Proposed resolution:

This wording is relative to N4944.

1. Modify [ranges.syn] as indicated:

   
     template<class T>
       using iterator_t = decltype(ranges::begin(declval<T&&gt;()));    // freestanding
     template<range R>
       using sentinel_t = decltype(ranges::end(declval<R&&gt;()));      // freestanding
     template<rangeclass R>
       using const_iterator_t = const_iterator<iterator_t<R>>;       // freestanding
     template<range R>
       using const_sentinel_t = const_sentinel<sentinel_t<R>>;       // freestanding
   

[ Kona 2023-11-07; Status changed: Open → NAD. ]

[ Kona 2023-11-07; SG9 ]

SG9 considered this and noted it's obsolete due to LWG 3946. To support this case now would require changing ranges::cbegin to add a special case for arrays of unknown bound, which would require a paper.

[ 2023-06-01; Reflector poll ]

Set priority to 3 after reflector poll. Several votes for NAD. Send to SG9 to be looked at with P2836.

ranges::iterator_t<T> is unconstrained; but in contrast, ranges::const_iterator_t<T> constrains T to be a range, i.e., it must have an end.

    static_assert(std::is_same_v<std::ranges::iterator_t<int(&)[2]>, int*>);
    static_assert(std::is_same_v<std::ranges::iterator_t<int(&)[]>, int*>);  // OK
    static_assert(std::is_same_v<std::ranges::const_iterator_t<int(&)[2]>, const int*>);
    static_assert(std::is_same_v<std::ranges::const_iterator_t<int(&)[]>, const int*>);  // hard error