Re: NULLs: theoretical problems?
Date: Sat, 11 Aug 2007 10:19:31 -0700
On 11 août, 15:39, "David Cressey" <cresse..._at_verizon.net> wrote:
> "David Portas" <REMOVE_BEFORE_REPLYING_dpor..._at_acm.org> wrote in message
> > On 11 Aug, 12:20, Jan Hidders <hidd..._at_gmail.com> wrote:
> > > Heck, I could even explain it in Dutch for you. ;-)
> > > But also then it wouldn't make sense. Any argumentation that is based
> > > on taking the definitions in standard normalization theory and
> > > applying them to relations with null values is based on a fallacy. If
> > > you really want a meaningful discussion you first need to come up with
> > > a proper formal definition of a relation with null values, update
> > > anomalies, redundancy, dependencies, et cetera. That, by the way, is
> > > actually not as hard as you might expect, but it still needs to be
> > > done.
> > > -- Jan Hidders
> > Jan, I agree with you. I don't know of any such formal definition that
> > deals with nulls. What I was trying (and probably failing) to do was
> > to show that by the present definitions we cannot find nulls to be
> > acceptable in dependencies or in normalized relations. I was trying to
> > do so without commenting on whether nulls can be permitted in
> > relations at all (but naturally I agree that they can't).
> It seems to me, after looking at it for a couple of decades, that nulls
> cannot be permitted in relations at all, but that they can be permitted in
> tables that represent relations.
For the latest part, I would rather say that they can *subjectively* be permitted in
tables that represent relations (each technology having its own proprietary definition of NULL). The problem of NULL acceptance is the *absence* of theory.
> SQL tables are approximately tables that represent relations. The mapping
> isn't perfect, so some of the formal properties of relations do not carry
> over to SQL tables. This has been the source of much debate and friction in
> this newsgroup and in other forums. Much of that debate and friction is
> unnecessary. Much ado about nothing.
We unfortunately can *not* ignore nonsense as it bites in the arse in a daily basis.
> The renaming of nulls to be a "special value" is, as far as I'm concerned,
> a deft side-step rather than a fundamental resolution. To me, the
> fundamental problem that nulls address is about the relationship between
> representations and the reality they purport to represent. This
> relationship between representation and reality is complex and subtle, and
> pervades almost all of computer science, not just database theory.
Agreed. Received on Sat Aug 11 2007 - 19:19:31 CEST