[Bug middle-end/58270] Wrong code while accessing trailing array elements in a global common structure

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58270

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|i386-pc-linux               |
             Status|WAITING                     |NEW
                 CC|                            |rguenth at gcc dot gnu.org
          Component|fortran                     |middle-end
               Host|i386-pc-linux               |
            Summary|Wrong code while accessing  |Wrong code while accessing
                   |array elements in a global  |trailing array elements in
                   |structure                   |a global common structure
      Known to fail|                            |4.9.0

--- Comment #19 from Richard Biener <rguenth at gcc dot gnu.org> ---
DECL_COMMONs could be special-cased in places that look at DECL_SIZE (I'll
declare this a QOI issue).  Or all decls that do not bind locally.  The
place that would "fix" tree-fre, tree-pre and tree-dse is in
get_ref_base_and_extent where it does

  if (DECL_P (exp))
    {
      /* If maxsize is unknown adjust it according to the size of the
         base decl.  */
      if (maxsize == -1
          && host_integerp (DECL_SIZE (exp), 1))
        maxsize = TREE_INT_CST_LOW (DECL_SIZE (exp)) - hbit_offset;
    }
  else if (CONSTANT_CLASS_P (exp))
    {
      /* If maxsize is unknown adjust it according to the size of the
         base type constant.  */
      if (maxsize == -1
          && host_integerp (TYPE_SIZE (TREE_TYPE (exp)), 1))
        maxsize = TREE_INT_CST_LOW (TYPE_SIZE (TREE_TYPE (exp))) - hbit_offset;
    }

In its general processing the function treats all trailing arrays as possibly
of undefined size.

Note that because -fcommon is still the default for all C dialects the
impact of changing the above for example in the simple && !DECL_COMMON (exp)
way is unknown.  -fcommon is a source of interesting bugs.