Proposed resolution:

This wording is relative to N4885.

1. Modify [pointer.traits.types] as indicated:

   As additional drive-by fix the improper usage of the term "instantiation" has been corrected.

   using element_type = see below;
   

   -1- Type: Ptr::element_type if the qualified-id Ptr::element_type is valid and denotes a type ([temp.deduct]); otherwise, T if Ptr is a class template instantiationspecialization of the form SomePointer<T, Args>, where Args is zero or more type arguments; otherwise, the specialization is ill-formedpointer_traits has no member element_type.

[ 2022-02-22 LEWG telecon; Status changed: LEWG → Open ]

No objection to unanimous consent for Jonathan's suggestion to make pointer_traits an empty class when there is no element_type. Jonathan to provide a paper.

[ 2022-01-29; Daniel comments ]

This issue has some overlap with LWG 3665 in regard to the question how we should handle the rebind_alloc member template of the allocator_traits template as specified by [allocator.traits.types]/11. It would seem preferable to decide for the same approach in both cases.

[ 2021-05-10; Reflector poll ]

Priority set to 2. Send to LEWG. Daniel: "there is no similar treatment for the rebind member template and I think it should be clarified whether pointer_to's signature should exist and in which form in the offending case."

P1474R1 chose to use std::to_address (a mechanism of converting pointer-like types to raw pointers) for contiguous iterators. std::to_address provides an optional customization point via an optional member in std::pointer_traits. However all iterators are not pointers, and the primary template of std::pointer_traits<Ptr> requires that either Ptr::element_type is valid or Ptr is of the form template<T, Args...> or the pointer_traits specialization is ill-formed. This requires specializing pointer_traits for those contiguous iterator types which is inconvenient for users. P1474 should have also made pointer_traits SFINAE friendly.