Re: namelist array objects in subprograms

["Paul Thomas" <paulthomas2 at wanadoo dot fr>]

> Anything that needs to handle array values expressons must do so 
> explicitly.
> See gfc_conv_intrinsic_transfer and gfc_conv_function_call for examples.

I took a long look at these functions and found that they only partially 
applicable because the call to get_new_var_expr does not fill the expr->ref, 
so preventing the  tree walk.  Of course, as in these two functions, the 
array dummy argument, added to a namelist, has to be an indirect reference.

Since all that is available in the transfer of the namelist elements is the 
gfc_symbol,  so the need for an indirect reference can be detected using 
(sym->attr.dummy && sym->as->rank).  Thus in trans-io.c:

for (nml = dt->namelist->namelist; nml; nml = nml->next)
 {
   gfc_init_se (&se, NULL);
   gfc_init_se (&se2, NULL);
   nmlvar = get_new_var_expr (nml->sym);
   nmlname = gfc_new_nml_name_expr (nml->sym->name);
   gfc_conv_expr_reference (&se2, nmlname);
   gfc_conv_expr_reference (&se, nmlvar);
   gfc_evaluate_now (se.expr, &se.pre);

   if  ( nml->sym->attr.dummy && nml->sym->as && nml->sym->as->rank)
     se.expr = gfc_build_indirect_ref (se.expr);                 /* Here it 
is! */

   transfer_namelist_element (&block, &nml->sym->ts, nml->sym->as, se.expr,
                                               se2.expr, se2.string_length, 
nml->sym->name);
 }

It works correctly, does not cause any regressions but is it acceptable? 
Presumably there are clever things that can be done with the backend 
decalaration?

Paul T