Re: Date's First Great Blunder

Using your most recent post, I'm going to quickly describe what I mean by "we're in different chapters". I don't want anyone to think that I'm trying to pull some superiority thing; we simply have different ways of looking at the same things, and our views are so fundamentally different, that it is unlikely that we can explain ourselves to each other.

Neo wrote:

> I can see that operations on data can give us more than 2 values.

From my perspective, predicate logic, and by extension, relational database logic, is not concerned with the result of arbitrary operations on data. It is concerned with predicates, truth statements. We would have to arrive at a common understanding here before continuing.

> We start at 2 values because it is the minimum needed to make
> distinctions and it is the simplest.

From my perspective, 2VL is simpler only in certain contexts. 2VL can actually cause increased complexity at a higher (but not necessarily much higher) level than the base single simple predicate. As I noted, to apply 2VL, you have to disallow many possible predicates. We would have to arrive at a common understanding of "simplicity" before continuing.

> More values should allow a db to model reality better, thus 5VL should
> be better than 4VL, and 6VL should be even better than 5VL, etc.

Maybe. A lot of issues would need to be agreed upon before thoroughly discussing the implications of the selection of a logic system.  

> The core problem is not the number of values derivable from operations
> on data, but when a modelling system trys to represent missing info.

As silly as it may sound, I believe that "missing info" is an arbitrary concept defined by context. This alone is an issue that would take much time to discuss.

> RDM does this with NULLs initially and can subsequently patch them
> with special values. Now the db has data that doesn't match reality
> (ie there is no color in reality that is unknown, this only occurs in
> some modelling systems such as RDM). Operations on such data produces
> values that sometimes don't match reality even if the modelling system
> is 3VL or higher. This is why C. J. Date ends his chapter 20 with
> "NULLs and 3VL undermine the entire foundation of the relational
> model".

As I mentioned before, we apparently have differing definitions for the word "NULL". Confusion over a concept as fundamental as NULL greatly impedes discussion.

Differing assumptions about what is "reality" are serious discussion-killers, too.

There are simply too many differences between our uses of terms and between our base assumptions. As an analogy, I could start posting in Esperanto. Even if what I was saying agreed with what you are saying, it would be hard to tell. So, sadly, as interesting as what we are discussing is, I don't know how we can salvage a productive debate from it. :/

I apologize for starting this thread of discussion only to kill it. I jumped in due to an emotional reaction to the citing of Date's comment, and I probably should have kept silent.

--Senny Received on Sat Apr 24 2004 - 21:06:20 CEST