Evolving ODF, Interop and Multiple Implementations

[robert_weir@us.ibm.com]

I want to toss in this idea, for a possible
criterion for how we evolve ODF into ODF 1.3.  

As we know ODF has powerful extensibility
mechanisms via foreign attributes and elements, in application-defined
namespaces.  The ODF 1.2 specification has extended conformance clauses
to handle these cases.  So any implementation should currently be
able to extend ODF and still claim conformance, so long as they have the
basics (the non-extended parts) in conformance.

This leads to the question:  As
we evolve ODF, when should we decide to add new capabilities to the specification
versus leaving them as implementation-defined extensions?

I'd like to suggest that we do this
based on the existence of *multiple implementations* of the new capability,
or at least of committed plans of multiple implementers to implement the
new capability.

In other words, I'd like to discourage
adding capabilities to ODF 1.3 unless it promotes interoperability.  If
only a single implementation will support a proposed feature, then existing
extensibility mechanisms should be used.

Thoughts?

-Rob