[Bug c++/55513] [4.7/4.8 Regression] Incorrect snprintf folding when building with -std=c++0x

["jakub at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55513

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-11-28 15:51:21 UTC ---
Reduced testcase:
void
f1 ()
{
  char s[10];
  const int t = __builtin_snprintf (s, 10, "Hello");
  __builtin_printf ("%d %s\n", t, s);
}

void
f2 ()
{
  char s[10];
  const int t = (__builtin_memcpy (s, "Hello", 6), 5);
  __builtin_printf ("%d %s\n", t, s);
}

int
main ()
{
  f1 ();
  f2 ();
}

Even f2 fails to call memcpy, so it isn't related to folding the snprintf into
what it folds to, but about handling COMPOUND_EXPRs with side-effects on the
LHS where the only side-effects are builtin calls.  If the LHS of the
COMPOUND_EXPR is a non-builtin, potential_rvalue_constant_expression returns
false and it is handled correctly, but even if it is say __builtin_exit (0) or
similar, it is optimized away.