Re: Can we solve this -- NFNF and non-1NF at Loggerheads

"Alan" <not.me_at_rcn.com> writes:

>> No, when he said "atomic", he meant something else: he meant that the
>> objects in question were atomic _with respect to the theory_. That
>> is, his relational theory was not concerned with what those objects
>> actually were; as far as the theory went, they were atomic. This does
>> not mean that they have to be scalar values; they can be _anything_.
>> However, if the value of an attribute is, say, a relation, there is
>> nothing in Codd's theory that looks into that relation, and lets its
>> contents affect operations on the outer, "real", relations.
>
> I refer you to the Elmasri/Navathe text, which states quite the opposite.
> Paraphrasing it (as I don't have it with me at the moment), 1NF means that
> all values are atomic (simple and indivisible - are terms used to explain
> "atomic" in the text), and any value must be a single value from the domain
> of that attribute. It does not allow nested data. It does not allow
> multi-valued or composite attributes. These constraints on 1NF are
> specifically stated in the text. Dawn has a copy. Perhaps she can look it up
> (look up First Normal Form).

As far as I can tell, the two paragraphs I'm quoting here are not at all in disagreement. In 1NF, attribute values are atomic with respect to the theory, meaning that their values have no meaning within the relational theory. What those values are is irrelevant, as long as it is possible to assign them, compare them for equality, order them for sorting, &c. Whether the domain of an attribute is given as "positive integers", "character strings", or "blarglefnorts", is not interesting.

Quick check of agreement: if I have a column for telephone numbers, and store them as text strings like '555-1234', '(201) 555-1234' &c, because that's what my application wants them to look like, you wouldn't therefore claim that my table fails to meet 1NF requirements, would you? My telephone numbers are _not_ atomic: there's an area code encoded within them, and that area code is even optional! :-)

Of course, I fully agree with you that the moment we expand our relational model with operators that peek into attribute values, treating them as composite objects equivalent to rows, or even tables, we can no longer claim that tables containing such attributes satisfy the classic definition of 1NF. At this point, new definitions of the normal forms are needed, that take the new features into consideration.

It seems intuitively obvious to me, though, that this can be done without breaking the underlying relational theory.

-tih

-- 
Don't ascribe to stupidity what can be adequately explained by ignorance.