Selecting a functions implementation based on iterator::value_type

Question

I'm looking for a reasonable way to select a sort algorithm based on the value type of the container.

In its current form I can deduce the proper sort(a, b) for integer/non-integer data.

#include <cstdlib>
#include <type_traits>
#include <algorithm>
#include <vector>
#include <iostream>

namespace sort_selector{
    template<typename T>
    void _radix_sort(T begin, T end){
        // radix implementation
    }

    template<typename T>
    typename std::enable_if<
                  std::is_integral<typename T::value_type>::value>::type
    sort(T begin, T end){
        std::cout << "Doing radix" << std::endl;
        sort_selector::_radix_sort(begin, end);
    }

    template<typename T>
    typename std::enable_if<
                  !std::is_integral<typename T::value_type>::value>::type
    sort(T begin, T end){
        std::cout << "Doing sort" << std::endl;
        std::sort(begin, end);
    }
}
int main(int argc, char** argv) {
    std::vector<double> for_stdsort = {1, 4, 6, 2};
    std::vector<int32_t> for_radixsort = {1, 4, 6, 2};
    //std::array<int32_t, 4> array_for_radixsort = {1, 4, 6, 2};

    sort_selector::sort(std::begin(for_stdsort), std::end(for_stdsort));
    sort_selector::sort(std::begin(for_radixsort), std::end(for_radixsort));
    //sort_selector::sort(std::begin(array_for_radixsort), 
    //                     std::end(array_for_radixsort));

    return 0;
}

1. I would like to be able to use array-like iterators. (they have no ::value_type).
2. I would like to be able to distinguish between say, int32_t and int64_t.

I'm at a complete loss as how to achieve this in any reasonably simple way. I.e. not specializing for every instance.

Answer 1

Use std::iterator_traits<T>::value_type to retrieve the value type of an iterator; it works for pointers as well as class-type iterators.

For dispatching, I would use template specialization to select the proper implementation (Live Demo):

namespace sort_selector {
// Default to using std::sort
template <typename T, typename = void>
struct dispatcher {
  template <typename Iterator>
  static void sort(Iterator begin, Iterator end) {
    std::cout << "Doing std::sort\n";
    std::sort(begin, end);
  }
};

// Use custom radix sort implementation for integral types
template <typename T>
struct dispatcher<T, typename std::enable_if<std::is_integral<T>::value>::type> {
  template <typename Iterator>
  static void sort(Iterator, Iterator) {
    std::cout << "Doing radix\n";
    // radix implementation
  }
};

// Use some other specific stuff for int32_t
template <>
struct dispatcher<int32_t, void> {
  template <typename Iterator>
  static void sort(Iterator, Iterator) {
    std::cout << "Specific overload for int32_t\n";
    // Do something
  }
};

// Dispatch appropriately
template <typename Iterator>
inline void sort(Iterator begin, Iterator end) {
  dispatcher<typename std::iterator_traits<Iterator>::value_type>::sort(begin, end);
}
} // namespace sort_selector

You should probably constrain sort_selector::sort to require random access iterators so your error messages are more digestible when someone inevitably tries to pass an improper iterator type:

namespace sort_selector {
// Dispatch appropriately
template <typename Iterator>
inline void sort(Iterator begin, Iterator end) {
  using traits = std::iterator_traits<Iterator>;
  static_assert(
    std::is_base_of<
      std::random_access_iterator_tag,
      typename traits::iterator_category
    >::value, "sorting requires random access iterators");
  dispatcher<typename traits::value_type>::sort(begin, end);
}
} // namespace sort_selector