Boost C++ Libraries: Ticket #2862: slist::swap is broken when cache_last< true > is used

Andrey Semashev — Tue, 17 Mar 2009 07:43:59 GMT

An addition: looks like the problem occurs only when one of the lists is empty.

Andrey Semashev — Tue, 17 Mar 2009 08:06:32 GMT

The following version of the priv_swap_cache_last method fixes the problem for me, although I'm not sure it's acceptable:

   void priv_swap_cache_last(slist_impl &other)
   {
      int choice = static_cast< int >(this->empty()) * 2 + static_cast< int >(other.empty());
      switch (choice)
      {
      case 0: // both lists are not empty
         {
            node_ptr other_last(other.get_last_node());
            node_ptr this_last(this->get_last_node());
            node_ptr other_bfirst(other.get_root_node());
            node_ptr this_bfirst(this->get_root_node());
            node_algorithms::transfer_after(this_bfirst, other_bfirst, other_last);
            node_algorithms::transfer_after(other_bfirst, other_last != other_bfirst? other_last : this_bfirst, this_last);
            node_ptr tmp(this->get_last_node());
            this->set_last_node(other.get_last_node());
            other.set_last_node(tmp);
         }
         break;
      case 1: // other.empty() == true
         other.splice_after(other.before_begin(), *this);
         break;
      case 2: // this->empty() == true
         this->splice_after(this->before_begin(), other);
         break;
      default:; // both lists are empty
      }
   }

Andrey Semashev — Tue, 17 Mar 2009 08:09:36 GMT

Oh, just noticed, there's no need for the "other_last != other_bfirst" comparison as it always yields true in the proposed priv_swap_cache_last version.

Andrey Semashev — Tue, 17 Mar 2009 08:15:03 GMT

On the second thought, ignore it. The splice_after breaks the constatnt_time_size behavior.

status changed; resolution set

anonymous — Wed, 25 Mar 2009 18:13:42 GMT

status new → closed
resolution → fixed

Fixed in revision 51971.