Tuple of vectors and push back

Question

I have a tuple of vectors, I would like to push_back() each value from an initializer list into the corresponding vector in the "vectors tuple". The create() function in the code is where I would like to do this.

template<typename...Fields>
class ComponentManager
{
  using Index = int;
public:
  /**
   * Provides a handle to a component
   **/
  struct ComponentHandle {
    static constexpr Index Nil = -1;
    bool nil() { return index == Nil; }
    const Index index;
  };

  ComponentHandle lookup(Entity e) { 
    return ComponentHandle{get(m_map,e,-1)}; 
  }

  template<int i>
  auto get(ComponentHandle handle) {
    return std::get<i>(m_field)[handle.index];
  }

  ComponentHandle create(Entity e, Fields ...fields) {
    m_entity.push_back(e);
    // m_fields.push_back ... ???
  }

private:
  std::vector<Entity>                m_entity;
  std::tuple<std::vector<Fields>...> m_field;
  std::map<Entity,Index>             m_map;
};

Example:

Entity e1, e2;
ComponentManager<int,float,int> cm{};
cm.create(e1, 42, 1337.0, 99);
cm.create(e2, 84, 2674.0, 198);
// Resulting contents of cm.m_field tuple
// { 
//    vector<int>   [ 42, 84 ], 
//    vector<float> [ 1337.0, 2674.0 ]
//    vector<int>   [ 99, 198 ]
// }

Answer 1

It may not be readily obvious, but the way to unpack a tuple in C++ is to use std::apply(). With C++17, this is as easy as:

void create(Entity e, Fields ...fields) {
    m_entity.push_back(e);
    std::apply([&](auto&... vs) {
       (vs.push_back(fields), ...);
    }, m_field);
}

With C++14, I'd suggest implementing apply() yourself anyway (it's a short function), and then you need to use the expander trick instead of using fold-expressions:

void create(Entity e, Fields ...fields) {
    m_entity.push_back(e);
    not_std::apply([&](auto&... vs) {
       using swallow = int[];
       (void)swallow{0, 
           (vs.push_back(fields), 0)...
           };
    }, m_field);
}

With C++11, most of the same applies, except we can't use generic lambdas and implementing std::apply is more verbose (but not more complicated) than in the linked reference. Thankfully, we don't actually need to for anything other than making the code shorter - we know all the vector types:

void create(Entity e, Fields ...fields) {
    m_entity.push_back(e);
    not_std::apply([&](std::vector<Fields>&... vs) {
       using swallow = int[];
       (void)swallow{0, 
           (vs.push_back(fields), 0)...
           };
    }, m_field);
}

Answer 2

template <class F, class... Args>
void for_each_argument(F f, Args&&... args) {
    (void) std::initializer_list<int>{(f(std::forward<Args>(args)), 0)...};
}

ComponentHandle create(Entity e, Fields ...fields) 
{
    for_each_argument([&](auto field)
                  { 
                      using field_type = std::vector<std::decay_t<decltype(field)>>;
                      std::get<field_type>(m_field).push_back(field); 
                  },
                  fields...);
}

wandbox example

---

I didn't mention in the question that it is a requirement that the fields can be the same type, so there can be several vector's in the tuple for example.

template <typename TVec, typename TFieldTuple, std::size_t... TIdxs>
void expander(TVec& vec, TFieldTuple ft, std::index_sequence<TIdxs...>)
{
    for_each_argument([&](auto idx)
    {
        std::get<idx>(vec).push_back(std::get<idx>(ft));
    }, std::integral_constant<std::size_t, TIdxs>{}...);
}

const auto create = [](auto& vec, auto ...fields) 
{
    expander(vec, 
             std::make_tuple(fields...), 
             std::make_index_sequence<sizeof...(fields)>());
};

wandbox example

Answer 3

C++14 solution that implements apply from C++17 half way down.

template<std::size_t I>
using index_t = std::integral_constant<std::size_t, I>;
template<std::size_t I>
constexpr index_t<I> index{};

template<class=void,std::size_t...Is>
auto index_over( std::index_sequence<Is...> ) {
  return [](auto&&f)->decltype(auto) {
    return decltype(f)(f)( index<Is>... );
  };
}
template<std::size_t N>
auto index_over( index_t<N> ={} ) {
  return index_over( std::make_index_sequence<N>{} );
}


template<class F>
auto for_each_arg( F&& f ) {
  return [f = std::forward<F>(f)](auto&&...args)->decltype(auto) {
    using discard=int[];
    (void)discard{0,(void(
      f(decltype(args)(args))
    ),0)...};
  };
}

These are useful, but not needed:

template<class F, class Tuple>
decltype(auto) apply( F&& f, Tuple&& tuple ) {
  auto count = index< std::tuple_size< std::decay_t<Tuple> >{} >;
  return index_over( count )( [&](auto...Is) {
    using std::get;
    return std::forward<F>(f)( get<decltype(Is)::value>( std::forward<Tuple>(tuple) )... );
  } );
}
template<class Tuple>
auto for_each_tuple_element(Tuple&& tuple) {
  return [&](auto&& f)->decltype(auto){
    return apply(
      for_each_arg( decltype(f)(f) ),
      std::forward<Tuple>(tuple)
    );
  };
}

test code:

int main() {
    std::tuple< std::vector<int>, std::vector<char> > tup;
    for_each_tuple_element( tup )( [](auto&& v) {
      v.push_back(3);
    });
    std::cout << std::get<0>(tup).size() << "," << std::get<1>(tup).size() << "\n";
}

live example

Then we can apply it to your problem.

ComponentHandle create(Entity e, Fields ...fields) {
  m_entity.push_back(e);
  auto indexer = index_over<sizeof...(fields>();
  auto fields_tuple = std::forward_as_tuple( std::forward<Fields>(fields)... );
  indexer( for_each_arg([&](auto Is){
    std::get<Is>(m_fields).push_back(
      std::get<Is>(decltype(fields_tuple)(fields_tuple))
    );
  } );
}

index_over takes a compile-time N and returns a lambda. This lambda takes a callable, and calls it with index_t<0> through index_t<N-1>.

for_each_arg takes a callable, and returns a lambda that takes any number of arguments. It calls the callable with each of those arguments in turn.

Stitched together we take our Fields...fields, build an index_over it to get us a compile-time set of indexes into it. We then store those Fields in a tuple of r and l value references.

We write an operation on a single index Is. We then pass that to for_each_arg, and pass the return value to the index_over, and get the single index-handling lambda to be called for each index.

Some compilers don't permit a non-constexpr std::integral_constant to convert to a scalar in a constexpr context. They are wrong and/or obsolete. For those, you'll have to do decltype(Is)::value.