LWG 22-04-29

[crtrott]

• Make proper concept [10a64f3c8e490199b3]
  • { M::is_always_strided() } -> same_as<bool>;
  • reorder do this before bool_constant

    template<class T> constexpr bool is-extents = false;
    template<class... T> constexpr bool is-extents<extents<T...>> = true;

Accessor

• a denotes a value of type A or const A [eb9bf80b6473]

• "The type A::pointer need not be dereferenceable." [eb9bf80b6473]

• p denotes a value of type A::pointer or const A::pointer [eb9bf80b6473]

• i, j, and j denotes values of type size_t [eb9bf80b6473]

• 5: drop swappable its in semiregular [02ddc845e8e]

  • do this for all other things

• Result: A type that models common_reference_with<A::reference&&, A::element_type&> [37ae082646]

• 11: its preconditions if any are met implicit in [233a72100553]

  • put Note first, since its unrelated accessible range,
  • drop "one could imagine" , [cd990d85246d6]
  • can impose instead of will generally, use i%size() [37ae082646]

Add Remarks: the expression is equality preserving. (both for a.access and a.offset) [37ae082646]

Front matter: [233a72100553]

An accessor policy defines types 
and operations by which a reference
to a single object is created from an abstract data handle to a number of such objects and an index.
For an accessor policy, a set of indicies `i` for  a given handle `p` for which access produces
valid reference to object, is defined as accesible range `[a, p)`.

• update accessible range wording [0,n) is an accessible range of p and a [233a72100553] [ecd50117d661df] [1ca5940c79209f]

q and b instead of p_sub and a_sub [233a721005]

default_accesor

• remove typename from offset_policy::pointer -> make it just pointer [fce6baeb0bd1cb]
• 1: requirements of accessor policy -> accessor policy requirements [fce6baeb0bd1cb]
• 4: for an instance of -> of an object of type default_accessor [b0673085842d0]
• 4: size_type -> size_t [b0673085842d0]

Constructor 1.1 OtherElementType()[], element_type()[] [fce6baeb0bd1cb] 2 Effects equivalent to return p+i; [fce6baeb0bd1cb]

3 Effects equivalent to return p[i] [fce6baeb0bd1cb]

move offset after access in synopsis and definitions [fce6baeb0bd1cb]

• submdspan

[crtrott]

mdspan:

move static functions before constructors [7a5a6ee2aaa35]

[mhoemmen]

Other points from Friday's review:

• Make the exposition-only thing is-extents a variable template with a variable template specialization.

  template<class T> constexpr bool is-extents = false;
  template<class... T> constexpr bool is-extents<extents<T...>> = true;

• With accessor: If it's move assignable, it's probably not a const type.

• It sounds like they prefer "A type such that common_reference_with<A::reference&&, A::element_type&> is modeled" to "A type that models common_reference_with<A::reference&&, A::element_type&>".

• Preferred accessible range notation: p + [0,n)

• Consider using "equality preserving" (a property of expressions; a word of power in [concepts.equality]) to talk about offset policy

• In [mdspan.accessor.default], change offset_policy::pointer to pointer, as it's a concrete type. Change size_type to size_t. Remove "for an instance of" in paragraph 4.

• Change "requirements of accessor policy" to "accessor policy requirements."

• In [mdspan.accessor.default.members], change typename default_accessor<OtherElementType>::element_type to OtherElementType.

• In [mdspan.mdspan], say "Effects: Equivalent to: return p + i;".