Reputation: 3258
I have a template class that accepts a variable number of types as arguments. The constructor accepts a pointer to an instance of the class that uses Args... - 1
as parameter type. Searching over the internet I found out that a std::tuple
is often used to handle this type of problems, but I don't understand how I could take the template arguments, create a tuple, remove the last type and then unpack the tuple again and store the result in a variable that can be retrieved later by the parent()
function.
template<typename ...Args>
class MyClass
{
public:
MyClass(MyClass<Args...> *parent) : parent_(parent) // Should be Args - 1
{
}
MyClass<Args...>* parent()
{
return parent_;
}
private:
MyClass<Args...> *parent_;
};
I found different answers here on StackOverflow about similar topics that involve tuples. This code has been posted on another question and should get a tuple with all parameters except the last one. The problem is that I don't know how to adapt it to unpack that tuple again.
template<typename, typename>
struct concat_tuple { };
template<typename... Ts, typename... Us>
struct concat_tuple<std::tuple<Ts...>, std::tuple<Us...>>
{
using type = std::tuple<Ts..., Us...>;
};
template <class T>
struct remove_last;
template <class T>
struct remove_last<std::tuple<T>>
{
using type = std::tuple<>;
};
template <class T, class... Args>
struct remove_last<std::tuple<T, Args...>>
{
using type = typename concat_tuple<std::tuple<T>, typename remove_last<std::tuple<Args...>>::type>::type;
};
Upvotes: 4
Views: 1668
Reputation: 48467
#include <type_traits>
#include <tuple>
#include <utility>
#include <cstddef>
template <template <typename...> class C, typename... Args, std::size_t... Is>
auto pop_back(std::index_sequence<Is...>) noexcept
-> C<std::tuple_element_t<Is, std::tuple<Args...>>...>&&;
template <typename... Args>
class MyClass
{
using Parent = std::remove_reference_t<
decltype(pop_back<::MyClass, Args...>(std::make_index_sequence<sizeof...(Args) - 1>{}))
>;
public:
explicit MyClass(Parent* parent) : parent_(parent)
{
}
Parent* parent()
{
return parent_;
}
private:
Parent* parent_;
};
template <>
class MyClass<> {};
int main()
{
MyClass<> a;
MyClass<int> b(&a);
MyClass<int, char> c(&b);
MyClass<int, char, float> d(&c);
}
The answer for the previous question, before the edit:
#include <tuple>
#include <utility>
#include <cstddef>
template <typename... Args>
class MyClass
{
public:
auto newInstance()
{
return newInstance(std::make_index_sequence<sizeof...(Args) - 1>{});
}
private:
template <std::size_t... Is>
MyClass<typename std::tuple_element<Is, std::tuple<Args...>>::type...> newInstance(std::index_sequence<Is...>)
{
return {};
}
};
Why the pop_back function has no body?
This is actually a trait implemented in terms of a function declaration. Alternatively, you could use a more classic solution with a structure specialization:
template <typename T, typename S>
struct pop_back;
template <template <typename...> class C, typename... Args, std::size_t... Is>
struct pop_back<C<Args...>, std::index_sequence<Is...>>
{
using type = C<std::tuple_element_t<Is, std::tuple<Args...>>...>;
};
and then use:
using Parent = typename pop_back<MyClass, std::make_index_sequence<sizeof...(Args) - 1>>::type;
That is, I used a function declaration to shorten the syntax. And it doesn't require a body, since noone is supposed to call this function in an evaluated context.
Why are you using noexcept?
Imagine you have a function:
void foo(MyClass<int, char>) noexcept {}
And elsewhere you want to check whether the call is noexcept
:
static_assert(noexcept(foo(pop_back<MyClass, int, char, float>(std::index_sequence<0, 1>{}))), "!");
Without the noexcept
specifier, the above assertion would fail, since the call to pop_back
would be considered as possibly throwing code.
Upvotes: 6