Reading JSON file with C++ and BOOST

An HTTP server sends me a JSON response (a string) like this :

{
    "folders" :
    [{
            "id" : 109,
            "parent_id" : 110,
            "path" : "\/1\/105\/110\/"
        },
        {
            "id" : 110,
            "parent_id" : 105,
            "path" : "\/1\/105\/"
        }
    ],

    "files" :
    [{
            "id" : 26,
            "parent_id" : 105,
            "name" : "picture.png",
            "hash" : "md5_hash",
            "path" : "\/1\/105\/"
        },
        {
            "id" : 25,
            "parent_id" : 110,
            "name" : "another_picture.jpg",
            "hash" : "md5_hash",
            "path" : "\/1\/105\/110\/"
        }
    ]
}

I want to compare this "tree of a remote folder" with a local folder tree (for example a string vector containing location of my local files), so I thought in converting this JSON on a map of (string, vector ( map(string, string) ) ) (I don't know if this is possible).

I'm developing a tool to synchronize files between a local and a remote folder, so I'm using boost to list a local folder, and I want to compare the local listing with the remote listing (the JSON response) to generate actions (download missing files that dont exist in the local folder, uploading files that dont exist in the remote folder).

with this function I found on another question :

void print(boost::property_tree::ptree const& pt)
{
    using boost::property_tree::ptree;
    ptree::const_iterator end = pt.end();
    for (ptree::const_iterator it = pt.begin(); it != end; ++it)
    {
        std::cout << it->first << ": " << it->second.get_value<std::string>() << std::endl;
        print(it->second);
    }
}

I succeeded in printing something like this :

folders:
:
id: 109
parent_id: 110
name: 2011_pictures
:
id: 110
parent_id: 105
name: Aminos
files:
id: 26
parent_id: 105
name: logo.png
:
id: 5
parent_id: 109
name: me.jpg

I want to know if it is possible to generate with this result a map<string, vector <map<string,string> > >, it will have 2 keys : "folders" and "files" and with those 2 keys we can access a vector of type map that contains informations for each object (file or folder). If this is feasible, it will reduce the complexity of the task (comparing two folders listing)

example : T["folder"][0]["id"] would return "109" ; T["files"][0]["name"] would return "logo.png"

UPDATE : this question is old but I want to give an advice : use RAPIDJSON whenever you want to deal with Json under C++.


Because the data structure in the other answer was deemed "very complex" and the target data structure was suggested to be:

struct Data {
    struct Folder { int id, parent_id; std::string path; };
    struct File   { int id, parent_id; std::string path, name, md5_hash; };

    using Folders = std::vector<Folder>;
    using Files   = std::vector<File>;

    Folders folders;
    Files   files;
};

I ended up writing a transformation from generic "JSON" to that data structure (see the other answer: Reading JSON file with C++ and BOOST).

However, perhaps the OP will be more pleased if we "skip the middle man" and parse the JSON specifically into the shown Data structure. This "simplifies" the grammar making it specific for this type of document only:

start    = '{' >> 
           (folders_ >> commasep) ^
           (files_ >> commasep)
         >> '}';

folders_ = prop_key(+"folders") >> '[' >> -(folder_ % ',') >> ']';
files_   = prop_key(+"files")   >> '[' >> -(file_   % ',') >> ']';

folder_  = '{' >> (
                (prop_key(+"id")        >> int_  >> commasep) ^
                (prop_key(+"parent_id") >> int_  >> commasep) ^
                (prop_key(+"path")      >> text_ >> commasep)
            ) >> '}';
file_    = '{' >> (
                (prop_key(+"id")        >> int_  >> commasep) ^
                (prop_key(+"parent_id") >> int_  >> commasep) ^
                (prop_key(+"path")      >> text_ >> commasep) ^
                (prop_key(+"name")      >> text_ >> commasep) ^
                (prop_key(+"hash")      >> text_ >> commasep)
            ) >> '}';

prop_key = lexeme ['"' >> lazy(_r1) >> '"'] >> ':';
commasep = &char_('}') | ',';

This grammar allows

  • insignificant whitespace,
  • re-ordering of properties within objects
  • and omitted object properties

Benefits:

  • early checking of property value types
  • lower compile times
  • less code indeed: 37 fewer LoC (not counting the sample JSON lines that's ~22%)

That last benefit has a flip side: if ever you want to read slightly different JSON, now you need to muck with the grammar instead of just writing a different extraction/transform. At 37 lines of code, my preference is with the other answer but I'll leave it to you to decide.

