If all of the members of std::tuple
are of standard layout types, is that std::tuple
itself standard layout? The presence of a user-defined copy-constructor makes it non-trivial, but I was wondering if it can still be standard layout.
A quote from the spec would be good.
Class template std::tuple is a fixed-size collection of heterogeneous values. It is a generalization of std::pair.
A tuple is an object capable to hold a collection of elements where each element can be of a different type. The class template needs the header <tuple> . std::tuple is a generalization of std::pair. You can convert between tuples with two elements and pairs.
Initializes each element of the tuple with the corresponding element of other . This constructor is constexpr if every operation it performs is constexpr. For the empty tuple std::tuple<>, it is constexpr.
The Tuple is an object capable to hold a collection of elements, where each element can be of different types. The pair can make a set of two values, which may be of different types. The pair is basically a special type of tuple, where only two values are allowed.
No, standard layout requires that all nonstatic data members belong to either one base subobject or directly to the most derived type, and typical implementations of std::tuple
implement one member per base class.
Because a member-declaration cannot be a pack expansion, in light of the above requirement, a standard layout tuple
cannot have more than one member. An implementation could still sidestep the issue by storing all the tuple
"members" inside one char[]
, and obtaining the object references by reinterpret_cast
. A metaprogram would have to generate the class layout. Special member functions would have to be reimplemented. It would be quite a pain.
Inspired by PotatoSwatter's answer, I've dedicated my day to creating a standard layout tuple for C++14.
The code actually works, but is not currently suited for use as it involves undefined behaviour. Treat it as a proof-of-concept. Here's the code I ended up with:
#include <iostream>
#include <type_traits>
#include <array>
#include <utility>
#include <tuple>
//get_size
template <typename T_head>
constexpr size_t get_size()
{
return sizeof(T_head);
}
template <typename T_head, typename T_second, typename... T_tail>
constexpr size_t get_size()
{
return get_size<T_head>() + get_size<T_second, T_tail...>();
}
//concat
template<size_t N1, size_t... I1, size_t N2, size_t... I2>
constexpr std::array<size_t, N1+N2> concat(const std::array<size_t, N1>& a1, const std::array<size_t, N2>& a2, std::index_sequence<I1...>, std::index_sequence<I2...>)
{
return { a1[I1]..., a2[I2]... };
}
template<size_t N1, size_t N2>
constexpr std::array<size_t, N1+N2> concat(const std::array<size_t, N1>& a1, const std::array<size_t, N2>& a2)
{
return concat(a1, a2, std::make_index_sequence<N1>{}, std::make_index_sequence<N2>{});
}
//make_index_array
template<size_t T_offset, typename T_head>
constexpr std::array<size_t, 1> make_index_array()
{
return {T_offset};
}
template<size_t T_offset, typename T_head, typename T_Second, typename... T_tail>
constexpr std::array<size_t, (sizeof...(T_tail) + 2)> make_index_array()
{
return concat(
make_index_array<T_offset, T_head>(),
make_index_array<T_offset + sizeof(T_head),T_Second, T_tail...>()
);
}
template<typename... T_args>
constexpr std::array<size_t, (sizeof...(T_args))> make_index_array()
{
return make_index_array<0, T_args...>();
}
template<int N, typename... Ts>
using T_param = typename std::tuple_element<N, std::tuple<Ts...>>::type;
template <typename... T_args>
struct standard_layout_tuple
{
static constexpr std::array<size_t, sizeof...(T_args)> index_array = make_index_array<T_args...>();
char storage[get_size<T_args...>()];
//Initialization
template<size_t T_index, typename T_val>
void initialize(T_val&& val)
{
void* place = &this->storage[index_array[T_index]];
new(place) T_val(std::forward<T_val>(val));
}
template<size_t T_index, typename T_val, typename T_val2, typename... T_vals_rest>
void initialize(T_val&& val, T_val2&& val2, T_vals_rest&&... vals_rest)
{
initialize<T_index, T_val>(std::forward<T_val>(val));
initialize<T_index+1, T_val2, T_vals_rest...>(std::forward<T_val2>(val2), std::forward<T_vals_rest>(vals_rest)...);
}
void initialize(T_args&&... args)
{
initialize<0, T_args...>(std::forward<T_args>(args)...);
}
standard_layout_tuple(T_args&&... args)
{
initialize(std::forward<T_args>(args)...);
}
//Destruction
template<size_t T_index, typename T_val>
void destroy()
{
T_val* place = reinterpret_cast<T_val*>(&this->storage[index_array[T_index]]);
place->~T_val();
}
template<size_t T_index, typename T_val, typename T_val2, typename... T_vals_rest>
void destroy()
{
destroy<T_index, T_val>();
destroy<T_index+1, T_val2, T_vals_rest...>();
}
void destroy()
{
destroy<0, T_args...>();
}
~standard_layout_tuple()
{
destroy();
}
template<size_t T_index>
void set(T_param<T_index, T_args...>&& data)
{
T_param<T_index, T_args...>* ptr = reinterpret_cast<T_param<T_index, T_args...>*>(&this->storage[index_array[T_index]]);
*ptr = std::forward<T_param<T_index, T_args...>>(data);
}
template<size_t T_index>
T_param<T_index, T_args...>& get()
{
return *reinterpret_cast<T_param<T_index, T_args...>*>(&this->storage[index_array[T_index]]);
}
};
int main() {
standard_layout_tuple<float, double, int, double> sltuple{5.5f, 3.4, 7, 1.22};
sltuple.set<2>(47);
std::cout << sltuple.get<0>() << std::endl;
std::cout << sltuple.get<1>() << std::endl;
std::cout << sltuple.get<2>() << std::endl;
std::cout << sltuple.get<3>() << std::endl;
std::cout << "is standard layout:" << std::endl;
std::cout << std::boolalpha << std::is_standard_layout<standard_layout_tuple<float, double, int, double>>::value << std::endl;
return 0;
}
Live example: https://ideone.com/4LEnSS
There's a few things I'm not happy with:
This is not yet suitable for use as-is, really only treat it as a proof-of-concept in this state. I will probably come back to improve on some of these issues. Or, if anyone else can improve it, feel free to edit.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With