<type_traits>: common_reference workaround still necessary?

[StephanTLavavej]

VS 2022 17.2 Preview 3 is now available. It contains internal MSVC-PR-382810 that fixed DevCom-876860 and its internal mirror VSO-1049320. We have workarounds for both bug numbers:

STL/stl/inc/type_traits

Lines 1248 to 1274 in 52505b9

	#if !defined(__EDG__) && !defined(__clang__) // TRANSITION, DevCom-876860
	template <class _Ty1, class _Ty2>
	using _Cond_res_if_right = // N4810 [meta.trans.other]/2.4
	decltype(false ? _Returns_exactly<_Ty1>() : _Returns_exactly<_Ty2>());
	template <class _Ty>
	using _Is_scalar_or_array = disjunction<is_scalar<_Ty>, is_array<_Ty>>;
	template <class _Ty1, class _Ty2, class = void>
	struct _Cond_res_workaround {};
	template <class _Ty1, class _Ty2>
	struct _Cond_res_workaround<_Ty1, _Ty2, void_t<_Cond_res_if_right<_Ty1, _Ty2>>> {
	using _Uty = remove_cvref_t<_Ty1>;
	using type = conditional_t<conjunction_v<is_same<_Uty, remove_cvref_t<_Ty2>>, _Is_scalar_or_array<_Uty>,
	disjunction<conjunction<is_lvalue_reference<_Ty1>, is_rvalue_reference<_Ty2>>,
	conjunction<is_rvalue_reference<_Ty1>, is_lvalue_reference<_Ty2>>>>,
	decay_t<_Copy_cv<remove_reference_t<_Ty1>, remove_reference_t<_Ty2>>>, _Cond_res_if_right<_Ty1, _Ty2>>;
	};
	template <class _Ty1, class _Ty2>
	using _Cond_res = typename _Cond_res_workaround<_Ty1, _Ty2>::type;
	#else // ^^^ workaround / no workaround vvv
	template <class _Ty1, class _Ty2>
	using _Cond_res = // N4810 [meta.trans.other]/2.4
	decltype(false ? _Returns_exactly<_Ty1>() : _Returns_exactly<_Ty2>());
	#endif // ^^^ no workaround ^^^

STL/tests/std/tests/P0784R7_library_support_for_more_constexpr_containers/test.cpp

Lines 217 to 221 in 52505b9

	#if defined(__clang__) || defined(__EDG__) // TRANSITION, VSO-1049320
	ranges::destroy_at(reinterpret_cast<U(*)[N]>(ptr));
	#else // ^^^ no workaround / workaround vvv
	ranges::destroy_at(reinterpret_cast<T(*)[N]>(const_cast<T*>(ptr)));
	#endif // TRANSITION, VSO-1049320

The P0784R7_library_support_for_more_constexpr_containers/test.cpp workaround can apparently be removed without issue. However, attempting to remove the <type_traits> workaround (which was added by GH-2592) causes a libcxx test to fail, llvm-project\libcxx\test\std\concepts\concepts.lang\concept.commonref\common_reference.compile.pass.cpp.

common_reference.compile.pass.cpp(38): error C2338: static_assert failed: 'std::common_reference_with<volatile T&, volatile U&&>'
common_reference.compile.pass.cpp(61): note: see reference to function template instantiation 'bool CheckCommonReferenceWith<int[5],int[5]>(void) noexcept' being compiled
common_reference.compile.pass.cpp(44): error C2607: static assertion failed

I don't know if this is:

• Remaining compiler bugginess that needs to be reported
• A library bug somewhere
• A libcxx test bug

[StephanTLavavej]

I suspect this may be remaining compiler bugginess - I experienced a bunch of compiler and test crashes with this workaround removal plus other changes, and I seem to be able to restore all my other changes without incident (I haven't tried running a full test pass with just the workaround removal yet).

Update: I found the root causes of the compiler crash (crash-after-error in an XFAILed ranges test) and test crashes (our assertion dialog avoidance machinery has bit-rotted), so the issue here is purely common_reference.compile.pass.cpp. 🎉

[CaseyCarter]

This is a duplicate of #2860.