TMP Part 3 - Tuple Iteration With Recursion

By on August 24, 2017

A fairly common task one encounters is iterating over a std::tuple. I’m going to keep this post fairly brief and go over how not to iterate over a tuple. We’ll discuss how you should iterate over a tuple in the next post.

The most naive way to iterate over a tuple is via recursive function calls using SFINAE to end the recursion. Here is one way to implement a transform over a tuple with recursion:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
template <size_t Index, typename... Elements, typename Func, typename... Args,
          std::enable_if_t<Index == 0>* = nullptr>
constexpr inline void tuple_transform(const std::tuple<Elements...>& tuple,
                                      Func&& func, Args&&... args) {
  func(std::get<Index>(tuple), std::integral_constant<size_t, Index>{},
       args...);
}

template <size_t Index, typename... Elements, typename Func, typename... Args,
          std::enable_if_t<Index != 0>* = nullptr>
constexpr inline void tuple_transform(const std::tuple<Elements...>& tuple,
                                      Func&& func, Args&&... args) {
  tuple_transform<Index - 1>(tuple, func, args...);
  func(std::get<Index>(tuple), std::integral_constant<size_t, Index>{},
       args...);
}

If you prefer, you can write the code such that the function counts up instead of down, but aside from that this is pretty much what we’re stuck with in C++11/14. In C++17, constexpr-if can be used so that we only have one function, but nevertheless we are doing recursion.

With the above could you can do both a fold, and a transform (referred to as map in functional languages). Here is an example of using a fold to compute the sum of elements in a tuple (yes, yes, in this case a std::vector would work fine, humor me for the purposes of this example).

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
const auto my_tupull = std::make_tuple(2, 7, -3.8, 20.9);
double sum_value = 0.0;
tuple_transform<3>(my_tupull, [](const auto& element, auto /*index*/,
                                 double& state) { state += element; },
                   sum_value);
std::cout << "Expected: 26.1   Computed: " << sum_value << "\n";

sum_value = 0.0;
tuple_transform<3>(my_tupull,
                   [](const auto& element, auto index, double& state) {
                     if (index.value != 1) {
                       state += element;
                     }
                   },
                   sum_value);
std::cout << "Expected: 19.1   Computed: " << sum_value << "\n";

A transform can be performed similarly but the compile time integral value must be extracted from index using decltype(index)::value. The compile time index can then be used to index a second tuple as follows:

1
2
3
4
5
6
7
8
9
10
const auto my_tupull = std::make_tuple(2, 7, -3.8, 20.9);
std::decay_t<decltype(my_tupull)> out_tupull;
tuple_transform<3>(my_tupull,
                   [](const auto& element, auto index, auto& out_tuple) {
                     constexpr size_t index_v = decltype(index)::value;
                     std::get<index_v>(out_tuple) = -element;
                   },
                   out_tupull);
std::cout << "Expected: (-2, -7, 3.8, -20.9)   Computed: " << out_tupull
          << "\n";

The lines of interest are 5 and 6. On 5 we assign the value held in the std::integral_constant to a constexpr variable, and on line 6 we use the constexpr variable to retrieve an element from the tuple.

As I mentioned at the beginning, this post is fairly short and I don’t go into much detail because this is the way you shouldn’t iterate over a tuple. I’m posting the correct way to iterate over a tuple at the same time here so you can compare them both.