On Fri, Oct 17, 2025 at 06:34:41PM +0200, Tomasz Kaminski wrote: > On Fri, Oct 17, 2025 at 6:33 PM Tomasz Kaminski <[email protected]> wrote: > > > > > > > On Fri, Oct 17, 2025 at 6:27 PM Hewill Kang <[email protected]> wrote: > > > >> I'm not sure what you mean, but it does weird to me that the underlying > >> types can't be compared, but the corresponding tuple-like can be compared: > >> > >> #include <tuple> > >> #include <array> > >> > >> struct S { > >> void operator== (const S&) const = delete; > >> void operator<=>(const S&) const = delete; > >> void operator <(const S&) const = delete; > >> }; > >> > >> int main() { > >> std::tuple<> t; > >> std::array<S, 0> a; > >> // a == a; // error > >> t < a; // ok > >> > > Yes, because the < is doing more efficient implementation of: > > t < std::tuple<>(a); > > > And this constructor alos does not look on a, and it can be non-movable. > I think this is more a problem of array<S, 0> not being comparable, despite > being copyable regardless of S. > > > > > > >> t == a; // ok > >> t <=> a; // ok > >> } > >> > >> Tomasz Kaminski <[email protected]> 於 2025年10月18日 週六 上午12:20寫道： > >> > >>> > >>> > >>> On Fri, Oct 17, 2025 at 6:00 PM Hewill Kang <[email protected]> wrote: > >>> > >>>> Wow, I just noticed that the current wording allows tuple<> to be > >>>> compared to any array<T, 0>, regardless of the type of T. > >>>> However, tuple<T> can only be compared to array<T, 1> if T itself is > >>>> comparable. > >>>> > >>> They common_reference is tuple<>, but they are not totally_ordered_with, > >>> as array is not ordered. > >>> > >>>> This seems to be a defect. > >>>> > >>>> Osama Abdelkader <[email protected]> 於 2025年10月17日 週五 > >>>> 下午11:27寫道： > >>>> > >>>>> Fixes #119721 > >>>>> > >>>>> This fixes a C++23 compliance issue where std::tuple<> cannot be > >>>>> compared > >>>>> with std::array<T, 0>. Both are empty tuple-like types and should be > >>>>> comparable according to the C++23 standard. > >>>>> > >>>>> The fix adds the missing comparison operators: > >>>>> - operator== and operator!= (return true and false respectively) > >>>>> - operator<, operator<=, operator>, operator>= (follow ordering rules > >>>>> for empty types) > >>>>> - operator<=> (returns strong_ordering::equal) > >>>>> > >>>>> This resolves the 'rejects-valid' bug where GCC was rejecting valid > >>>>> C++23 code > >>>>> that Clang with libc++ already supports correctly. > >>>>> > >>>>> libstdc++-v3/ChangeLog: > >>>>> > >>>>> * include/std/tuple: Add comparison operators between tuple<> > >>>>> and array<T, 0>. > >>>>> > >>>>> Signed-off-by: Osama Abdelkader <[email protected]> > >>>>> --- > >>>>> libstdc++-v3/include/std/tuple | 88 ++++++++++++++++++++++++++++++++++ > >>>>> 1 file changed, 88 insertions(+) > >>>>> > >>>>> diff --git a/libstdc++-v3/include/std/tuple > >>>>> b/libstdc++-v3/include/std/tuple > >>>>> index 0ca616f1b..d3621dd16 100644 > >>>>> --- a/libstdc++-v3/include/std/tuple > >>>>> +++ b/libstdc++-v3/include/std/tuple > >>>>> @@ -1880,6 +1880,94 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION > >>>>> using _Cat = typename > >>>>> __tuple_like_common_comparison_category<_UTuple>::type; > >>>>> return std::__tuple_cmp<_Cat>(__t, __u, > >>>>> index_sequence_for<_Elements...>()); > >>>>> } > >>>>> + > >>>>> + // Comparison operators between tuple<> and array<T, 0> > >>>>> + // These are needed because both are empty tuple-like types in C++23 > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr bool > >>>>> + operator==(const tuple<>&, const array<_Tp, 0>&) > >>>>> + { return true; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr bool > >>>>> + operator==(const array<_Tp, 0>&, const tuple<>&) > >>>>> + { return true; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr bool > >>>>> + operator!=(const tuple<>&, const array<_Tp, 0>&) > >>>>> + { return false; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr bool > >>>>> + operator!=(const array<_Tp, 0>&, const tuple<>&) > >>>>> + { return false; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr bool > >>>>> + operator<(const tuple<>&, const array<_Tp, 0>&) > >>>>> + { return false; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr bool > >>>>> + operator<(const array<_Tp, 0>&, const tuple<>&) > >>>>> + { return false; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr bool > >>>>> + operator<=(const tuple<>&, const array<_Tp, 0>&) > >>>>> + { return true; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr bool > >>>>> + operator<=(const array<_Tp, 0>&, const tuple<>&) > >>>>> + { return true; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr bool > >>>>> + operator>(const tuple<>&, const array<_Tp, 0>&) > >>>>> + { return false; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr bool > >>>>> + operator>(const array<_Tp, 0>&, const tuple<>&) > >>>>> + { return false; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr bool > >>>>> + operator>=(const tuple<>&, const array<_Tp, 0>&) > >>>>> + { return true; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr bool > >>>>> + operator>=(const array<_Tp, 0>&, const tuple<>&) > >>>>> + { return true; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr strong_ordering > >>>>> + operator<=>(const tuple<>&, const array<_Tp, 0>&) > >>>>> + { return strong_ordering::equal; } > >>>>> + > >>>>> + template<typename _Tp> > >>>>> + [[nodiscard]] > >>>>> + constexpr strong_ordering > >>>>> + operator<=>(const array<_Tp, 0>&, const tuple<>&) > >>>>> + { return strong_ordering::equal; } > >>>>> + > >>>>> #endif // C++23 > >>>>> > >>>>> #else // ! (concepts && consteval) > >>>>> -- > >>>>> 2.43.0 > >>>>> > >>>>>
Thanks for the review, I just sent v2 which addresses your feedback, would appreciate your review there. Thanks, Osama
