Re: [BLD] My comments on version dated 5/10

There is only ONE way to serialize each particular rule.
However, there are logical equivalences, so quite different rules may mean
the same. There is little we can do about it, because these are the laws of
nature. Forcing the implementers to recognize even a small number of such
equivalences is a HUGE burden for the more expressive languages. However,
this simplifies the task for the less expressive languages (provided they
support just the right kinds of rules!).

So, it is a matter of who bears the burden.  I strongly believe that
requiring normal forms for the rules (through logical equivalences) is
counterproductive.


	--michael  

> Christian de Sainte Marie wrote:
> > 
> > Michael Kifer wrote:
> > 
> >>> Am I the only one in the group to think that normal forms can make 
> >>> the life of implementors easier?
> >>
> >> Easier? They'll make the life harder! Instead of a straightforward
> >> translation they will force the implementors to recognize non-normal 
> >> forms
> >> and do translation.
> > 
> > Hmmm... There might be a misunderstanding, here: what I say is that, if 
> > RIF specifies a normal form, we can have a level of compliance where 
> > only the recognition of normal forms is required. That might make 
> > implementation easier.
> > 
> > But I agree that, as I already pointed in my previous emails, the 
> > downside is that it would probably require such implementations to 
> > publish only normal forms, which might make implementation more complex. 
> > Although most actual rule languages have probably more expressive power 
> > than RIF basics, and will have to do some recognition and transformation 
> > anyway; but, well...
> > 
> > And so, I have no decisive opinion either way: I was just wondering if 
> > this is something we should discuss.
> > 
> > But I do not seem to get much echo, which probably means that the answer 
> > is: no, this is not something we should bother about. And that we should 
> > probably just end the subject here.
> > 
> > If somebody disagrees and thinks that we should raise an issue; please 
> > speak up now or shut up forever! :-)
> 
> Sorry, I haven't followed what the specific test cases are that you are 
> talking about.
> 
> In general having a single normal form is a substantial simplification 
> for implementers. The fact that the same set of triples can be 
> serialized so many different ways in RDF/XML is a barrier to uptake. In 
> practice it means that translators have to support multiple "normal 
> forms" for output too. It is also a significant support headache (a 
> rather more significant cost than software implementation).
> 
> So in general I would advocate having a small number of normal forms. 
> However, not having followed the details I don't feel strongly enough to 
> raise it as an issue.
> 
> Dave
> -- 
> Hewlett-Packard Limited registered Office:
> Cain Road, Bracknell, Berks RG12 1HN
> Registered No: 690597 England
> 
>

Received on Friday, 12 October 2007 17:02:28 UTC