Here's the same demo program using this grammar directly:

Live On Coliru

//#define BOOST_SPIRIT_DEBUG
#include <boost/fusion/adapted.hpp>
#include <boost/spirit/include/qi.hpp>
#include <boost/spirit/include/phoenix.hpp>

namespace qi = boost::spirit::qi;

static std::string const sample = R"(
    {
        "folders" :
        [{
                "id" : 109,
                "parent_id" : 110,
                "path" : "\/1\/105\/110\/"
            },
            {
                "id" : 110,
                "parent_id" : 105,
                "path" : "\/1\/105\/"
            }
        ],

        "files" :
        [{
                "id" : 26,
                "parent_id" : 105,
                "name" : "picture.png",
                "hash" : "md5_hash",
                "path" : "\/1\/105\/"
            },
            {
                "id" : 25,
                "parent_id" : 110,
                "name" : "another_picture.jpg",
                "hash" : "md5_hash",
                "path" : "\/1\/105\/110\/"
            }
        ]
    })";

struct Data {
    struct Folder { int id, parent_id; std::string path; };
    struct File   { int id, parent_id; std::string path, name, md5_hash; };

    using Folders = std::vector<Folder>;
    using Files   = std::vector<File>;

    Folders folders;
    Files   files;
};

BOOST_FUSION_ADAPT_STRUCT(Data::Folder, (int,id)(int,parent_id)(std::string,path))
BOOST_FUSION_ADAPT_STRUCT(Data::File,   (int,id)(int,parent_id)(std::string,path)(std::string,name)(std::string,md5_hash))
BOOST_FUSION_ADAPT_STRUCT(Data,         (Data::Folders,folders)(Data::Files,files))

namespace folder_info { // adhoc JSON parser

    template <typename It, typename Skipper = qi::space_type>
    struct grammar : qi::grammar<It, Data(), Skipper>
    {
        grammar() : grammar::base_type(start) {
            using namespace qi;

            start    = '{' >> 
                       (folders_ >> commasep) ^
                       (files_ >> commasep)
                     >> '}';

            folders_ = prop_key(+"folders") >> '[' >> -(folder_ % ',') >> ']';
            files_   = prop_key(+"files")   >> '[' >> -(file_   % ',') >> ']';

            folder_  = '{' >> (
                            (prop_key(+"id")        >> int_  >> commasep) ^
                            (prop_key(+"parent_id") >> int_  >> commasep) ^
                            (prop_key(+"path")      >> text_ >> commasep)
                        ) >> '}';
            file_    = '{' >> (
                            (prop_key(+"id")        >> int_  >> commasep) ^
                            (prop_key(+"parent_id") >> int_  >> commasep) ^
                            (prop_key(+"path")      >> text_ >> commasep) ^
                            (prop_key(+"name")      >> text_ >> commasep) ^
                            (prop_key(+"hash")      >> text_ >> commasep)
                        ) >> '}';

            prop_key = lexeme ['"' >> lazy(_r1) >> '"'] >> ':';
            commasep = &char_('}') | ',';

