Re: The MySQL/PHP pair

Dan:

"Dan" <guntermann_at_verizon.com> wrote in message news:bjOid.12$6t.4_at_trnddc05...
> "A domain is <italics>simple</italics> if all of its
> values are atomic (nondecomposable by the database management system)."
>
> I don't think this could be any clearer in terms of what Codd said in this
> particular instance. I've always associated this with the definition of
1NF
> as the starting point for normal forms, as a natural consequence of the
> mathematical definition of a relation, but I'll have to admit the linkage
is
> not explicit and might not be self-evident for some, especially if
> literalism is important. If the repeating groups are logically
> decomposable, they must be eliminated. If they are not, then they don't
> have to be. If we assume, which I really don't think we need to do since
it
> seems obvious to me, that a relation is in 1NF by definition, then there
is
> a mathematical reason why a repeating group (your old definition of 1NF)
> must be eliminated.

I think this is the crux of the debate. 1NF cannot properly model an audit system. It can model parts of it but not the fundamental requirement of it unless delusional assumptions are made (that nothing can go wrong and we can foresee everything).

We know this. We make assumptions and adjustments to try to account for this. We joke about auditors and accountants. But it is, nonetheless, true!

So we move beyond 1NF to try to define a model that accounts for the vagaries of a particular reality. This doesn't mean 1NF is stupid. It doesn't mean those who use it as part of their modeling tool chest are wrong. It simply means the limited (by design) model doesn't adequately describe a particular reality. As far as I know, every theory suffers, to one degree or another, from this limitation.

WOL talked about this previously when he was passionately describing the need to measure theory against reality. I don't understand Dawn to be debating the specifics of 1NF, just that it isn't complete and, therefore, cannot be used as the sole model for software development.

> I would say that there is nothing about modeling data logically that
> requires 1NF. But by definition, a relation is in 1NF. I know we've gone
> through the paces on this before, but you keep going back. Why not just
> leave the "Relation" and "First Normal Form" alone and start with
something
> clean, fresh, and new. I think Laconic2 mentioned something like "Dawn'
> Normal Forms".

This is, of course, a good suggestion. I've noticed that when a fresh approach happens to be mentioned, a bevy of complaints arise that these new solutions don't meet the "understood by all to be the mathematically perfect" 1NF model definitions. In addition, these definitions seem to be quite fluid and their imperfections downplayed.

So, it appears we keep "going through the paces", again. :-)

Bill Received on Sat Nov 06 2004 - 19:28:14 CET