Proposed resolution:

This wording is relative to N4791.

1. Change [range.subrange.ctor] as indicated:

   constexpr subrange(I i, S s) requires (!StoreSize);
   

   -?- Expects: [i, s) is a valid range.

   -1- Effects: Initializes begin_ with i and end_ with s.

   constexpr subrange(I i, S s, iter_difference_t<I> n)
     requires (K == subrange_kind::sized);
   

   -2- Expects: [i, s) is a valid range, and n == ranges::distance(i, s).

[ 2019-01-20 Reflector prioritization ]

Set Priority to 0 and status to Tentatively Ready

The constructors of subrange place no requirements on the iterator and sentinel values from which a subrange is constructed. They allow constructions like subrange{myvec.end(), myvec.begin()} in which the resulting subrange isn't in the domain of the Range concept which requires that "[ranges::begin(r), ranges::end(r)) denotes a range" ([range.range]/3.1). We should forbid the construction of abominable values like this to enforce the useful semantic that values of subrange are always in the domain of Range.

(From ericniebler/stl2#597.)