Date: Sat, 5 Jan 2008 13:57:51 -0800 (PST)
On Jan 5, 1:25 am, "Brian Selzer" <br..._at_selzer-software.com> wrote:
> "Marshall" <marshall.spi..._at_gmail.com> wrote in message
> > The connection between the model and what is being modeled is
> > only in our head. We can consider the model without considering
> > what the state of the real world is. Indeed, we can have a model
> > that doesn't even *have* a corresponding real-world aspect. So
> > when we talk about "missing" information, that's an attribute of
> > the map between the model and reality in our head. It's not
> > an aspect or an attribute of the model. At all.
> I disagree. With nulls there is an explicit indication that there is
> information that exists but hasn't been supplied. It is not a matter of
> interpretation. Without nulls, there is no explicit indication so it
> becomes a matter of interpretation.
The interpretation you're supplying may well be the one *intended* by the designers of SQL, and certainly very frequently used, but it definitely is *not* the only one possible. Furthermore we can use nulls without *any* interpretation. All we need is an operational semantics.
Here's a relation:
Does the tuple with the b value being NULL specify that the b value for this tuple is missing, or does it specify inapplicable? Or does it specify that opening Christmas presents is nifty? What does b mean anyway? These questions are only sensible relative to a particular interpretation, and I haven't supplied one. And in addition to the option of supplying zero interpretations, I also have the possibility of supplying three, or eleven interpretations, each with different answers to the above questions.
Your assertion that NULL is an explicit indication of xyz is exactly *your interpretation* of what NULL is. It is not an attribute possessed intrinsically by NULL.
> > If we have a set A, and for each member of A we have either
> > zero or one members of set B, then we can do that in a system
> > without nulls, or in a system with nulls. The question is, which
> > way is better? The answer is, the way without nulls is better.
> You're oversimplifying: If we have a set A, and for each member of A we
> /can/ have either zero or one members of set B, and for each case where we
> /can/ have one member of set B, a member /may/ have been specified.
That is not what I specified. You are of course free to specify some other thing for the purposes of example, but specifying some other thing doesn't make my thing go away.
> And yes, we can do that in a system without nulls, or in a
> system with nulls. Which way is better? I'm not ready to
> dismiss nulls simply because it is politic to do so.
How fortunate then that no one wants you to.
Marshall Received on Sat Jan 05 2008 - 22:57:51 CET