C++ iterate into nested struct field with boost fusion adapt_struct
Asked Answered
E

2

9

Two stackoverflow answers suggest the approach using fusion adapt_struct to iterate over struct fields. The approach looks nice. However, how do you iterate into a field which itself is a struct?

Following the previous answers, I come up with the code below. The problem is at the "#if 0" clause the code does not compile. As an alternative solution I created "decode()" function to take a void pointer to the target argument. That works, but loses the type information at compile time. Is there a better solution?

struct Foo_s { int i; };
BOOST_FUSION_ADAPT_STRUCT( Foo_s,  (int, i) )

struct Bar_s { int v; Foo_s w; };
BOOST_FUSION_ADAPT_STRUCT( Bar_s, (int, v)  (Foo_s, w) )

struct AppendToTextBox {
    template <typename T> void operator()(T& t) const {
        int status = 0;
        const char *realname = abi::__cxa_demangle(typeid(t).name(), 0, 0, &status);
        printf("  typename: %s  value: %s  realname: %s\n", typeid(t).name(),
               boost::lexical_cast<std::string>(t).c_str(), realname);
        std::string rn(realname);
        if ( rn.rfind("_s") == rn.size()-2 ) {
#if 0 /* this can not compile */
            for_each(t, AppendToTextBox());
#else
            decode(&t, rn);
#endif
        }
    }
};

void decode(void *f, std::string & intype ) {
    if ( intype.find("Foo_s") == 0 ) 
        for_each( *(Foo_s *)f, AppendToTextBox());
};

int main(int argc, char *argv[]) {
  Bar_s f = { 2, { 3 } };
  for_each(f, AppendToTextBox());
  return 0;
}

I have seen on wikipedia instead of passing a type string "intype" you can use typeid and dynamic_cast. But that will only be a minor improvement. I'm looking for a solution that is more intrinsic to C++ or boost language design.

Eberta answered 23/8, 2012 at 4:32 Comment(0)
O
10

I made an example of what you want that you can see at my blog site. In this is case it's a JSON serializer that works with nested structs. It uses a 'more Boost' solution since I saw it in the Boost.Serialization library. (See also below and live on Coliru.)

The solution uses Fusion Sequence adaptation of structs and a metafunction that walks object members (recursively) - using Boost.TypeTraits and different traits for specific types.

You can see a more complex example of the same solution at the site for googlecode corbasim project for creating an run-time reflexive API.

Code listing for the generic JSON serializer:

See it Live on Coliru

#ifndef JSON_SERIALIZER_HPP
#define JSON_SERIALIZER_HPP

#include <boost/type_traits.hpp> // is_array, is_class, remove_bounds

#include <boost/mpl/eval_if.hpp>
#include <boost/mpl/identity.hpp>
#include <boost/mpl/next_prior.hpp>

#include <boost/fusion/mpl.hpp>
#include <boost/fusion/adapted.hpp> // BOOST_FUSION_ADAPT_STRUCT

// boost::fusion::result_of::value_at
#include <boost/fusion/sequence/intrinsic/value_at.hpp>
#include <boost/fusion/include/value_at.hpp>

// boost::fusion::result_of::size
#include <boost/fusion/sequence/intrinsic/size.hpp>
#include <boost/fusion/include/size.hpp>

// boost::fusion::at
#include <boost/fusion/sequence/intrinsic/at.hpp>
#include <boost/fusion/include/at.hpp>

namespace json
{

// Forward
template < typename T >
struct serializer;

namespace detail
{

namespace iterator
{

template < typename S, typename N >
struct Comma
{
    template < typename Ostream >
    static inline void comma(Ostream& os)
    {
        os << ", ";
    }
};

template < typename S >
struct Comma< S, typename boost::mpl::prior< typename boost::fusion::result_of::size< S >::type >::type >
{
    template < typename Ostream >
    static inline void comma(Ostream& os)
    {
    }
};

