json_value;

        std::string str_v;
        read_json_value( str_v, from );

        if( "low" == str_v ) v = log_level::low;
        else if( "normal" == str_v ) v = log_level::normal;
        else if( "high" == str_v ) v = log_level::high;
        else throw json_dto::ex_t{ "invalid value for log_level" };
    }

    void
    write(
        const log_level & v,
        rapidjson::Value & to,
        rapidjson::MemoryPoolAllocator<> & allocator ) const
    {
        using json_dto::write_json_value;
        using json_dto::string_ref_t;

        switch( v )
        {
            case log_level::low:
                write_json_value( string_ref_t{ "low" }, to, allocator );
            break;

            case log_level::normal:
                write_json_value( string_ref_t{ "normal" }, to, allocator );
            break;

            case log_level::high:
                write_json_value( string_ref_t{ "high" }, to, allocator );
            break;
        }
    }
};

struct log_config
{
    std::string path_;
    log_level level_;

    template< typename Json_Io >
    void json_io( Json_Io & io )
    {
        io & json_dto::mandatory( "path", path_ )
            & json_dto::mandatory( textual_log_level{}, "level", level_ );
    }
};

Note that Reader_Writer class should have two const methods read and write those signatures are the same with the signatures of read_json_value and write_json_value functions.

[See full example with Reader_Writer](./dev/sample/tutorial18/main.cpp)

Custom Reader_Writer classes can also be used for handling non-standard representation of some values in JSON document. For example, sometimes string-values like "NAN" or "nan" are used for NaN (Not-a-Number) values. RapidJSON can only parsed special value NaN, but not "NAN" nor "nan" values. In such case a custom Reader_Writer like the following one can be used:

struct custom_floating_point_reader_writer
{
    template< typename T >
    void read( T & v, const rapidjson::Value & from ) const
    {
        if( from.IsNumber() )
        {
            json_dto::read_json_value( v, from );
            return;
        }
        else if( from.IsString() )
        {
            const json_dto::string_ref_t str_v{ from.GetString() };
            if( equal_caseless( str_v, "nan" ) )
            {
                v = std::numeric_limits<T>::quiet_NaN();
                return;
            }
            else if( equal_caseless( str_v, "inf" ) )
            {
                v = std::numeric_limits<T>::infinity();
                return;
            }
            else if( equal_caseless( str_v, "-inf" ) )
            {
                v = -std::numeric_limits<T>::infinity();
                return;
            }
        }

        throw json_dto::ex_t{ "unable to parse value" };
    }

    template< typename T >
    void
    write(
        T & v,
        rapidjson::Value & to,
        rapidjson::MemoryPoolAllocator<> & allocator ) const
    {
        using json_dto::write_json_value;
        using json_dto::string_ref_t;

        if( std::isnan(v) )
            write_json_value( string_ref_t{"nan"}, to, allocator );
        else if( v > std::numeric_limits<T>::max() )
            write_json_value( string_ref_t{"inf"}, to, allocator );
        else if( v < std::numeric_limits<T>::min() )
            write_json_value( string_ref_t{"-inf"}, to, allocator );
        else
            write_json_value( v, to, allocator );
    }
};

struct struct_with_floats_t
{
    float m_num_float;
    double m_num_double;

    template< typename Json_Io >
    void
    json_io( Json_Io & io )
    {
        io
            & optional( custom_floating_point_reader_writer{},
                    "num_float", m_num_float, 0.0f )
            & optional( custom_floating_point_reader_writer{},
                    "num_double", m_num_double, 0.0 );
    }
};

Note also that read and write methods of Reader_Writer class can be template methods.

Custom Reader_Writer with containers and nullable_t, and std::optional

If a custom Reader_Writer is used then a reference to the whole field is passed to Reader_Writer's methods. For example:

struct my_int_reader_writer
{
    void read(int & v, ...) const {...} // Custom read procedure for an int.

    void write(const int & v, ...) const {...} // Custom write procedure for int.
};
...
struct my_data
{
    int field_;
    ...
    template<typename Io> void json_io(Io & io)
    {
        io & json_dto::mandatory(my_int_reader_writer{},
                "field", field_)
            ...
            ;
    }
};

In that case a reference to an int will be passed to my_int_reader_writer's read and write methods.

In the case when my_data isn't int but a std::vector<int> then a reference to std::vector<int> instance will be passed to read/write. And there will be a compiler error because read/write expects a reference to an int.

