[libcxx-commits] [libcxx] [libc++] Collapse `optional<T&>` inheritance hierarchy (PR #215286)

[A. Jiang via libcxx-commits]

================
@@ -54,40 +54,47 @@ _LIBCPP_PUSH_MACROS
 _LIBCPP_BEGIN_NAMESPACE_STD
 
 template <class _Tp>
-struct __optional_ref_base {
-  using value_type                 = _Tp;
-  using __raw_type _LIBCPP_NODEBUG = remove_reference_t<_Tp>;
-  __raw_type* __value_;
+struct __optional_ref_iterator_base {};
 
-  _LIBCPP_HIDE_FROM_ABI constexpr __optional_ref_base() noexcept : __value_(nullptr) {}
+#  if _LIBCPP_HAS_EXPERIMENTAL_OPTIONAL_ITERATOR
+
+template <class _Tp>
+  requires(is_object_v<_Tp> && !__is_unbounded_array_v<_Tp>)
+struct __optional_ref_iterator_base<_Tp&> {
+#    ifdef _LIBCPP_ABI_BOUNDED_ITERATORS_IN_OPTIONAL
+  using iterator = __bounded_iter<_Tp*>;
+#    else
+  using iterator = __capacity_aware_iterator<_Tp*, optional<_Tp&>, 1>;
+#    endif
+};
+
+#  endif
+
+template <class _Tp>
+class _LIBCPP_DECLSPEC_EMPTY_BASES optional<_Tp&> : public __optional_ref_iterator_base<_Tp&> {
----------------
frederick-vs-ja wrote:

It seems to me that `_LIBCPP_DECLSPEC_EMPTY_BASES` doesn't really affect the size of `optional<T&>` here as we're using single inheritance now. See https://learn.microsoft.com/en-us/cpp/cpp/empty-bases?view=msvc-170.

Maybe it's be better to drop addition of `_LIBCPP_DECLSPEC_EMPTY_BASES` in this PR and add test coverage on sizes (i.e. various `static_assert(sizeof(std::optional<T&>) == sizeof(T*));`) in a following-up PR.

https://github.com/llvm/llvm-project/pull/215286