Please consider the following code snippet:
template<class E>
class vector_expression
{
public:
auto size() const {
return static_cast<E const&>(*this).size();
}
auto operator[](/* type equal to E::size_type */ i) const
{
if (i >= size())
throw std::length_error("");
return static_cast<E const&>(*this)[i];
}
}; // class vector_expression
template<typename T, class Tuple = std::vector<T>>
class vector
: public vector_expression<vector<T, Tuple>>
{
public:
using value_type = T;
using size_type = typename Tuple::size_type;
size_type size() const {
return m_elements.size();
}
value_type operator[](size_type i) const { /* ... */ }
private:
Tuple m_elements;
}; // class vector
The type of the argument i
of vector_expression<E>
should equal E::size_type
. For a plausible reason, typename E::size_type
doesn't work here. For the same reason, std::result_of_t<decltype(&size)(vector_expression)>
doesn't work here.
So, how can we do it, if we can do it?
You can pass it explicitly as a template parameter to vector_expression
:
template<class E, class size_type>
class vector_expression ...
template<typename T, class Tuple = std::vector<T>>
class vector
: public vector_expression<vector<T, Tuple>,
typename Tuple::size_type> ...
Edit:
It is also possible to turn the problematic function into a member function template, so that it is not instantiated until the full class definition is seen:
template <typename K = E>
auto operator[](typename K::size_type i) const
{
if (i >= size())
throw std::length_error("");
return static_cast<K const&>(*this)[i];
}