vn-verdnaturachat/ios/Pods/boost-for-react-native/boost/fusion/tuple/detail/tuple.hpp

123 lines
3.9 KiB
C++

/*=============================================================================
Copyright (c) 2001-2011 Joel de Guzman
Distributed under the Boost Software License, Version 1.0. (See accompanying
file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
==============================================================================*/
#if !defined(FUSION_TUPLE_10032005_0810)
#define FUSION_TUPLE_10032005_0810
#include <boost/fusion/support/config.hpp>
#include <boost/fusion/tuple/detail/tuple_fwd.hpp>
#include <boost/fusion/container/vector/vector.hpp>
#include <boost/fusion/sequence/intrinsic/size.hpp>
#include <boost/fusion/sequence/intrinsic/value_at.hpp>
#include <boost/fusion/sequence/intrinsic/at.hpp>
#include <boost/fusion/sequence/comparison.hpp>
#include <boost/fusion/sequence/io.hpp>
#include <boost/utility/enable_if.hpp>
#include <boost/type_traits/is_const.hpp>
#include <boost/config/no_tr1/utility.hpp>
#if !defined(BOOST_FUSION_DONT_USE_PREPROCESSED_FILES)
#include <boost/fusion/tuple/detail/preprocessed/tuple.hpp>
#else
#if defined(__WAVE__) && defined(BOOST_FUSION_CREATE_PREPROCESSED_FILES)
#pragma wave option(preserve: 2, line: 0, output: "preprocessed/tuple" FUSION_MAX_VECTOR_SIZE_STR ".hpp")
#endif
/*=============================================================================
Copyright (c) 2001-2011 Joel de Guzman
Distributed under the Boost Software License, Version 1.0. (See accompanying
file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
This is an auto-generated file. Do not edit!
==============================================================================*/
#if defined(__WAVE__) && defined(BOOST_FUSION_CREATE_PREPROCESSED_FILES)
#pragma wave option(preserve: 1)
#endif
namespace boost { namespace fusion
{
template <BOOST_PP_ENUM_PARAMS(FUSION_MAX_VECTOR_SIZE, typename T)>
struct tuple : vector<BOOST_PP_ENUM_PARAMS(FUSION_MAX_VECTOR_SIZE, T)>
{
typedef vector<
BOOST_PP_ENUM_PARAMS(FUSION_MAX_VECTOR_SIZE, T)>
base_type;
BOOST_FUSION_GPU_ENABLED tuple()
: base_type() {}
BOOST_FUSION_GPU_ENABLED tuple(tuple const& rhs)
: base_type(static_cast<base_type const&>(rhs)) {}
template <typename U1, typename U2>
BOOST_FUSION_GPU_ENABLED
tuple(std::pair<U1, U2> const& rhs)
: base_type(rhs) {}
#include <boost/fusion/tuple/detail/tuple_expand.hpp>
template <typename T>
BOOST_FUSION_GPU_ENABLED
tuple& operator=(T const& rhs)
{
base_type::operator=(rhs);
return *this;
}
BOOST_FUSION_GPU_ENABLED
tuple& operator=(tuple const& rhs)
{
base_type::operator=(static_cast<base_type const&>(rhs));
return *this;
}
template <typename U1, typename U2>
BOOST_FUSION_GPU_ENABLED
tuple& operator=(std::pair<U1, U2> const& rhs)
{
base_type::operator=(rhs);
return *this;
}
};
template <typename Tuple>
struct tuple_size : result_of::size<Tuple> {};
template <int N, typename Tuple>
struct tuple_element : result_of::value_at_c<Tuple, N> {};
template <int N, typename Tuple>
BOOST_FUSION_GPU_ENABLED
inline typename
lazy_disable_if<
is_const<Tuple>
, result_of::at_c<Tuple, N>
>::type
get(Tuple& tup)
{
return at_c<N>(tup);
}
template <int N, typename Tuple>
BOOST_FUSION_GPU_ENABLED
inline typename result_of::at_c<Tuple const, N>::type
get(Tuple const& tup)
{
return at_c<N>(tup);
}
}}
#if defined(__WAVE__) && defined(BOOST_FUSION_CREATE_PREPROCESSED_FILES)
#pragma wave option(output: null)
#endif
#endif // BOOST_FUSION_DONT_USE_PREPROCESSED_FILES
#endif