[Bug middle-end/94940] [10/11 Regression] spurious -Warray-bounds for a zero length array member of union since r10-4300-g49fb45c81f4ac068

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94940

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think array_at_struct_end_p conservatively returns true for p->a[i] though.
Indeed all calls to the function return the correct value.  So is it somebody
invented a "more clever" variant of said check?  Or do we simply fail to call
it?  Ok, more clever it is:

bool
vrp_prop::check_array_ref (location_t location, tree ref,
                           bool ignore_off_by_one)
{
...
  if (!up_bound
      || TREE_CODE (up_bound) != INTEGER_CST
      || (warn_array_bounds < 2
          && array_at_struct_end_p (ref)))
    {
...
          const bool compref = TREE_CODE (arg) == COMPONENT_REF;
          if (compref)
            {
              /* Try to determine the size of the trailing array from
                 its initializer (if it has one).  */
              if (tree refsize = component_ref_size (arg, &interior_zero_len))
                if (TREE_CODE (refsize) == INTEGER_CST)
                  maxbound = refsize;

and refsize == 0 here.  Parsing an initializer for a union is tricky but
this one clearly gets the wrong answer from that.