This prevents needing to do complex and poorly-defined updates of the
mapping table if the new parser has different token types than the old.
Per discussion.