The Load/Save tests had been validated against the schema, however there were a
substantial number of issues that were discovered when the test suite was run
against Xerces-J's implementation.

The only significant test change was DOMWriterFilterTest2.  In that test a
filter was used to strip all text nodes from the document.  It then tested
whether an attribute node had child nodes and expected it to have none.  Most
implementations use a specialized class for attribute and an attribute with no
child nodes doesn't exist in nature.  I've changed the test so that it checks
that the text content of the attribute is a empty string.

See http://lists.w3.org/Archives/Public/www-dom-ts/2003Dec/0000.html