<string>: Consider improving conformance for char_traits

[achabense]

1. According to the standard(ref), char_traits::eq/lt/to_char_type/to_int_type/eq_int_type/not_eof should take values instead of references as parameters.
   (
   Also in
   https://github.com/microsoft/STL/blob/a62109595b6d89e08172fdf4beb75a2670fe0cc9/stl/inc/xstring#L305-L327?plain=1
   https://github.com/microsoft/STL/blob/a62109595b6d89e08172fdf4beb75a2670fe0cc9/stl/inc/xstring#L449-L471?plain=1
   )
   

   STL/stl/inc/xstring

   Lines 193 to 215 in a621095

   	_NODISCARD static constexpr bool eq(const _Elem& _Left, const _Elem& _Right) noexcept {
   	return _Left == _Right;
   	}
   	_NODISCARD static constexpr bool lt(const _Elem& _Left, const _Elem& _Right) noexcept {
   	return _Left < _Right;
   	}
   	_NODISCARD static constexpr _Elem to_char_type(const int_type& _Meta) noexcept {
   	return static_cast<_Elem>(_Meta);
   	}
   	_NODISCARD static constexpr int_type to_int_type(const _Elem& _Ch) noexcept {
   	return static_cast<int_type>(_Ch);
   	}
   	_NODISCARD static constexpr bool eq_int_type(const int_type& _Left, const int_type& _Right) noexcept {
   	return _Left == _Right;
   	}
   	_NODISCARD static constexpr int_type not_eof(const int_type& _Meta) noexcept {
   	return _Meta != eof() ? _Meta : !eof();
   	}

2. Also, int_types for char_traits<char16_t>, <char32_t> and <wchar_t> should be uint_least16_t, uint_least32_t and wint_t respectively. Currently they are defined as unsigned short, unsigned int and unsigned short. --They just works now. However, I think it will be better to use aliases directly, or at least add some static_asserts(e.g. static_assert(std::is_same_v<wint_t,unsigned short>).
   

   STL/stl/inc/xstring

   Lines 337 to 344 in a621095

   	template <>
   	struct char_traits<char16_t> : _WChar_traits<char16_t> {};
   	template <>
   	struct char_traits<char32_t> : _Char_traits<char32_t, unsigned int> {};
   	template <>
   	struct char_traits<wchar_t> : _WChar_traits<wchar_t> {};

   
   (char_traits<char>,<char8_t> has no such issue)

3. (off topic) We can safely add noexcept /*strengthened*/ for this function:
   

   STL/stl/inc/xstring

   Lines 1402 to 1404 in a621095

   	_NODISCARD constexpr int compare(_In_z_ const _Elem* const _Ptr) const { // compare [0, _Mysize) with [_Ptr, <null>)
   	return compare(basic_string_view(_Ptr));
   	}

4. (off topic; question) What does the template(line2547~2549) do here? _Alloc2 is not used in the body, and _Alloc is directly used as allocator in other member functions.
   (
   Also here; the only another use of this pattern
   https://github.com/microsoft/STL/blob/a62109595b6d89e08172fdf4beb75a2670fe0cc9/stl/inc/xstring#L2560-L2566?plain=1
   )
   

   STL/stl/inc/xstring

   Lines 2547 to 2553 in a621095

   	#if _HAS_CXX17
   	template <class _Alloc2 = _Alloc, enable_if_t<_Is_allocator<_Alloc2>::value, int> = 0>
   	#endif // _HAS_CXX17
   	_CONSTEXPR20 basic_string(_In_z_ const _Elem* const _Ptr, const _Alloc& _Al)
   	: _Mypair(_One_then_variadic_args_t{}, _Al) {
   	_Construct<_Construct_strategy::_From_ptr>(_Ptr, _Convert_size<size_type>(_Traits::length(_Ptr)));
   	}

[frederick-vs-ja]

4. _Alloc2 is not used in the body, and _Alloc is directly used as allocator in other member functions.

This is needed by SFINAE. In order to constrain this constructor, we need to make it a function template and introduce a template parameter on which substitution may fail.

[achabense]

@frederick-vs-ja Can we simply remove this constraint? The constraint is not used elsewhere other than the two cases(basic_string(const CharT*, const _Alloc& alloc = _Alloc()) and basic_string(size_type, CharT, const _Alloc& alloc = _Alloc())) --_Alloc is the allocator for basic_string. Also, the impl defines separate basic_string(...) and basic_string(..., const _Alloc&) for basic_string(..., const _Alloc& alloc = _Alloc());.

[frederick-vs-ja]

Can we simply remove this constraint?

No. This constraint is added by LWG-3076, which should be considered as a defect report against C++17 (in which CTAD is introduced).

[frederick-vs-ja]

1. According to the standard(ref), char_traits::eq/lt/to_char_type/to_int_type/eq_int_type/not_eof should take values instead of references as parameters.

Wow, this is a missed change from WG21-N2349. We need to implement these C++11 changes.

FYI libstdc++ can't make the changes for char and wchar_t specializations due to ABI compatibility (https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80810).

Edit: find and one overload of assign take the character argument by reference, which looks weird to me. I've mailed to LWG Chair to submit an LWG issue for this.