[Bug c++/106212] New: Code becomes non-constexpr with _GLIBCXX_DEBUG

[Bug c++/106212] New: Code becomes non-constexpr with _GLIBCXX_DEBUG

[fiesh at zefix dot tv]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106212

            Bug ID: 106212
           Summary: Code becomes non-constexpr with _GLIBCXX_DEBUG
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: fiesh at zefix dot tv
  Target Milestone: ---

Consider the following C++20 code:

#include <array>
struct P {
  constexpr P(int i) : e{i} {}
  auto operator<=>(P const &) const = default;
  std::array<int, 1> e;
};
constexpr auto operator-(P lhs, P) { return lhs; }
constexpr P p{1}, q{0};
static_assert(p - q != 0);


It behaves as follows:
* g++ compiles it successfully
* g++ with -D_GLIBCXX_DEBUG does not compile it
* clang++ compiles it successfully
* clang++ -D_GLIBCXX_DEBUG compiles it successfully

See also https://godbolt.org/z/ve19v1re9

[Bug c++/106212] Code becomes non-constexpr with _GLIBCXX_DEBUG

[fiesh at zefix dot tv]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106212

--- Comment #1 from fiesh at zefix dot tv ---
Oh and this appears to be a regression introduced in GCC 12.

[Bug c++/106212] Code becomes non-constexpr with _GLIBCXX_DEBUG

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106212

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2022-07-06
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1

--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to fiesh from comment #1)
> Oh and this appears to be a regression introduced in GCC 12.

Seems to be a library change in GCC 12 which triggers a pre-existing compiler
bug. This fails with all versions of GCC -std=c++20 but compiles with clang
-std=c++20 and EDG --c++20


namespace std
{
  using size_t = decltype(sizeof(0));

  template<typename _Tp, size_t _Nm>
    struct array
    {
      _Tp _M_elems[_Nm];

      constexpr const _Tp*
      data() const noexcept
      { return _M_elems; }
    };

    template<typename _Tp>
    constexpr
    bool
    __check_singular(_Tp* const& __ptr)
    { return __ptr == 0; }

  template<typename _Tp, size_t _Nm>
    constexpr
    bool
    operator==(const array<_Tp, _Nm>& __one, const array<_Tp, _Nm>& __two)
    {
      if (__check_singular(__one.data()))
        throw 1;
      return &__one == &__two;
    }
}

struct P {
  constexpr P(int i) : e{i} {}
  bool operator==(P const &) const = default;
  std::array<int, 1> e;
};
constexpr auto operator-(P lhs, P) { return lhs; }
constexpr P p{1}, q{0};
static_assert(p - q != 0);

[Bug c++/106212] Code becomes non-constexpr with _GLIBCXX_DEBUG

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106212

--- Comment #3 from Jonathan Wakely <redi at gcc dot gnu.org> ---
a.C:39:21: error: non-constant condition for static assertion
   39 | static_assert(p - q != 0);
      |               ~~~~~~^~~~
a.C:39:24:   in 'constexpr' expansion of 'operator-(p, q).P::operator==(P(0))'
a.C:34:8:   in 'constexpr' expansion of 'std::operator==<int, 1>(((const
P*)this)->P::e, <anonymous>.P::e)'
a.C:19:20: error: '(((const int*)(&<anonymous>.P::e.std::array<int,
1>::_M_elems)) == 0)' is not a constant expression
   19 |     { return __ptr == 0; }
      |              ~~~~~~^~~~

[Bug c++/106212] Code becomes non-constexpr with _GLIBCXX_DEBUG

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106212

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |DUPLICATE
             Status|NEW                         |RESOLVED

--- Comment #4 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Looks like a dup of PR 85944 since it only happens at global scope. The same
static_assert works fine in a function:

void f()
{
static_assert(p - q != 0);
}

*** This bug has been marked as a duplicate of bug 85944 ***