Re: [opendocument-users] New question (2): Reference implementation?

[Rick Jelliffe <rjelliffe@allette.com.au>]

Cody, John (OFT) wrote:
> (Straightforward question, I hope, and hopefully will not engender off-topic disagreeability):
>
>         -- Does ODF have a reference implementation?
>   
 
Document format standards are really two different standards 
intertwined: a standard for the document format and a standard for some 
kinds of application.

The recent drafts of HTML 5 have taken the step of teasing these apart, 
so that there one standard for the markup language HTML 5 and another 
standard which has more of the semantic, user agent and API aspects.

A document format is application-neutral and therefore cannot really 
have a reference implementation. Or, perhaps better, it should not need 
a reference implementation because all its significant constraints will 
be dealt with by schemas. In the case of HTML 5, they use ISO RELAX NG 
schemas for structural constraints and but they also list as assertions 
(suitable for implementing in ISO Schematron) the kinds of extra rules 
that go beyond what is convenient or possible in RELAX NG. 

If we can exclude user agent semantics (how an application acts) then 
the second kind of standard is based on linking from the markup to an 
abstract model. The schema tells us that a <table> must have <row> 
elements, but the semantics say that <table> is a table and a <row> is a 
row.

In most cases, the semantic constraints (tables must have rows) are not 
necessary to model, because the schema looks after that. But for 
semantics that relates to rendition of information in an 
application-neutral manner, there has been an unwillingness to provide 
automated verification. (For example, that one row starts after the 
previous on a page can be made into constraints that can be tested on 
the PDF positions of the contents.) Until the constraints being 
specified on renditions are put into a form susceptible for automated 
testing of implementations, conformance only has the weak method of 
checking by eye, using test suites, or deferring to the most popular 
application (which must swing in Microsoft's favour as it becomes the 
dominant provider of ODF applications. )

For other standards bodies, I have been often told that JTC1 does not 
allow them, because the standard has to speak for itself. And in W3C 
there is an idea that every major feature has to have had an 
implementation, to prove the practicality and implementability of the 
feature, to move the specification to full Recommendation: but this does 
not require any application has implemented all parts, nor do the 
implementations become references that augment the standard in any way.

The other issue with reference implementations is that there are two 
kinds of standards: ones which codify a dominant product to open up the 
market and ones which consolidate related technologies to promote a new 
market.  In the first case, there is a de facto reference implementation 
that everyone can look at (e.g. MS Office for IS29500 OOXML and Open 
Office for IS26300 ODF); but you might be wary about treating them as 
official reference implementations because you want to ameliorate their 
market domination not entrench it further. In the second case 
(consolidation),  very often a simple open source reference 
implementation is prepared as part of the implementation process as a 
co-operative effort, but without the bells and whistles that a 
full-fledged application would have.

Cheers
Rick Jelliffe