Re: Why not XHTML+RDF? was Re: Links are links

Jonathan Borden wrote:
> Paul Prescod wrote:
> 
>>I want my cake and to eat it to and you haven't offered a technical
>>reason why this is impossible.
> 
> 
> 
> Sure, the standard answer seems to be: use XSLT to convert your syntax into
> something standard.

Despite its many virtues, XSLT is not appropriate for this job. It is 
too heavy, inefficient and unreliable in its performance. Most damning, 
it does not preserve node identity which is a crucial issue for both 
XLink and RDF.

All we are talking about is renaming a few elements and attributes, 
maybe synthesizing or hiding one from time to time. This is trivials 
stuff that does not require a Turing-complete language like XSLT. If you 
were given the problem of mapping between two RDF vocabularies would you 
use XSLT or something more declarative like daml:equivalentTo or 
daml:subProperty.

It wasn't considered rocket science when it was made a requirement of 
the XLink WG years ago and could probably be *simpler* than the various 
rules that attempt to make RDF's XML syntax readable.

 > No doubt. I wonder why Hytime never took off.

It was about five standards in one. And although the instance syntax was 
good, all of the declarative stuff was shoehorned into FIXED attributes 
in DTDs. That gets ugly. But really it was the former that was the 
bigger problem.

 > Was it too complicated?

Yes.

 > Hard to implement?

Yes.

 > Should we reconsider decisions made in 1992-1996 in light of
 > stuff available today?

Yes.

 > Yet even if we reconsider, can Hytime gain the mindset of web folks c.
 >  2002?

No, for the same reasons it did not the first time around. I'm not 
suggesting a HyTime revival. I'm just pointing out that HyTime solved a 
problem so therefore the problem is soluable. You can have both 
interoperability and whatever darn linking syntax you want *at the same 
time* without a Turing-complete transformation language.

> Similarly RDF allows you to get to the semantics (to the extent that this
> describes a 'graph' or internal model) easily without worrying about the
> details of elements, attributes etc. etc.

People like to worry about those details. Otherwise s-expressions would 
have taken off years ago. But syntax matters.

> On the other hand when its _close_ to
> either XLink or RDF, then dealing with a namespaced attribute or a few child
> elements seems not a huge price to pay assuming that generalized software
> exists that can do interesting things with either RDF or XLink.

Seems to you. But I've been hearing Lisp programmers say similar things 
for years. And look at where that got Lisp. I hope for more success for 
RDF and XLink. Syntax matters. Names of things matter.

  Paul Prescod

Received on Thursday, 3 October 2002 03:37:25 UTC