MessagePack C++ - How to iterate through an unknown data structure?

Question

I want to share structured data between C++ and Python languages using MessagePack like this one:

{
  "t" : [ [t00,...,t0N], ... , [tM0,...,tMN] ],
  "x" : [ x0,..,xN],
  "P" : [ [P00, ..., P0N], ..., [PM0,...,PMN] ]
}

The number of variables is optional so in some cases I will have for example only:

{
 "t" : [ [t00,...,t0N], ... , [tM0,...,tMN] ]
}

Decoding this in Python is pretty simple, my problem is to figure out how to decode this in C++ if I don't know in advance the structure of the data ? or the exact number of variables that I would have; is it possible to iterate the structure in these cases?

I managed to handle a "fixed" data structure ( always with the same number of variables ) defining a struct for example:

struct variables
{
   std::vector< std::vector<double> > t;
   std::vector< double > x;
   std::vector< std::vector<double> > P;
   MSPACK_DEFINE_MAP( t, x, P );
};

std::stringstream inBuffer;

.... (read data )

std::string str( inBuffer.str() );
msgpack::object_handle oh = msgpack::unpack( str.data(), str.size() );
msgpack::object deserialized = oh.get();

variables var;
deserialized.convert( var );

Is there a better way to accomplish this ?, how could manage optional variables that could not appear in the structure ?; I repeat the previous question: could I iterate an unknown data structure in C++?, how ?

Thanks in advance!

Regards, Ernesto

Answer 1

There are two ways to treat unknown data structure.

The first way is using parse/visitor mechanism. Here is an example:

#include <msgpack.hpp>
#include <sstream>
#include <iostream>

// This is a simple print example visitor.
// You can do any processing in your visitor.
struct my_visitor : msgpack::null_visitor {
    bool start_map_key() {
        processing_map_key = true;
        return true;
    }
    bool end_map_key() {
        processing_map_key = false;
        return true;
    }
    bool start_array(uint32_t size) {
        std::cout << "array (size:" << size << ")[" << std::endl;
        return true;
    }
    bool end_array() {
        std::cout << "]" << std::endl;
        return true;
    }

    bool visit_str(const char* v, uint32_t size) {
        if (processing_map_key) {
            std::cout << "map key:" << std::string(v, size) << std::endl;
        }
        return true;
    }
    bool visit_positive_integer(uint64_t v) {
        std::cout << "found value:" << v << std::endl;
        return true;
    }

    bool processing_map_key = false;
    std::string indent;
};


int main() {
    // create test data
    std::stringstream ss;
    msgpack::packer<std::stringstream> pk(ss);
    pk.pack_map(1);
    pk.pack("t");
    pk.pack_array(2);
    pk.pack_array(3);
    pk.pack(1);
    pk.pack(2);
    pk.pack(3);
    pk.pack_array(3);
    pk.pack(4);
    pk.pack(5);
    pk.pack(6);

    // print data (for debug)
    {
        auto oh = msgpack::unpack(ss.str().data(), ss.str().size());
        std::cout << oh.get() << std::endl;
    }

    // apply visitor
    {
        my_visitor mv;
        msgpack::parse(ss.str().data(), ss.str().size(), mv);
    }
}

Running demo: https://wandbox.org/permlink/3NrR4IMDIuLTk9e9

See https://github.com/msgpack/msgpack-c/wiki/v2_0_cpp_visitor.

The other way is using msgpack::type::variant or `msgpack::type::variant_ref. The former copies data, you can update it. The latter doesn't copy data. You cannot update it. This approach requires boost. So you need to define MSGPACK_USE_BOOST. I recommend defining as a compiler option.

// Boost is required
#define MSGPACK_USE_BOOST

#include <msgpack.hpp>
#include <sstream>
#include <iostream>

struct my_visitor:boost::static_visitor<void> {
    void operator()(uint64_t v) const {
        std::cout << "positive insteger:" << v << std::endl;
    }
    // const is required for map key because std::multimap's key (first) is const.
    void operator()(std::string const& v) const {
        std::cout << "string:" << v << std::endl;
    }
    void operator()(std::vector<msgpack::type::variant>& v) const {
        std::cout << "array found" << std::endl;
        for (auto& e : v) {
            boost::apply_visitor(*this, e);
        }
    }
    void operator()(std::multimap<msgpack::type::variant, msgpack::type::variant>& v) const {
        std::cout << "map found" << std::endl;
        for (auto& e : v) {
            std::cout << "key:" << std::endl;
            boost::apply_visitor(*this, e.first);
            std::cout << "value:" << std::endl;
            boost::apply_visitor(*this, e.second);
        }
    }
    template <typename T>
    void operator()(T const&) const {
        std::cout << "  match others" << std::endl;
    }
};

int main() {
    // create test data
    std::stringstream ss;
    msgpack::packer<std::stringstream> pk(ss);
    pk.pack_map(1);
    pk.pack("t");
    pk.pack_array(2);
    pk.pack_array(3);
    pk.pack(1);
    pk.pack(2);
    pk.pack(3);
    pk.pack_array(3);
    pk.pack(4);
    pk.pack(5);
    pk.pack(6);

    auto oh = msgpack::unpack(ss.str().data(), ss.str().size());
    std::cout << oh.get() << std::endl;

    msgpack::type::variant v = oh.get().as<msgpack::type::variant>();
    boost::apply_visitor(my_visitor(), v);
}

Running demo: https://wandbox.org/permlink/HQwJjfwW8rLEMi0d

See https://github.com/msgpack/msgpack-c/wiki/v2_0_cpp_variant

Here are exampless: https://github.com/msgpack/msgpack-c/blob/master/example/boost/msgpack_variant_capitalize.cpp https://github.com/msgpack/msgpack-c/blob/master/example/boost/msgpack_variant_mapbased.cpp

Both ways can treat unpredictable data structure. You need to do some visitor processing. If the data structure is predictable some extent, your original approach is also good way.

Answer 2

Actually there is a simpler way, if you are dealing with maps (like stated in the question), not arrays.

msgpack::object_handle oh = msgpack::unpack(/* some data */);
std::map<std::string,msgpack::type::variant> map = obj.convert();

This way you will get a map with all the data, no need for a visitor or boost.