Re: So what's null then if it's not nothing?

In article <1134925992.542178.22230_at_z14g2000cwz.googlegroups.com>, boston103_at_hotmail.com says...
>
> > No, the set of truth values does not include NULL, just like the set of
> > integers don't.
>
> How do you define addition over your integers then ? How do you
> interpret 2 + NULL ? *What* is '+' in the expression ?

"2 + NULL" is not a valid expression.

> > NULL is not a value, it is a property of a variable (or
> > an expression).
>
> How do you define any operation with NULL over the domain if integers ?

In short (and informally): null expressions propagate; if nothing is null, everything works as normal.

> > Metadata.
>
> Please provide a definition of the word 'metadata' applicable to NULL
> and explain how relevant such definition is to the operation of integer
> addition.

A variable's value is the data. Information *about* the variable---e.g. whether it's constant, passed by value or reference, its physical storage characteristics, whether it has been assigned a value, or is null, is metadata. And it is not relevant. I don't change integer addition.

> > I disagree with their presentation. To use truth tables to show how null
> > expressions work with boolean operators leads people to believe that
> > NULL is a (truth) value.
>
> How would you define, say, OR when one of the operands is NULL ? What
> exactly is 'OR' in an expression like TRUE OR NULL ?

"TRUE OR NULL" is not a valid expression.

> > And I'm not saying I necessarily agree with their position, but I
> > understand the justification for that position. Given the way NULL works
> > in other contexts in SQL, I think the 'boolean NULL' makes sense.
>
> Please explain what you mean by 'I think the 'boolean NULL' makes
> sense' (provide an example).

I mean that it fits into the pattern of how NULL is treated elsewhere in SQL. Other approaches break NULL propagation, and/or require truth value columns (and expressions) to be NOT NULL.

> > > Also, ok, you forbid expressions like FALSE AND NULL, but what does it
> > > change ? How do you evaluate (1=0) AND (x>y), where x or y happen to
> > > be NULL ?
> >
> > The change is just for making it clear(er) that NULL is not a value.
>
> What is gained at the logical/algebraic level by 'making it clear(er)
> that NULL is not a value' ?

Hopefully, endless discussions such as this one are avoided. As you might remember, the presence of null/w in a truth table was your principal argument against my interpretation of Codd.

-- 
Jon