parser_tuple2st() and a failure to propogate an error in
build_node_children() (masking yet another leak, of course!).
This closes SF bug #485133 (confirmed by Insure++).