Metacat

This issue was being caused by offset indexing that involved 3 methods in
DocumentImpl.java:

buildIndex()
traverseParents()
updatePathIndex()

The problem arose when the node data for ATTRIBUTE nodes and ELEMENT nodes is stored in different locations within the xml_nodes table. ATTRIBUTES store
their own data in a column of the same record, whereas ELEMENTS store their node data in a child TEXT node (an entirely different node record altogether, on linked by the parent node id relationship).

Now, these methods work together to create a HashMap of PathIndexEntry objects that associate the correct node data, path, and node id to be indexed, regardless of whether it is an ATTRIBUTE or an ELEMENT that is being indexed. Testing needs to be done.

Note: all metacat databases installed prior to this fix should run the action=buildindex task in order to remove all of the erroneous entries from before and create new path and node indices.

Fixed the implementation of the buildIndex function which was not
working for new document insertions. A previous fix in updatePathIndex
for ATTRIBUTE data inadvertantly caused a foreign key duplication
exception for insertions of ELEMENT nodes when multiple relative paths
exist. This fix simply reverts to the old behaviour of allowing the
primary key of xml_path_index to be set using its sequence instead of
manually matching it to the xml_nodes.nodeid (which the current code
did and which caused the duplicate key problem). See bug 2469 for
related details regaring the indexing changes.

As a continued fix for http://bugzilla.ecoinformatics.org/show_bug.cgi?id=2469,
I've fixed the indexing implementation in both buildIndex() and
traverseParents(). Duane pointed out that the incorrect parent node ids
were being indexed in xml_path_index, causing some stylesheets to render
metadata incorrectly. I've changed buildindex() to index the correct parent id,
and have changed traverseParents() to only index paths that actually contain
node data (some TEXT nodes exist with empty data).

One more patch for bug #2469:
Although the correct parentid values were being indexed in xml_path_index
for leaf node xpaths, they were still incorrect for relative and absolute
paths. This patch modifies traverseParents() and changes the parent node id
to be indexed to that of the leaf node, no matter if the path is a leaf,
relative, or absolute. Thanks for catching this Duane.

Original Bugzilla ID was 2469