// Iteracion sobre una estructura
template < typename S, typename N >
struct StructImpl
{
    // Tipo del campo actual
    typedef typename boost::fusion::result_of::value_at< S, N >::type current_t;
    typedef typename boost::mpl::next< N >::type next_t;
    typedef boost::fusion::extension::struct_member_name< S, N::value > name_t;

    template < typename Ostream >
    static inline void serialize(Ostream& os, const S& s)
    {
        os << "\"" << name_t::call() << "\": ";
        ::json::serializer< current_t >::serialize(os, boost::fusion::at< N >(s));

        // Insert comma or not    
        Comma< S, N >::comma(os);

        StructImpl< S, next_t >::serialize(os, s);
    }
};

// Fin de la iteracion sobre estructuras.
template < typename S >
struct StructImpl< S, typename boost::fusion::result_of::size< S >::type >
{
    template < typename Ostream >
    static inline void serialize(Ostream& os, const S& s)
    {
        // Nada que hacer
    }
};

// Iterador sobre una estructura. Template fachada.
template < typename S >
struct Struct : StructImpl< S, boost::mpl::int_< 0 > > {};

} // iterator

template < typename T >
struct array_serializer 
{
    typedef array_serializer< T > type;

    typedef typename boost::remove_bounds< T >::type slice_t;

    static const size_t size = sizeof(T) / sizeof(slice_t);

    template < typename Ostream >
    static inline void serialize(Ostream& os, const T& t)
    {
        os << "[";
        for(size_t idx=0; idx<size; idx++)
        {
            ::json::serializer< slice_t >::serialize(os, t[idx]);
            if (idx != size-1)
                os << ", ";
        }
        os << "]";
    }

};

template < typename T >
struct struct_serializer 
{
    typedef struct_serializer< T > type;

    template < typename Ostream >
    static inline void serialize(Ostream& os, const T& t)
    {
        os << "{";
        iterator::Struct< T >::serialize(os, t);
        os << "}";
    }
};

template < typename T >
struct arithmetic_serializer 
{
    typedef arithmetic_serializer< T > type;

    template < typename Ostream >
    static inline void serialize(Ostream& os, const T& t)
    {
        os << t;
    }
};

template < typename T >
struct calculate_serializer
{
    typedef
        typename boost::mpl::eval_if< boost::is_array< T >,
            boost::mpl::identity< array_serializer < T > >,
        //else
        typename boost::mpl::eval_if< boost::is_class< T >,
            boost::mpl::identity< struct_serializer < T > >,
        //else
            boost::mpl::identity< arithmetic_serializer < T > >
        >
        >::type type;

};

} // detail

template < typename T >
struct serializer : public detail::calculate_serializer < T >::type
{
};


} // json

#endif // JSON_SERIALIZER_HPP

//#include "json.hpp"
#include <iostream>

struct my_other_struct
{
    int my_other_integer;
};

struct my_struct
{
    int my_integer;

    typedef int my_array_t[2];
    my_array_t my_array;

    typedef my_other_struct my_other_structs_t[3];
    my_other_structs_t my_other_structs;
};

BOOST_FUSION_ADAPT_STRUCT(my_struct, (int, my_integer) (my_struct::my_array_t, my_array) (my_struct::my_other_structs_t, my_other_structs))
BOOST_FUSION_ADAPT_STRUCT(my_other_struct, (int, my_other_integer))


int main(int argc, char *argv[])
{
    my_struct s1 = my_struct { 1, { 42, -42 }, { { 11 }, { 22 }, { 33 } } };

    json::serializer< my_struct >::serialize(std::cout, s1);

    std::cout << std::endl;
}
Outwear answered 23/8, 2012 at 6:27 Comment(4)
Thanks, great answers! I elaborated a bit and put the working code in a separate answer, just to make it clear.Eberta
@Eberta you could edit the answer in order to share the reputation with the answer authorTreviso
The two links Andres pointed to are very helpful. If you are looking for an answer, do follow them.Eberta
@AndrésSenac Ah, ironically I seem to have found the same code via Minghua's blog for my answer hereRikki
E
3

Andres gives an excellent answer. The problem in my original code is that "for_each" takes only sequence types. When compiler evaluates the T for an int, it passes to "for_each" an int argument thus it fails. The idea behind Adries' solution is to hide "for_each" in a sequence-specific class (DecImplSeq_s below), and provide an alternative class (DecImplVoid_s) for non-sequence fields. Then create a facade class to divide the decoding of sequence and non-sequence fields (DecCalc_s).

