Proposed resolution:

This wording is relative to N5014.

1. Modify [sequence.reqmts] as indicated:

   a.assign_range(rg)
   

   -60- Result: void

   -61- Mandates: assignable_from<T&, ranges::range_reference_t<R>> is modeled. For `inplace_vector`, if `ranges::size(rg)` is a constant expression then `ranges::size(rg)` ≤ `a.max_size()`.

2. Modify [inplace.vector.cons] as indicated:

   template<container-compatible-range<T> R>
     constexpr inplace_vector(from_range_t, R&& rg);
   

   -?- Mandates: If `ranges::size(rg)` is a constant expression then `ranges::size(rg)` ≤ `N`.

   -9- Effects: Constructs an `inplace_vector` with the elements of the range `rg`.

   -10- Complexity: Linear in `ranges::distance(rg)`.

Consider:

std::array<int, 42> a;
std::inplace_vector<int, 5> v(std::from_range, a);

The above throws `std::bad_alloc` at runtime because the size of `array` is larger than capacity of `inplace_vector`. However, we should reject it at compile time since the `array` size is a constant expression.

Given that we do a lot of compile-time size checking in <simd>, it's worth applying that here as well. Compile-time errors are better than runtime ones.