https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111242
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
#define N 3
constexpr bool
foo ()
{
int *v = new int[N];
auto p = &v[N+1];
bool t = &v[0] == p;
delete[] v;
return t;
}
static_assert (!foo ());
constexpr bool
bar ()
{
int v[N];
auto p = &v[N+1];
bool t = &v[0] == p;
return t;
}
static_assert (!bar ());
The difference is that in the bar case, there is cxx_eval_array_reference
called which diagnoses this. But in the foo case we should I think diagnose
this in cxx_eval_binary_expression when handling POINTER_PLUS_EXPR, the lhs is
(int *) &heap []
where heap [] is an artificial VAR_DECL which has int[1][3] type, and the
second operand evaluates to 16 (i.e. 4 * sizeof (int)), so this is clearly out
of bounds pointer arithmetics.
