Bug 29458

This is fixed by the following patch:

Index: trans-array.c
===================================================================
--- trans-array.c       (revision 121280)
+++ trans-array.c       (working copy)
@@ -1280,6 +1280,7 @@
          gfc_conv_expr (&se, c->iterator->var);
          gfc_add_block_to_block (pblock, &se.pre);
          loopvar = se.expr;
+         TREE_NO_WARNING(loopvar) = 1;
 
          /* Make a temporary, store the current value in that
             and return it, once the loop is done.  */

Why not create a new i for the inner loop instead of saving it off?
Or is:
  integer :: n, i
  n = 5
  n = sum((/(i,i=1,f())/))
contains
  function f()
  integer :: f
  f = i+1
  end

valid and well defined?

*** Bug 31731 has been marked as a duplicate of this bug. ***

Will not be fixed in 4.2.0; retargeting at 4.2.1.

(In reply to comment #2)
> Why not create a new i for the inner loop instead of saving it off?

This is indeed what we should do; however, I can't find a way to work it out. Unassigning myself.

I'd just like to remind everyone that this is still an issue in gcc version 4.3.0 20071109

This bug also shows up when executing the example code for initializing random_seed in the gfortran documentation.  Is it a regression w.r.t.  4.0 or 4.1?  Maybe not, but it would be nice if something in the docs didn't generate a warning ;)

(In reply to comment #7)
> This bug also shows up when executing the example code for initializing
> random_seed in the gfortran documentation.  Is it a regression w.r.t.  4.0 or
> 4.1?  Maybe not, but it would be nice if something in the docs didn't generate
> a warning ;)

No, it is not a regression w.r.t to any previous version of gfortran.
The problem has always existed.  You're more than welcomed to fix it.

I tried this:

Index: gcc/fortran/trans-array.c
===================================================================
--- gcc/fortran/trans-array.c   (revision 131893)
+++ gcc/fortran/trans-array.c   (working copy)
@@ -1246,36 +1246,27 @@ gfc_trans_array_constructor_value (stmtb
          tree end;
          tree step;
          tree loopvar;
          tree exit_label;
          tree loopbody;
-         tree tmp2;
-         tree tmp_loopvar;

          loopbody = gfc_finish_block (&body);

          if (c->iterator->var->symtree->n.sym->backend_decl)
            {
-             gfc_init_se (&se, NULL);
-             gfc_conv_expr (&se, c->iterator->var);
-             gfc_add_block_to_block (pblock, &se.pre);
-             loopvar = se.expr;
+             loopvar = gfc_typenode_for_spec (&c->iterator->var->ts);
+             loopvar = gfc_create_var (loopvar, c->iterator->var->symtree->n.sym->name);
            }
          else
            {
              /* If the iterator appears in a specification expression in
                 an interface mapping, we need to make a temp for the loop
                 variable because it is not declared locally.  */
              loopvar = gfc_typenode_for_spec (&c->iterator->var->ts);
              loopvar = gfc_create_var (loopvar, "loopvar");
            }

-         /* Make a temporary, store the current value in that
-            and return it, once the loop is done.  */
-         tmp_loopvar = gfc_create_var (TREE_TYPE (loopvar), "loopvar");
-         gfc_add_modify_expr (pblock, tmp_loopvar, loopvar);
-
          /* Initialize the loop.  */
          gfc_init_se (&se, NULL);
          gfc_conv_expr_val (&se, c->iterator->start);
          gfc_add_block_to_block (pblock, &se.pre);
          gfc_add_modify_expr (pblock, loopvar, se.expr);
@@ -1293,10 +1284,12 @@ gfc_trans_array_constructor_value (stmtb
          /* If this array expands dynamically, and the number of iterations
             is not constant, we won't have allocated space for the static
             part of C->EXPR's size.  Do that now.  */
          if (dynamic && gfc_iterator_has_dynamic_bounds (c->iterator))
            {
+             tree tmp2;
+
              /* Get the number of iterations.  */
              tmp = gfc_get_iteration_count (loopvar, end, step);

              /* Get the static part of C->EXPR's size.  */
              gfc_get_array_constructor_element_size (&size, c->expr);
@@ -1339,13 +1332,10 @@ gfc_trans_array_constructor_value (stmtb
          gfc_add_expr_to_block (pblock, tmp);

          /* Add the exit label.  */
          tmp = build1_v (LABEL_EXPR, exit_label);
          gfc_add_expr_to_block (pblock, tmp);
-
-         /* Restore the original value of the loop counter.  */
-         gfc_add_modify_expr (pblock, loopvar, tmp_loopvar);
        }
     }
   mpz_clear (size);
 }


But something is still using the wrong 'i':

   <D.937>:;
    if (i.4D.918 > D.919)
      {
        goto L.1D.922;
      }
    else
      {

      }
    D.932 = atmp.2D.915.dataD.906;
    D.938 = (integer(kind=4)D.8[0] *) D.932;
    offset.6D.939 = offset.3D.917;
    (*D.938)[offset.6D.939] = iD.902; <<===== This is the wrong "i"!!
    offset.3D.917 = offset.3D.917 + 1;
    i.4D.918 = i.4D.918 + 1;
    goto <D.937>;

I also give up. Too time-consuming understanding what is going on...

For my classification of -Wuninitialized problems (see http://gcc.gnu.org/wiki/Better_Uninitialized_Warnings), this is caused by a representation issue.

Patch:
  http://gcc.gnu.org/ml/fortran/2009-04/msg00035.html

Subject: Bug 29458

Author: dfranke
Date: Sun Apr  5 18:02:00 2009
New Revision: 145564

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=145564
Log:
gcc/fortran/:
2009-04-05  Daniel Franke  <franke.daniel@gmail.com>

        PR fortran/29458
        * trans-array.c (gfc_trans_array_constructor_value): Shadow implied
        do-loop variable to avoid spurious middle-end warnings.


gcc/testsuite/:
2009-04-05  Daniel Franke  <franke.daniel@gmail.com>

        PR fortran/29458
        * gfortran.dg/implied_do_1.f90: New.


Added:
    trunk/gcc/testsuite/gfortran.dg/implied_do_1.f90
Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/trans-array.c
    trunk/gcc/testsuite/ChangeLog

Fixed in trunk. Backport unlikely as it's not a regression. Closing.