            ////////////////////////////////////////
            // Bonus: properly decoding the string:
            text_   = '"' >> *ch_ >> '"';

            ch_ = +(
                    ~char_("\"\\")) [ _val += _1 ] |
                       qi::lit("\x5C") >> (               // \ (reverse solidus)
                       qi::lit("\x22") [ _val += '"'  ] | // "    quotation mark  U+0022
                       qi::lit("\x5C") [ _val += '\\' ] | // \    reverse solidus U+005C
                       qi::lit("\x2F") [ _val += '/'  ] | // /    solidus         U+002F
                       qi::lit("\x62") [ _val += '\b' ] | // b    backspace       U+0008
                       qi::lit("\x66") [ _val += '\f' ] | // f    form feed       U+000C
                       qi::lit("\x6E") [ _val += '\n' ] | // n    line feed       U+000A
                       qi::lit("\x72") [ _val += '\r' ] | // r    carriage return U+000D
                       qi::lit("\x74") [ _val += '\t' ] | // t    tab             U+0009
                       qi::lit("\x75")                    // uXXXX                U+XXXX
                            >> _4HEXDIG [ append_utf8(qi::_val, qi::_1) ]
                    );

            BOOST_SPIRIT_DEBUG_NODES((files_)(folders_)(file_)(folder_)(start)(text_))
        }
    private:
        qi::rule<It, Data(),            Skipper> start;
        qi::rule<It, Data::Files(),     Skipper> files_;
        qi::rule<It, Data::Folders(),   Skipper> folders_;
        qi::rule<It, Data::File(),      Skipper> file_;
        qi::rule<It, Data::Folder(),    Skipper> folder_;
        qi::rule<It, void(const char*), Skipper> prop_key;

        qi::rule<It, std::string()> text_, ch_;
        qi::rule<It> commasep;

        struct append_utf8_f {
            template <typename...> struct result { typedef void type; };
            template <typename String, typename Codepoint>
            void operator()(String& to, Codepoint codepoint) const {
                auto out = std::back_inserter(to);
                boost::utf8_output_iterator<decltype(out)> convert(out);
                *convert++ = codepoint;
            }
        };
        boost::phoenix::function<append_utf8_f> append_utf8;
        qi::uint_parser<uint32_t, 16, 4, 4> _4HEXDIG;
    };

    template <typename Range, typename It = typename boost::range_iterator<Range const>::type>
    Data parse(Range const& input) {
        grammar<It> g;

        It first(boost::begin(input)), last(boost::end(input));
        Data parsed;
        bool ok = qi::phrase_parse(first, last, g, qi::space, parsed);

        if (ok && (first == last))
            return parsed;

        throw std::runtime_error("Remaining unparsed: '" + std::string(first, last) + "'");
    }
}

int main()
{
    auto parsed = folder_info::parse(sample);

    for (auto& e : parsed.folders) 
        std::cout << "folder:\t" << e.id << "\t" << e.path << "\n";
    for (auto& e : parsed.files) 
        std::cout << "file:\t"   << e.id << "\t" << e.path << "\t" << e.name << "\n";
}

Output:

folder: 109 /1/105/110/
folder: 110 /1/105/
file:   26  /1/105/ picture.png
file:   25  /1/105/110/ another_picture.jpg

Disclaimer: The sample below is not a full blown JSON parser. Consider using a library that supports your needs. You can see a more evolved JSON parser here https://github.com/sehe/spirit-v2-json

A quick-and-dirty Spirit grammar (assuming you don't need too much conformance) would be:

    text_   = '"' >> raw [*('\\' >> char_ | ~char_('"'))] >> '"'; // ¹
    value_  = null | bool | text_ | double_ | object_ | array_; // ²
    member_ = text_ >> ':' >> value_;
    object_ = '{' >> -(member_ % ',') >> '}';
    array_  = '[' >> -(value_  % ',') >> ']';

    // ¹ as a bonus I added utf8 escape decoding in the full sample
    // ² as another bonus I threw in the missing `null` and `bool` types

Which translates into C++ types without further effort using an AST like:

using text   = std::string;
using value  = boost::make_recursive_variant<
        null,
        bool,
        text,                                      // "string" (roughly!)
        double,                                    // number
        std::map<text, boost::recursive_variant_>, // object
        std::vector<boost::recursive_variant_>     // array
    >::type;
using member = std::pair<text, value>;
using object = std::map<text, value>;
using array  = std::vector<value>;

