Re: A few lessons I have learned (June, '03)

In a message dated 6/6/2003 12:47:14 PM US Mountain Standard Time, 
costello@mitre.org writes:

> 3. Minimize exacting requirements on the *form* of instance documents. 
> Expect diversity of expression. 
> 
>   Corollary: In designing schemas apply liberal quantities 
>         of <any> and <all>; minimize use of <sequence>
>         and minOccurs="1".
> 

I strongly disagree with this principle as it reduces the ability to 
robustly validate documents.  The principle you are expressing is 
useful when the instance documents are subject to change.  That is
not the case for all instance documents in all vertical domains.
Thus the principle is only valid on one side of the change spectrum and 
cannot be considered a universal principle.  

Thus, the issue is really to understand what type of data modeling you
are performing -- specifically, contextual modeling versus non-contextual
modeling.  Contextual modeling is where the document name, structure,
and order convey the context of the information in the document.  
Non-contextual
modeling is where the context is built up by an understanding of the atomic
statements (not fixed).  I discuss this in detail in our new book; however, 
suffice it to say
that RDF is well suited to non-contextual modeling as is your principle 
above.  There are many domains where contextual modeling is more
appropriate.  The issue boils down to whether it is better to fix the context 
or
evolve the context.

 - Mike
---------------------------------------------------
Michael C. Daconta
Chief Scientist, APG, McDonald Bradley, Inc.
www.daconta.net