[Bug c++/99851] Warn about operator new that takes std::nothrow_t but is potentially-throwing

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99851

Martin Sebor <msebor at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
                 CC|                            |msebor at gcc dot gnu.org
   Last reconfirmed|                            |2021-03-31

--- Comment #1 from Martin Sebor <msebor at gcc dot gnu.org> ---
Confirmed, thanks!  Just to make sure I understand: we want a warning for the
operator new declaration (irrespective of its definition) because the nothrow_t
argument suggests it doesn't throw but the absence of noexcept implies it
might.  I.e., the warning can be emitted as early as in the C++ front end.

The following modified test case shows that the the null test in main() is
eliminated on the basis of the missing noexcept:

$ cat pr99851.C && g++ -O1 -S -Wall -fdump-tree-optimized=/dev/stdout pr99851.C
namespace std {
  typedef __SIZE_TYPE__ size_t;
  const struct nothrow_t { } nothrow;
}

struct X
{
  void* operator new[](std::size_t n, const std::nothrow_t&) {
    return __builtin_malloc (n);
  }

  unsigned data = 0;
};

int main()
{
  void *p = new (std::nothrow) X[2];
  if (!p)
    __builtin_abort ();
  __builtin_printf ("%p\n", p);
}

;; Function main (main, funcdef_no=1, decl_uid=2393, cgraph_uid=5,
symbol_order=5) (executed once)

int main ()
{
  void * _9;

  <bb 2> [local count: 357878154]:
  _9 = __builtin_malloc (8);
  MEM[(struct X *)_9].data = 0;
  MEM[(struct X *)_9 + 4B].data = 0;
  __builtin_printf ("%p\n", _9);
  return 0;

}