The common header goes with the first example below to show Adres' idea.

/* compile with g++ 4.4.6: g++ -I boost_1_35_0 test.cpp */
#include <typeinfo>
#include <string>
#include <boost/fusion/include/sequence.hpp>
#include <boost/fusion/include/algorithm.hpp>
#include <boost/fusion/include/adapt_struct.hpp>
#include <boost/fusion/include/is_sequence.hpp>
#include <boost/mpl/eval_if.hpp>
#include <boost/lexical_cast.hpp>
#include <cxxabi.h>
#include <stdio.h>
using namespace boost::fusion;

The common code of the solution derived directly from Adres' sample:

template <typename T2> struct Dec_s;
struct AppendToTextBox {
  template <typename T> void operator()(T& t) const {
        //decode T and t as the original code here...
        Dec_s<T>::decode(t);
  }
};
template <typename T2> struct DecImplSeq_s {
  typedef DecImplSeq_s<T2> type;
  static void decode(T2   & f) { for_each(f, AppendToTextBox()); };
};
template <typename T2> struct DecImplVoid_s {
  typedef DecImplVoid_s<T2> type;
  static void decode(T2   & f) { };
};

template <typename T2> struct DecCalc_s {
  typedef typename
    boost::mpl::eval_if< traits::is_sequence<T2>, DecImplSeq_s<T2>, DecImplVoid_s<T2> >
  ::type type;
};

template <typename T2> struct Dec_s : public DecCalc_s<T2>::type { };

Here is how you can use the common code above:

struct Foo_s { int i; char k[100]; };
struct Bar_s { int v; Foo_s w; };

BOOST_FUSION_ADAPT_STRUCT( Foo_s,  (int, i)  (char, k[100]) )
BOOST_FUSION_ADAPT_STRUCT( Bar_s, (int, v)  (Foo_s, w) )

int main(int argc, char *argv[]) {
  Bar_s f = { 2, { 3, "abcd" } };
  Dec_s<Bar_s>::decode(f);
  return 0;
}

Another solution that is more straightforward without using advanced boost tricks, you can implement a specialized decoder class for each primitive types, without using "eval_if". To use this solution, you need to do a specialization for each primitive type in your structs.

struct Foo_s { int i; char k[100]; };
BOOST_FUSION_ADAPT_STRUCT( Foo_s,  (int, i)  (char, k[100]) )

struct Bar_s { int v; Foo_s w; };
BOOST_FUSION_ADAPT_STRUCT( Bar_s, (int, v)  (Foo_s, w) )

template <typename T2> struct Dec_s {  static void decode(T2   & f); };
struct AppendToTextBox {
    template <typename T>
    void operator()(T& t) const {
        //decode T and t as the original code here...
        Dec_s<T>::decode(t);
    }
};

template <typename T2> void Dec_s<T2>::decode(T2 & f) {
    for_each(f, AppendToTextBox());
};
template<> void Dec_s<int >::decode(int  & f) {};
template<> void Dec_s<char>::decode(char & f) {};

int main(int argc, char *argv[]) {
  Bar_s f = { 2, { 3, "abcd" } };
  Dec_s<Bar_s>::decode(f);
  return 0;
}

After some progressive exploration, here is a complete example. It uses more recent boost features, but does not build with early boost versions like 1.35.0. It works well with boost 1.47.0 and 1.51.0.

The common header part:

#include <typeinfo>
#include <string>
#include <boost/fusion/include/sequence.hpp>
#include <boost/fusion/include/algorithm.hpp>
#include <boost/fusion/include/adapt_struct.hpp>
#include <boost/fusion/include/is_sequence.hpp>
#include <boost/mpl/eval_if.hpp>
#include <boost/type_traits.hpp> // is_array, is_class, remove_bounds
#include <boost/lexical_cast.hpp>
#include <cxxabi.h>
#include <stdio.h>

extern int dec_indents; /* 0, 4, 8, ... */
struct NL {
    static void print() { printf("\n");
        for (int i=0; i<dec_indents; i++) printf(" ");
    }
};

using namespace boost::fusion;

