>Why do you say it works for only relatively simple queries? Because the probability of encountering something that can't be analyzed (for example use of the ancestor axis) increases with the size and complexity of the query, and my guess is that for queries of more than 100 lines or so, such a construct occurs more often than not.