Identifier Invariance in ODF Documents

["Dennis E. Hamilton" <dennis.hamilton@acm.org>]

This may be tangential to CT, but maybe not.

I will use the current ODF 1.2 specification since I have not followed the intricacies of MCT.  I think the considerations are the same.

In ODF 1.2, there is now increased use of xml:id as an attribute.  It is used, for example, to cross-reference among places where changes have been made to the details about the change in a <text:tracked-changes> element.

There are also uses of identifiers that cross-reference by other than ID and IDREF attribute-value types.  That figures in as well, but the use of ID, IDREF, and xml:id is what I want to focus on.

Another use of xml:id is as a fragment identifier for reference via an HREF from elsewhere in the same XML document (e.g., content.xml) or between two XML documents (e.g., an RDF document and content.xml) in the package.

Because, in general, it is not possible to know what xml:id attributes will be the target of HREFs from other parts of the package, including RDF parts and parts of an unknown character, there is a problem if the xml:id values and references to them are not preserved when the elements involved are carried between the consumption of a document and the production of a derived document (a revision or some repurposing).

The problem is that an ODF processor is not required to be prepared to recognize where all the references into a part by fragment identifier may be, and to update them.

This is a similar kind of problem to ones that Oliver has been pointing out.

So, a processor that is intended to preserve round-trip use in an interchange setting may need to keep existing identifiers invariant so long as the document components that establish them are retained in the processed result.

Whatever is done in change tracking, it should not prevent the preservation of identifiers in this manner.

Note that the preservation of identifiers works consistently when an element is decapitated or completely deleted, its start tag is part of material preserved from in the change-tracking of the deletion and any xml:id in that start tag is going to go wherever that start tag goes.  In ODF 1.2 change tracking, that also means identifiers in change-mark elements will go with those elements when swept up in a deletion.  (There are other elements that cross-cut XML hierarchy that can have their associations severed in this kind of activity too.) 

Of course, processors could preserve referential integrity by dealing with all inbound references and repair them when an  xml:id is emitted in the result of document processing.  But I don't think ODF can required that.  ODF also should not prevent processors maintaining identifier invariance as a simple interoperability assurance.  (This also deals with foreign-element cases, although I don't particularly favor preservation of not-understood foreign elements.)

Something y'all may need to think about.

 - Dennis