C++20 std::span: Uses of converting constructor

Question

The converting constructor of C++ std::span of the form:

template<class OtherElementType, size_t OtherExtent>
  constexpr explicit(see below) span(
     const span<OtherElementType, OtherExtent>& s) noexcept;

with constraints, can be used for the following purposes:

int arr[]{1, 2, 3};

span<int, 3> sf{arr};
// static span to dynamic span.
span<int> sd{sf};

// non-const to const span.
span<const int> scd{sd};
span<const int, 3> sfc{sf};

// dynamic to static span.
span<int, 3> sf2{sd};

What are some of the other uses this constructor which satisfy this constraint?

is_convertible_v<OtherElementType(*)[], element_type(*)[]> is true. [ Note: The intent is to allow only qualification conversions of the OtherElementType to element_type. — end note ]

As the note says it allows for _any_ qualification conversion, e.g. `int**` to `int const* const* const`, etc. — user17732522, Jun 27 '23 at 06:45

Jan Schultke · Answer 1 · 2023-06-27T08:55:19.073

The purpose is to allow implicitly converting between spans, as long as this conversion is a qualifying conversion and the sizes match. It can also be used for explicit conversions like in your exaples, but (8) is special because it's explicit only in a few cases. For example:

void process_all(std::span<const int> numbers);

std::span<int> get_range();

void foo() {
    auto r = get_range(); // ok, sizes match and qualifying conversion takes place
    process_all(r);
}

This is possible, because there is a qualifying conversion from int to const int, so we can convert std::span<int> to std::span<const int>. This use case is quite common, because making const std::span<T> doesn't ensure immutability of the elements, only a std::span<const T> does that.

This qualifying conversion can actually be multiple levels deep, e.g. we can:

take a std::span<const int * const * const> as a function parameter
pass a std::span<int**> to it

However, that's not a unique property of this constructor, (7) does the same. You may ask: "Why don't we just use constructor (7)?"

template< class R >
explicit(extent != std::dynamic_extent)
constexpr span( R&& range ); // (7)

This constructor works with any sized contiguous range, and a span is one. However, it is always explicit for a static extent, and only constructor (8) allows an implicit conversion between equally sized static spans. For example:

#include <span>

void process_all(std::span<const int, 100> numbers);

std::span<int, 100> get_range();
std::span<int, 10> get_short_range();
std::span<int> get_dynamic_range();

void foo() {
    process_all(get_range());         // OK
    process_all(get_short_range());   // ill-formed, need explicit conversion
    process_all(get_dynamic_range()); // ill-formed, need explicit conversion
}

The examples that you've given all use explicit conversions, so they don't demonstrate what makes (8) special, namely that it's conditionally explicit in fewer situations than other constructors.

C++20 std::span: Uses of converting constructor

1 Answers1