Then the common decoder with output formatting:

template <typename T2> struct Dec_s;

template <typename S, typename N> struct Comma {
  static inline void comma() { printf(" , "); }
};
template <typename S> struct Comma<S, typename
 boost::mpl::prior<typename boost::fusion::result_of::size<S>::type >::type> {
   static inline void comma() {}
};

template <typename S, typename N> struct DecImplSeqItr_s {
  typedef typename boost::fusion::result_of::value_at<S, N>::type current_t;
  typedef typename boost::mpl::next<N>::type next_t;
  typedef boost::fusion::extension::struct_member_name<S, N::value> name_t;
  static inline void decode(S& s) {
    printf(" \"%s\" = ", name_t::call() );
    Dec_s<current_t>::decode(boost::fusion::at<N>(s));
    Comma<S, N>::comma();  // Insert comma or not
    DecImplSeqItr_s<S, next_t>::decode(s);
  }
};
template <typename S>
struct DecImplSeqItr_s<S, typename boost::fusion::result_of::size<S>::type > {
    static inline void decode(S& s) { }
};
template <typename S>
struct DecImplSeqStart_s:DecImplSeqItr_s<S, boost::mpl::int_<0> > {};

template <typename S> struct DecImplSeq_s {
  typedef DecImplSeq_s<S> type;
  static void decode(S & s) {
    printf("  struct  start --- { --- ");
    dec_indents += 4;
    NL::print();
    DecImplSeqStart_s<S>::decode(s);
    dec_indents -= 4;
    NL::print();
    printf("  struct  done  --- } --- ");
    NL::print();
  };
};

template <typename T2> struct DecImplArray_s {
  typedef DecImplArray_s<T2> type;
  typedef typename boost::remove_bounds<T2>::type slice_t;
  static const size_t size = sizeof(T2) / sizeof(slice_t);
  static inline void decode(T2 & t) {
    printf("  array start --- [ --- ");
    dec_indents += 4;
    NL::print();
    for(size_t idx=0; idx<size; idx++) {
        Dec_s<slice_t>::decode(t[idx]);
        if (idx < size-1) {
            NL::print(); printf(" , ");
        }
    }
    dec_indents -= 4;
    NL::print();
    printf("  array done  --- ] --- \n");
    NL::print();
  }
};

template <typename T2> struct DecImplVoid_s {
  typedef DecImplVoid_s<T2> type;
  static void decode(T2   & t) {
    int status = 0;
    const char *realname = abi::__cxa_demangle(typeid(t).name(),0,0,&status);
    printf(" type %s", realname);
    NL::print();
  };
};

template <typename T2> struct DecCalc_s {
  typedef
    typename boost::mpl::eval_if< traits::is_sequence<T2>, DecImplSeq_s<T2>,
    typename boost::mpl::eval_if< boost::is_array<T2>,
                                 boost::mpl::identity< DecImplArray_s<T2> >,
    DecImplVoid_s<T2>   > >
  ::type type;
};

template <typename T2> struct Dec_s : public DecCalc_s<T2>::type { };

To use this common decoder, you can put it into a .h file, and use the following .c code:

/* compile with g++ 4.5.1: g++ -I boost_1_47_0 test.cpp */

#include "common_decoder.h"

using namespace boost::fusion;

int dec_indents=0;

struct Foo_s { int i; typedef char j_t[10]; Foo_s::j_t j; };
BOOST_FUSION_ADAPT_STRUCT( Foo_s, (int, i) (Foo_s::j_t, j) )

struct Bar_s { int v; typedef Foo_s w_t[2]; Bar_s::w_t w; };
BOOST_FUSION_ADAPT_STRUCT( Bar_s, (int, v) (Bar_s::w_t, w) )

int main(int argc, char *argv[]) {
  Bar_s f = { 2, {{ 3, "abcd" },{ 4, "defg" }} };
  Dec_s<Bar_s>::decode(f);
  return 0;
}
Eberta answered 24/8, 2012 at 1:27 Comment(0)

© 2022 - 2024 — McMap. All rights reserved.