Re: A question for Mr. Celko

"--CELKO--" <jcelko212_at_earthlink.net> wrote in message news:18c7b3c2.0407181042.321c8893_at_posting.google.com...
>
> The classic example for relation-valued columns is to put the table
> inside itself.

Sure, that'll cause boatloads of problems. But it is not the case that having RVAs and having a table nested inside itself are the same thing. It's the fully-recursive part that's the problem, not the nesting. In fact, simple nesting is informationally equivalent to two tables, one with a foreign key.

> Lazurus Long is his own great grandfather, if you are
> a Heinlein fan,

But of course!

> and you fall into infinite recursion. A parts
> explosion suddenly has more assemblies than there are atoms in the
> Universe. I nest a table using the Ackermann function as my model
> (non-primitive recursion), Etc.

All these examples require a fully-recursive table definition. That is, table A has an attribute that is itself table A. Not even just the same type; the same actual relvar.

> In fact, any data structure more complex than a scalar can run into
> Mr. Godel and his friends. I am not bothered by having just simple
> scalar columns and unimaginative tables -- Number theory gets along
> fine with nothing but integers.

Mr. Godel and his friends are our enemies.

Number theory may get along just fine with just integers, but I'm a working programmer; I need better tools than that! 1NF isn't strong enough. Then again, I'm quite concerned about the influence of the evil Dr. Godel and his minions. But I'm not convinced that there isn't some middle ground between lame 1NF/FOL and full on self-referential table definitions and HOL. Although I'm clearly not educated enough to fully understand the implications of Incompleteness.

Marshall Received on Sun Jul 18 2004 - 21:21:56 CEST