Re: more closed-world chatter

On 5 mai, 02:53, paul c <toledobythe..._at_oohay.ac> wrote:
> Cimode wrote:
>
> ...
>
> > paul
> > I am a worried that the question you have raised brings quickly to the
> > computing model involved in the interval based representation of sets
> > and away from the logical underlying issues.
>
> Sorry and no offence but I'm not sure what this sentence means.
> "interval based representation of sets" sounds like it might be a
> well-known term, but I'm ignorant of it. Do you think that's what I
> meant? If so, I had no idea that's what I was talking about.
I apologize for being so vague (I was in a hurry when I caught up the thread).

It seems to me that the spirit of your question leads to the computing model that actually allows the representation of sets. What I mean is that the arbitration of whether exceptions should or shoud not occur is tightly bound to the chronological availability of the information to the db engine of set disjointness/non disjointness. As a consequence, I believe the arbitration is more related to the computing model that would allow the implementation of a TRDBMS, rather than the abstract aspect of RM.

Maybe an example of what I mean would help. For the last 10 years, I have been working onto an experimental dbms core that would truly separate the physical and logical layers. In this long term work, one of the fundamental questions that have arised is: how do I represent the characteristics of a set as opposed to another set ? In other words, if one considers set R1 and set R2, what would be the most efficient way of establishing *declaratively* whether the two sets are either disjoint and, if they are not, under what conditions they would be joined. What I came up with is that such information could be made available to the engine if:

> The data layer (domain values) is represented by a serie of N-intervals having each lower and upper logical bounds.
> There should be a mathematical dependency between the logical intervals (therefore the lower and upper bound) of one set and another set.
> The representation of the data layer must meet the criteria of estabilshing domain characterization *declaratively*

If the three above principles are met, then one can safely state that some of the characteritics of the relationship between two sets could be available to the core db engine as compile time depending. In more simplyistic terms, when doing a JOIN between R1 and R2, the information of disjointness should be established before even attempting to execute the query. The TRDBMS would have some kind of internal function that would allow it to preestablish the information of disjointness. In the case of R1 /\ R2 to be ran on the core:

--> If data layer of R1 and R2 is represented by *smart intervals* that would allow the implicit inference by the core that they are indeed disjoint.
--> if disjointness is *pre-established* by a mathematical function of type f(R1, R2) = NONE)

--> then there is no need to go any further and execute anything. therefore no real exception and the system truly have represented the information *declaratively*.

So far I have programmed succeded to get the system to let me know that I would be having no results at the moment I would type the R1 /\ R2 in case both would be disjoint. (But I am still struggling with the opposite case).

I sincerely hope this makes sense (online exchange and my limited english proficiency is quite frustrating). Received on Sat May 05 2007 - 09:36:35 CEST