[Patch, Fortran, F08] PR 55207: Variables declared in the main program should implicitly get the SAVE attribute

[Janus Weil]

>> (1) The test in pr47399 comment 0 gives an ICE:
>>
>> pr47399.f90: In function 'test':
>> pr47399.f90:32:0: internal compiler error: in gfc_trans_auto_array_allocation, at fortran/trans-array.c:5682
>>     integer, dimension (a_const%i()) :: x        ! #3
>
> I can confirm this. Will investigate.

Ok, my first impulse to fix this was ...


Index: gcc/fortran/decl.c
===================================================================
--- gcc/fortran/decl.c    (revision 208590)
+++ gcc/fortran/decl.c    (working copy)
@@ -3831,7 +3831,8 @@
      implicitly have the SAVE attribute.  */
   if ((gfc_current_state () == COMP_MODULE
        || gfc_current_state () == COMP_PROGRAM)
-      && !current_attr.save && (gfc_option.allow_std & GFC_STD_F2008) != 0)
+      && !current_attr.save && (gfc_option.allow_std & GFC_STD_F2008) != 0
+      && current_attr.flavor != FL_PARAMETER)
     current_attr.save = SAVE_IMPLICIT;

   colon_seen = 1;


... which unfortunately does not help. Second I simply removed the
assert which causes the ICE ...


Index: gcc/fortran/trans-array.c
===================================================================
--- gcc/fortran/trans-array.c    (revision 208590)
+++ gcc/fortran/trans-array.c    (working copy)
@@ -5679,7 +5679,6 @@
   type = TREE_TYPE (type);

   gcc_assert (!sym->attr.use_assoc);
-  gcc_assert (!TREE_STATIC (decl));
   gcc_assert (!sym->module);

   if (sym->ts.type == BT_CHARACTER


... which actually helps (i.e. it removes the ICE and makes the
example work correctly at runtime). Third I looked at the dump that is
generated:


  integer(kind=8) ubound.0;
  integer(kind=8) size.1;
  static integer(kind=4)[0:D.2341] * restrict x;
  integer(kind=8) D.2341;
  bitsizetype D.2342;
  sizetype D.2343;
  void * restrict D.2345;
  integer(kind=8) D.2344;

  try
    {
      ubound.0 = (integer(kind=8)) get_i ();
      size.1 = NON_LVALUE_EXPR <ubound.0>;
      size.1 = MAX_EXPR <size.1, 0>;
      D.2341 = size.1 + -1;
      D.2342 = (bitsizetype) (sizetype) NON_LVALUE_EXPR <size.1> * 32;
      D.2343 = (sizetype) NON_LVALUE_EXPR <size.1> * 4;
      D.2344 = size.1 * 4;
      D.2345 = (void * restrict) __builtin_malloc (MAX_EXPR <(unsigned
long) D.2344, 1>);
      x = (integer(kind=4)[0:D.2341] * restrict) D.2345;


What looks funny to me is that the ubound in the declaration of x
(i.e. 'D.2341') is set only some time after that declaration. This
makes me wonder whether the ICE we're seeing is just a symptom of a
dormant wrong-code problem here? I think I need to take a closer look
at the PR again to understand what's going on ...

Cheers,
Janus