[Bug fortran/50981] [4.4/4.5/4.6 Regression] Wrong-code for scalarizing ELEMENTAL call with absent OPTIONAL argument

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50981

--- Comment #23 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-01-20 11:28:23 UTC ---
Created attachment 26392
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=26392
Small test case for polymorphic optional dummies

(In reply to comment #22)
> Updated patch.
> This passes the full comment #13 test fixed as follows:

Truly awesome! Will you submit the patch?

 * * *

Related, but a bit separate issue: OPTIONAL CLASS dummies. See attached test
case. It seems to mostly work with the patch below, except for one ICE and one
SEGV at run time. (And - out-commented part - I'm hitting an ICE which might be
the same as PR 46356 comment 2.)

TODO: In the fixed-up test case of comment 13, copy sub_t and change "type" to
"class" - copy all calls to sub_t - and modify those to call the new function.
See what will break ...

TODO2: Call both the original sub_t and the new function with polymorphic
components of derived types.

--- trans-expr.c        (revision 183328)
+++ trans-expr.c        (working copy)
@@ -576,8 +576,16 @@ gfc_conv_expr_present (gfc_symbol * sym)
   gcc_assert (sym->attr.dummy);

   decl = gfc_get_symbol_decl (sym);
-  if (TREE_CODE (decl) != PARM_DECL)
+
+  if (sym->ts.type == BT_CLASS)
     {
+      decl = gfc_class_data_get (decl);
+      if (CLASS_DATA (sym)->attr.dimension
+         || CLASS_DATA (sym)->attr.codimension)
+       decl = gfc_build_addr_expr (NULL_TREE, decl);
+    }
+  else if (TREE_CODE (decl) != PARM_DECL)
+    {
       /* Array parameters use a temporary descriptor, we want the real
          parameter.  */
       gcc_assert (GFC_DESCRIPTOR_TYPE_P (TREE_TYPE (decl))