Which way to go with PL/SQL and XML

Question

Hi fellows,
I have a general question to XML on Oracle.
I read already some parts of the three manuals coming with the online documentation of 9i
- XML developer kit XDK
- XML API - XDK and Oracle XML DB
- XML database developer guide Oracle XML DB.
Also I wrote already a PL/SQL package using DBMS_XMLGEN to generate XML data from relational Oracle tables.
Actually I like the simplicity how I can create XML data with this DBMS package but I still missing the features like using a DTD/XML Schema to generate XML data.
I still don't know what the best way would be to use these features of XML in Oracle.
Hopefully someone can give me some advise and possibly can point to some code samples so that I can see the proper use of the package.
How about DBMS_XMLQuery?
To use this package, would that be the better solution for my problem?
Any comment much appreciated.
Thanks
Fred

Answer 1

Hi Fred,
Having written a fair amount of PL/SQL code to generate and process XML in the past years, I can give you my opinion on the matter. If you look at the PL/SQL API's Oracle has been written to handle XML, you will find two types/categories of API's: PL/SQL code based on wrapping java (i.e. DBMS_XMLQUERY) and PL/SQL code based on C (i.e. DBMS_XMLGEN). The first category is usually slower, so I tend to use these only if there are no other options. You can measure execution times by executing the same query using both DBMS_XMLGEN. vs. DBMS_XMLQUERY.
I usually tend to store xml into relational tables by extracting the xml document in PL/SQL using the functions XMLType data type offers (i.e. extract, existNode, getStringVal, etc) function rather than using DBMS_XMLSTORE for flexibility reason (more control over what's being stored).
Besides the PL/SQL API's there are some rather neat SQL functions to handle XML (i.e. XMLElement, XMLForest, XMLAgg, XMLAttribute, etc) I like to use to build XML documents that are to complex for DBMS_XMLGEN. But, on the downside, I think these functions might cost considerable amounts of CPU, because behind the scene a DOM-tree is build in the SGA/UGA(?), especially on large XML structures or high concurrency systems. I have not measured this yet, so I say might.
Further, what I think is lacking from the Oracle API's in general is handing of CDATA sections and the fact that some XML functions default generate "pretty-printing" XML, which can not be switched of as far as I know. In my opinion XML should never be "pretty printed", because XML basically is about data transportation, not about lay-out. CDATA sections are not handled, even worse "<![CDATA[ ]]>" is being escaped (to &lt and &gt) and not properly interpreted when processed back to text.