If you have two qd_json::value objects, you can just compare them:

qd_json::value local_tree, remote_tree;
if (local_tree == remote_tree)
{
    std::cout << "the tree is unchanged\n";
}

Here's a demo program:

Updated Demo

The demonstration was updated to show you how to get to the "user-friendly" data structure you suggested in the edit of the question:

int main() {
    auto json = qd_json::parse(sample);

    // extract into user friendly datastructure from the question:
    auto extracted = Data::extract_from(json);

    for (auto& e : extracted.folders) std::cout << "folder:\t" << e.id << "\t" << e.path << "\n";
    for (auto& e : extracted.files)   std::cout << "file:\t"   << e.id << "\t" << e.path << "\t" << e.name << "\n";
}

Live On Coliru

#include <boost/fusion/adapted/std_pair.hpp>
#include <boost/spirit/include/qi.hpp>
#include <boost/spirit/include/phoenix.hpp>
#include <map>

namespace qi = boost::spirit::qi;

static std::string const sample = R"(
    {
        "folders" :
        [{
                "id" : 109,
                "parent_id" : 110,
                "path" : "\/1\/105\/110\/"
            },
            {
                "id" : 110,
                "parent_id" : 105,
                "path" : "\/1\/105\/"
            }
        ],

        "files" :
        [{
                "id" : 26,
                "parent_id" : 105,
                "name" : "picture.png",
                "hash" : "md5_hash",
                "path" : "\/1\/105\/"
            },
            {
                "id" : 25,
                "parent_id" : 110,
                "name" : "another_picture.jpg",
                "hash" : "md5_hash",
                "path" : "\/1\/105\/110\/"
            }
        ]
    })";

namespace qd_json { // quick and dirty JSON handling
    struct null {
        bool operator==(null) const { return true; }
    };

    inline static std::ostream& operator<<(std::ostream& os, null) { return os << "null"; }

    using text   = std::string;
    using value  = boost::make_recursive_variant<
            null,
            text,                                      // "string" (roughly!)
            double,                                    // number
            std::map<text, boost::recursive_variant_>, // object
            std::vector<boost::recursive_variant_>,    // array
            bool
        >::type;
    using member = std::pair<text, value>;
    using object = std::map<text, value>;
    using array  = std::vector<value>;

    template <typename It, typename Skipper = qi::space_type>
    struct grammar : qi::grammar<It, value(), Skipper>
    {
        grammar() : grammar::base_type(value_) {
            using namespace qi;

            text_   = '"' >> raw [*('\\' >> char_ | ~char_('"'))] >> '"';
            null_   = "null" >> attr(null{});
            bool_   = "true" >> attr(true) | "false" >> attr(false);
            value_  = null_ | bool_ | text_ | double_ | object_ | array_;
            member_ = text_ >> ':' >> value_;
            object_ = '{' >> -(member_ % ',') >> '}';
            array_  = '[' >> -(value_  % ',') >> ']';

            ////////////////////////////////////////
            // Bonus: properly decoding the string:
            text_   = lexeme [ '"' >> *ch_ >> '"' ];

            ch_ = +(
                    ~char_("\"\\")) [ _val += _1 ] |
                       qi::lit("\x5C") >> (               // \ (reverse solidus)
                       qi::lit("\x22") [ _val += '"'  ] | // "    quotation mark  U+0022
                       qi::lit("\x5C") [ _val += '\\' ] | // \    reverse solidus U+005C
                       qi::lit("\x2F") [ _val += '/'  ] | // /    solidus         U+002F
                       qi::lit("\x62") [ _val += '\b' ] | // b    backspace       U+0008
                       qi::lit("\x66") [ _val += '\f' ] | // f    form feed       U+000C
                       qi::lit("\x6E") [ _val += '\n' ] | // n    line feed       U+000A
                       qi::lit("\x72") [ _val += '\r' ] | // r    carriage return U+000D
                       qi::lit("\x74") [ _val += '\t' ] | // t    tab             U+0009
                       qi::lit("\x75")                    // uXXXX                U+XXXX
                            >> _4HEXDIG [ append_utf8(qi::_val, qi::_1) ]
                    );

