[Bug c++/119149] New: Bogus "control reaches end of non-void function" warning with ill-formed expression on RHS of comma operator

[redi at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119149

            Bug ID: 119149
           Summary: Bogus "control reaches end of non-void function"
                    warning with ill-formed expression on RHS of comma
                    operator
           Product: gcc
           Version: 15.0
            Status: UNCONFIRMED
          Keywords: diagnostic
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: redi at gcc dot gnu.org
  Target Milestone: ---

void g(long&);

int good(int i)
{
  if (i > 0)
    return i;
  __builtin_abort();
  g(i);
}

int bad(int i)
{
  if (i > 0)
    return i;
  (__builtin_abort(), (void) g(i));
}

Both functions are ill-formed due to calling g(i) but the second one gives a
bogus warning:

warn.cc: In function 'int good(int)':
warn.cc:8:5: error: cannot bind non-const lvalue reference of type 'long int&'
to a value of type 'int'
    8 |   g(i);
      |     ^
warn.cc:1:8: note: initializing argument 1 of 'void g(long int&)'
    1 | void g(long&);
      |        ^~~~~
warn.cc: In function 'int bad(int)':
warn.cc:15:32: error: cannot bind non-const lvalue reference of type 'long
int&' to a value of type 'int'
   15 |   (__builtin_abort(), (void) g(i));
      |                                ^
warn.cc:1:8: note: initializing argument 1 of 'void g(long int&)'
    1 | void g(long&);
      |        ^~~~~
warn.cc:16:1: warning: control reaches end of non-void function [-Wreturn-type]
   16 | }
      | ^


The problematic construct in the second one matches what libstdc++ does since
r15-7856-gd87c0d5443ba86

In std::expected::value() we have:

      constexpr const _Tp&
      value() const &
      {
        static_assert( is_copy_constructible_v<_Er> );
        if (_M_has_value) [[likely]]
          return _M_val;
        _GLIBCXX_THROW_OR_ABORT(bad_expected_access<_Er>(_M_unex));
      }

When the static_assert fails, the bad_expected_access<_Er>(_M_unex) expression
is ill-formed.

With -fexceptions we get the static_assert and an error for the ill-formed
bad_expected_access initialization (I would prefer to not get the latter, but
that's Bug 96286).

Previously with -fno-exceptions we would just get the failed static_assert, now
we get the failed static_assert and the ill-formed initialization, *and* a
-Wreturn-type warning.