Re: MV Keys

In article <1141239957.788010.239620_at_j33g2000cwa.googlegroups.com>, dawnwolthuis_at_gmail.com says...
> Jon Heggland wrote:
> > Fair enough. Perhaps I should ask Dawn instead. Dawn, what definition of
> > 1NF do you use? Why do *you* think list attributes breaks the RM?
>
> I use it the same way that Codd uses the term "normalize" in his 1970
> paper. He refers to removing "non-simple domains" which is not exactly
> tightly defined, but is used related to "repeating groups."

Yes, he says "repeating groups" and "nonsimple domains" are "roughly analogous", and uses a relational-valued attribute (RVA) as the (sole) example of an attribute with a non-simple domain. But he presents normalisation as something done for convenience. Why do you say that using RVAs or list-valued attributes amounts to throwing away the entire RM? Is this something you've come up with yourself, is it something Codd has said, is it based on current SQL systems not handling RVAs, or is it something else?

> > The others have formal definitions, based on functional
> > dependencies. 1NF is mostly informally defined (except for Date's
> > definition, which however renders 1NF pointless and irrelevant---but
> > then, so is 2NF, and arguably 3NF),
>
> Yes, and this damages the def of all NF's because they are all defined
> first as requiring 1NF.

Have you thought about this, or is this just based on some textbook presenting e.g. 2NF as "1NF plus no partial dependencies"? Have you read Date's articles on RVAs and 1NF? If not, you might be in for a surprise.

> > This definition also has the unfortunate property that a relvar can be
> > in 5NF, but not in 1NF.
>
> Although all NF's are defined as first requiring 1NF, at least with
> Codd and his camps.

Indeed? Many textbooks define BCNF independently of 3NF, 2NF and 1NF--- despite the historical background. What matters is substance, not tradition.

> > 3. "1NF is subsumed in the definition of a relation." Date's position,
> > which means it is really pointless to talk about 1NF in any case. If a
> > "relation" is not in 1NF, it isn't a relation, and normalisation doesn't
> > apply.
>
> I agree with Date tossing out normalization as defined by Codd, but not
> with his attempt to redefine 1NF to be meaningless. It is difficult to
> move foward if people still have the mistaken impression that database
> theory requires that we remove repeating groups.

Date certainly don't. It sounds like you might agree with him more than you think.

-- 
Jon