> The second question, is could this be done other ways, for example a
> package of UDF functions that call an external XML parser/engine, to
> do the grunt work. What are the benefits vs costs of doing it this
> way instead?

There was an article in Russian community about using XML in
relational databases (http://www.ibase.ru/devinfo/xmldb.htm, if I
understand correctly they used Yaffil). Author claimed that in many
cases it is convenient to use UDFs (in particular XPath UDF,
http://www.ibase.ru/devinfo/xpathdll.zip) to manipulate
semi-structured information in XML format. For better performance he
was suggesting to use indices on expressions (Yaffil team has
implemented this).

From the discussion that followed his article I understood that many
people already use XML stored in BLOBs as well as XML decomposed to
relational structure. Also I did not hear (but maybe I missed it) any
request for some special treatment in the engine (except indices on
expressions).

Best regards,
Roman Rokytskyy