            BOOST_SPIRIT_DEBUG_NODES((text_)(value_)(member_)(object_)(array_)(null_)(bool_))
        }
    private:
        qi::rule<It, text()>            text_, ch_;
        qi::rule<It, null()>            null_;
        qi::rule<It, bool()>            bool_;
        qi::rule<It, value(),  Skipper> value_;
        qi::rule<It, member(), Skipper> member_;
        qi::rule<It, object(), Skipper> object_;
        qi::rule<It, array(),  Skipper> array_;

        struct append_utf8_f {
            template <typename...> struct result { typedef void type; };
            template <typename String, typename Codepoint>
            void operator()(String& to, Codepoint codepoint) const {
                auto out = std::back_inserter(to);
                boost::utf8_output_iterator<decltype(out)> convert(out);
                *convert++ = codepoint;
            }
        };
        boost::phoenix::function<append_utf8_f> append_utf8;
        qi::uint_parser<uint32_t, 16, 4, 4> _4HEXDIG;
    };

    template <typename Range, typename It = typename boost::range_iterator<Range const>::type>
    value parse(Range const& input) {
        grammar<It> g;

        It first(boost::begin(input)), last(boost::end(input));
        value parsed;
        bool ok = qi::phrase_parse(first, last, g, qi::space, parsed);

        if (ok && (first == last))
            return parsed;

        throw std::runtime_error("Remaining unparsed: '" + std::string(first, last) + "'");
    }

}

#include <boost/range/algorithm.hpp>
#include <boost/range/adaptors.hpp>
#include <boost/range/algorithm_ext/push_back.hpp>

struct Data {
    struct Folder { int id, parent_id; std::string path; };
    struct File   { int id, parent_id; std::string path, name, md5_hash; };

    using Folders = std::vector<Folder>;
    using Files   = std::vector<File>;

    Folders folders;
    Files   files;

    static Data extract_from(qd_json::value const& json) {
        using namespace boost::adaptors;

        return {
            boost::copy_range<Folders>(arr(obj(json).at("folders")) | transformed(obj) | transformed(&Data::extract_folder)),
            boost::copy_range<Files>  (arr(obj(json).at("files"))   | transformed(obj) | transformed(&Data::extract_file)),
        };
    }
 private:
    static Folder extract_folder(qd_json::object const& obj) {
        return {
            id   (obj.at("id")),
            id   (obj.at("parent_id")),
            text (obj.at("path"))
        };
    }
    static File extract_file(qd_json::object const& obj) {
        return {
            id   (obj.at("id")),
            id   (obj.at("parent_id")),
            text (obj.at("path")),
            text (obj.at("name")),
            text (obj.at("hash")),
        };
    }

    static int             id  (qd_json::value const&v) { return boost::get<double>(v); };
    static std::string     text(qd_json::value const&v) { return boost::get<qd_json::text>(v); };
    static qd_json::array  arr (qd_json::value const&v) { return boost::get<qd_json::array>(v); };
    static qd_json::object obj (qd_json::value const&v) { return boost::get<qd_json::object>(v); };
};

int main()
{
    auto json = qd_json::parse(sample);

    // compare json documents
    qd_json::value clone = json;
    assert(json == clone);

    // extract into user friendly datastructure from the question:
    auto extracted = Data::extract_from(json);

    for (auto& e : extracted.folders) std::cout << "folder:\t" << e.id << "\t" << e.path << "\n";
    for (auto& e : extracted.files)   std::cout << "file:\t"   << e.id << "\t" << e.path << "\t" << e.name << "\n";
}

Output:

folder: 109 /1/105/110/
folder: 110 /1/105/
file:   26  /1/105/ picture.png
file:   25  /1/105/110/ another_picture.jpg