[Bug fortran/31608] wrong types in character array/scalar binop

[pault at gcc dot gnu dot org]

------- Comment #61 from pault at gcc dot gnu dot org  2007-11-16 09:23 -------
Richard,

I believe that this is the right outcome for achar_4?

            (*(char[0:][1:1] *) atmp.4.data)[S.5][1]{lb: 1 sz: 1} =
*(_gfortran_
compare_string (D.529, &(*(char[0:][1:1] *) atmp.2.data)[S.5][1]{lb: 1 sz: 1},
1
, &"a"[1]{lb: 1 sz: 1}) >= 0 && _gfortran_compare_string (D.541,
&(*(char[0:][1:
1] *) atmp.3.data)[S.5][1]{lb: 1 sz: 1}, 1, &"z"[1]{lb: 1 sz: 1}) <= 0 ?
&char.6
 : &(*(char[0:][1:1] *) atmp.1.data)[S.5][1]{lb: 1 sz: 1});

Obtained with the following, not yet regtested, patch:

Index: gcc/fortran/trans-array.c
===================================================================
*** gcc/fortran/trans-array.c   (révision 130174)
--- gcc/fortran/trans-array.c   (copie de travail)
*************** gfc_conv_expr_descriptor (gfc_se * se, g
*** 4734,4748 ****
        gfc_add_block_to_block (&block, &rse.pre);
        gfc_add_block_to_block (&block, &lse.pre);

!       if (TREE_CODE (rse.expr) != INDIRECT_REF)
!       {
!         lse.string_length = rse.string_length;
!         tmp = gfc_trans_scalar_assign (&lse, &rse, expr->ts, true,
!                                 expr->expr_type == EXPR_VARIABLE);
!         gfc_add_expr_to_block (&block, tmp);
!       }
!       else
!       gfc_add_modify_expr (&block, lse.expr, rse.expr);

        /* Finish the copying loops.  */
        gfc_trans_scalarizing_loops (&loop, &block);
--- 4734,4743 ----
        gfc_add_block_to_block (&block, &rse.pre);
        gfc_add_block_to_block (&block, &lse.pre);

!       lse.string_length = rse.string_length;
!       tmp = gfc_trans_scalar_assign (&lse, &rse, expr->ts, true,
!                                    expr->expr_type == EXPR_VARIABLE);
!       gfc_add_expr_to_block (&block, tmp);

        /* Finish the copying loops.  */
        gfc_trans_scalarizing_loops (&loop, &block);
Index: gcc/fortran/trans-expr.c
===================================================================
*** gcc/fortran/trans-expr.c    (révision 130174)
--- gcc/fortran/trans-expr.c    (copie de travail)
*************** gfc_conv_string_parameter (gfc_se * se)
*** 3597,3604 ****
    type = TREE_TYPE (se->expr);
    if (TYPE_STRING_FLAG (type))
      {
!       gcc_assert (TREE_CODE (se->expr) != INDIRECT_REF);
!       se->expr = gfc_build_addr_expr (pchar_type_node, se->expr);
      }

    gcc_assert (POINTER_TYPE_P (TREE_TYPE (se->expr)));
--- 3597,3611 ----
    type = TREE_TYPE (se->expr);
    if (TYPE_STRING_FLAG (type))
      {
!       if (TREE_CODE (se->expr) != INDIRECT_REF)
!         se->expr = gfc_build_addr_expr (pchar_type_node, se->expr);
!       else
!       {
!         type = gfc_get_character_type_len (gfc_default_character_kind,
!                                            se->string_length);
!         type = build_pointer_type (type);
!         se->expr = gfc_build_addr_expr (type, se->expr);
!       }
      }

    gcc_assert (POINTER_TYPE_P (TREE_TYPE (se->expr)));

I am just putting it on to regtest now.  I am rather convinced that it will
cause some regressions but that it is the right way to go. We'll see!

Cheers

Paul


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31608