[Bug fortran/58754] ICE/Regression in [4.7/4.8/4.9] with allocatable character arrays

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58754

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-valid-code
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2013-10-16
                 CC|                            |burnus at gcc dot gnu.org
      Known to work|                            |4.6.3
         Depends on|                            |57456
     Ever confirmed|0                           |1
      Known to fail|                            |4.7.3, 4.8.1, 4.9.0

--- Comment #1 from Tobias Burnus <burnus at gcc dot gnu.org> ---
The segfault occurs for trans-array.c's gfc_array_init_size:

5064      if (expr3_elem_size != NULL_TREE)
5065        tmp = expr3_elem_size;
5066      else if (expr3 != NULL)
5067        {
...
5089        tmp = TYPE_SIZE_UNIT (gfc_get_element_type (type));
5092      element_size = fold_convert (size_type_node, tmp);

And here tmp == NULL. I think the problem is that characters have to be handled
separately.

There should be somewhere a PR which is about moving all the memory size
calculation to gfc_trans_allocate - which already does it for scalars - instead
of partially repeating it in gfc_array_init_size. (The current algorithm might
also cause multiple function evaluation, which is wrong, too.)

I believe unifying the could would be the proper fix (and good in terms of code
size/maintenance as well). -- The PR I meant is: PR57456.