Proposed resolution:

This wording is relative to N5001.

1. Modify [mdspan.layout.reqmts] as indicated:

   m.stride(r)
   

   -19- Preconditions: `m.is_strided()` is `true`.

   -20- Result: `typename M::index_type`

   -21- Returns: `sr` as defined in `m.is_strided()` above.

   [Note ?: It is not required for `m.stride(r)` to be well-formed if `m.extents().rank()` is zero, even if `m.is_always_strided()` is `true`. — end note]

[mdspan.layout.reqmts] p19-21 says that a layout mapping needs to provide `m.stride(r)`. However, [mdspan.layout.left.obs] p5 constrains layout_left::mapping<Extents>::stride(r) on Extents::rank() > 0 being `true`. The same is true of `layout_right::mapping` ([mdspan.layout.right.obs] p5). (The other Standard mappings in [views.multidim] and [linalg] do not have this constraint.) This suggests that a rank-zero `layout_{left,right}::mapping` does not conform with the layout mapping requirements.

On the other hand, other parts of [mdspan.layout.reqmts] imply that `r` must be in the range [0, `rank()`) for the layout mapping's extents type. If such an `r` does not exist, which is the case for a rank-zero layout mapping, then the `m.stride(r)` requirement is vacuous. This implies that a rank-zero `layout_{left,right}::mapping` fully conforms with the layout mapping requirements.

It is definitely the design intent for rank-zero `mdspan` to work, and for it to represent a view of a single element. Users can create rank-zero `mdspan` by invoking its constructor, or by using `submdspan` where all the slice arguments are convertible to `index_type`. Even though the normative wording permits this, adding a Note would clarify the design intent without making the wording redundant. This was the preferred change per LWG reflector discussion.