A few thoughts on ODF change tracking

[robert_weir@us.ibm.com]

A few thoughts on change tracking.

What would make this feature "complete"?  I think change tracking is 
complete if an application can record any change to the end-user semantics 
of the document.  This includes things like changing the visible content 
of the document, but also changing the metadata.  It does not include 
changes that have no end-user semantic value, e.g., changing an xml:id or 
automatic style name to a different value, provided the referential 
integrity is maintained. 

I should be able to take two arbitrary ODF documents of the same type, say 
text documents A and B, generate a diff of them and encoded that diff of A 
to B as change tracks on A, such that when I accept all tracked changes in 
A I get something semantically equivalent to B.

Given that as a goal, I think we need to get it out of our heads that any 
standard or implementation does change tracking completely today. 
Everything I can find has some notable areas of missing functionality.

Aside from the issues already noted in JIRA, consider:

- It is possible for a named style to change its definition, without 
changing anything in content.xml.  For example, I could have a style 
called "hidden" that is defined as white text on a white background, and 
then unhide it later by merely redefining the definition in styles.xml. 
How is that tracked?  MS Word does this, and in the UI appears to hint 
that it is a change to the document as-a-whole rather than any particular 
content.  This seems reasonable.

- It is possible to replace an image file with an identically named image 
file but with different content.  How is that tracked? 

- Similarly, for any object embedding, OLE or whatever, how are changes 
tracked? (MS Word doesn't seem to track these changes)

- And what about MathML, a special kind of object in ODF?

- And what about sub-documents?

- Or RDF XML metadata?

- Or Changes to the manifest, such as adding a new resource, not 
necessarily references from content.xml

- And what about presentation files, which appears to lack support even in 
 OOXML? 

So change tracking, if you want to be complete, needs to deal with changes 
at the packaging level, in all of the defined XML's, as well as changes in 
linked resources, such as jpeg images and RDF XML.  As far as I can tell, 
no application and no standard does complete change tracking.  So I 
suggest we don't take it as a last-minute goal for ODF 1.2 to achieve 
completeness there. 

However, inability to do everything is not an excuse not to do nothing. If 
we can extend the current change tracking approach to some of the more 
commonly-reported pain points, like change tracking if table row 
deletions, etc., then this would be good.  But I'd propose that before we 
move to a new and incompatible representation of change tracking that we 
sit down and work through the entire semantic content of an ODF document 
and ensure that we're using an approach that is capable of recording 
changes across the entire range of possible document edits.  This is 
important, and worth doing, but worth doing right.  The proposal submitted 
via the comment list appears to be an excellent start, but we need to 
expand it to cover some of these other use cases, if we want to claim 
completeness. 

Regards,

-Rob