If we want our custom Reader_Writer to be applied for every member of a container then json_dto::apply_to_content_t proxy should be used as Reader_Writer type:

struct my_complex_data
{
    std::vector<int> field_;
    ...
    template<typename Io> void json_io(Io & io)
    {
        io & json_dto::mandatory(
                json_dto::apply_to_content_t<my_int_reader_writer>{},
                "field", field_)
            ...
            ;
    }
};

The apply_to_content_t proxy works very simple way: it holds an instance of an actual Reader_Writer and applies that actual Reader_Writer to every member of a container (or to the content of json_dto::nullable_t and std::optional, see bellow).

The same rule is applied to nullable_t and std::optional:

struct my_data
{
    std::optional<int> field_;
    ...
    template<typename Io> void json_io(Io & io)
    {
        io & json_dto::optional(my_int_reader_writer{},
                "field", field_, std::nullopt)
            ...
            ;
    }
};

Such code leads to compiler error because my_int_reader_writer's read and write methods expect a reference to int, not to std::optional<int>. So we have to use apply_to_content_t here too:

struct my_data
{
    std::optional<int> field_;
    ...
    template<typename Io> void json_io(Io & io)
    {
        io & json_dto::optional(
                // Now my_int_reader_writer will be applied to the content
                // of std::optional<int>, not to std::optional<int> itself.
                json_dto::apply_to_content_t<my_int_reader_writer>{},
                "field", field_, std::nullopt)
            ...
            ;
    }
};

Note that apply_to_content_t can be nested:

struct my_complex_data {
    json_dto::nullable_t< std::vector<int> > params_;
    ...
    template<typename Io> void json_io(Io & io) {
        io & json_dto::mandatory(
                // The first occurence of apply_to_content_t is for nullable_t.
                json_dto::apply_to_content_t<
                    // The second occurence of apply_to_content_t is for std::vector.
                    json_dto::apply_to_content_t<
                        // This is for the content of std::vector.
                        my_int_reader_writer
                    >
                >{},
                "params", params_)
            ...
            ;
    }
};

Overloading of read_json_value/write_json_value for const_map_key_t/mutable_map_key_t

Since v.0.2.11 json_dto (de)serializes keys of map-like containers (std::map, std::multimap, std::unordered_map and so on) by using new proxy types const_map_key_t and mutable_map_key_t.

A new type const_map_key_t<T> is used for serializing a key of type T.

A new type mutable_map_key_t<T> is used for deserializing a key of type T.

It means that if someone wants to make overloads of read_json_value and write_json_value for types that are used as keys in map-like structures, then such overloads should be placed into json_dto namespace and should have the following prototypes:

namespace json_dto {

void read_json_value(
    mutable_map_key_t<UserType> key,
    const rapidjson::Value & from);

void write_json_value(
    const_map_key_t<UserType> key,
    rapidjson::Value & to,
    rapidjson::MemoryPoolAllocator<> & allocator);

} /* namespace json_dto */

[See full example with overloading of read/write_json_value for mutable/const_map_key_t](./dev/sample/tutorial19/main.cpp)

Custom Reader_Writer and mutable/const_map_key_t

The addition of mutable_map_key_t/const_map_key_t in the v.0.2.11 means that custom Reader_Writers should take the presence of those types into the account.

For example, if a custom Reader_Writer is used for (de)serializing a content of std::map then that Reader_Writer should have implementations of read/write methods for keys and values from the map:

struct my_kv_formatter
{
    // Read a key.
    void read(
        json_dto::mutable_map_key_t<KeyType> & key,
        const rapidjson::Value & from) const {...}

    // Read a value.
    void read(
        ValueType & value,
        const rapidjson::Value & from) const {...}

    // Write a key.
    void write(
        const json_dto::const_map_key_t<KeyType> & key,
        rapidjson::Value & to,
        rapidjson::MemoryPoolAllocator<> & allocator) const {...}

    // Write a value.
    void write(
        const ValueType & value,
        rapidjson::Value & to,
        rapidjson::MemoryPoolAllocator<> & allocator) const {...}
};

Please note that a references to instances of mutable_map_key_t/const_map_key_t are passed to read/write methods.

[See full example with overloading of Reader_Writer for mutable/const_map_key_t](./dev/sample/tutorial19.1/main.cpp)

License

json_dto is distributed under BSD-3-Clause license. See LICENSE file for more information.

For the license of rapidson library see LICENSE file in rapidson distributive.

For the license of rapidson_mxxru library see LICENSE file in rapidson_mxxru distributive.

For the license of CATCH library see LICENSE file in CATCH distributive.