Ada's use of mark_all_temps_used

[Mike Stump <mrs at apple dot com>]

Ada does:

/* Insert the code for GNAT_NODE at the position saved for that node.  
*/

void
insert_code_for (gnat_node)
     Node_Id gnat_node;
{
  if (global_bindings_p ())
    {
      push_pending_elaborations ();
      gnat_to_code (gnat_node);
      Check_Elaboration_Code_Allowed (gnat_node);
      insert_elaboration_list (get_gnu_tree (gnat_node));
      pop_pending_elaborations ();
    }
  else
    {
      rtx insns;

      do_pending_stack_adjust ();
      start_sequence ();
      mark_all_temps_used ();
      gnat_to_code (gnat_node);
      do_pending_stack_adjust ();
      insns = get_insns ();
      end_sequence ();
      emit_insn_after (insns, RTL_EXPR_RTL (get_gnu_tree (gnat_node)));
    }
}

I have a question, do you really mean to mark all temp slots, including 
those that had been allocated and freed already as used and kept?

mark_all_temps_used ()
{
  struct temp_slot *p;

  for (p = temp_slots; p; p = p->next)
    {
      p->in_use = p->keep = 1;
      p->level = MIN (p->level, temp_slot_level);
    }
}

It seems slightly beyond clean to do this.  What if the compiler grabs 
a temp slot for pic or something after you freed a slot, and before you 
mark all temps used?  Also, if the temp slot had been in use before but 
not now, why preserve the previous level?  Isn't temp_slot_level enough?

I'm redoing the slot code and came across these interesting semantics.