Re: [UCR] Design constraints: early example goal/csf hierarchy --> PR / PRR

On Apr 23, 2006, at 2:25 PM, Vincent, Paul D wrote:

> Gerd - I don't think comparing the state of standardization in current 
> PR systems with HTML is very useful.

I disagree.

>  The latter was defined as a "standard" from the start,

And companies added tags left right and center from the start. It also 
had a very liberal (vauge even) semantics. Browsers added tags to be 
competitive and distinguish themselves. People relied on browser 
dependent understanding of the meaning of tags. I think of production 
rules as having  a fairly similar history.

>  for distributed (+ public) internet access. A better analogy would be 
> to complain that the semantics of Java/Java byte code and .NET/CLR are 
> different (and we don't want to open that can of worms).

Both of these were single source, primarily defined by implementation 
(though then good descriptions and standard, or semistandard, 
definition). .NET/CLR was also pretty certainly designed to be 
incompatible :)

> The reason for PRR and PR in RIF is to share rules across different 
> systems.

Which is not remotely the purpose of Java byte code and .NET/CLR (at 
least if "different system" is defined as JVMs vs. .NET systems). 
Hence, I think the analogy fails :)

> In time, if the standard is successful, then it may take on some role 
> of leading the semantic definitions for PR engines / use.

Consider Common Lisp, or, indeed, any language standard. If you want 
interoperabilty and portability, you have to accept some compromise. 
The goal is to agree upon the stuff that no one wants to compete on 
anymore, so that the entire market can grow. So, I'd be pretty shocked 
if vendors were completely unwilling to evolve a bit. Of course, 
sensibly, they'll try to evolve as little as they can (while hopefully 
making everyone else evolve more), but that's ok. We just have to 
recognize that absolute fidelity to the quirks will make the task MUCH 
harder and the result less useful. For everyone.

> But there is a long way to go yet!
>
> PS: I would expect all PR vendors to have well defined semantics for 
> their rule engines.

This would be great. I would ask that all PR vendors on the working 
group supply these documents as soon as possible so we can study them. 
Or is there a wiki page for these already?

>  I would not expect RIF to take on the challenge of deciding which 
> particular PR semantic was "correct" - what may be a quirk to one 
> person may be an essential productivity saver to another!

And some people will be, or will feel that they have been, screwed. 
It's unfortunate, but too bad for them. We risk the whole endeavor if 
we try to deal with everything.

Note that there are ways and there are ways. One way is to make some 
quirk *illegal*. Another is to make an area in which quirks abound 
*undefined*. *Both* are problematic. One of our jobs is to decide when 
the tradeoff are worth it. Vendors and customers will give pushback, so 
it's not like we are making these decisions entirely in isolation. In 
fact, it's important to gather data about what people are willing to 
give up, and what the costs are.

Balancing legacy and sanity is a tricky task...but they must be 
*balanced*, not relentlessly adhered to.

Cheers,
Bijan.

Received on Sunday, 23 April 2006 22:42:06 UTC