We don't use Lexer.unread anywhere, so it's not worth putting in the
effort to fix this properly.