dawn

Could you point me to proof that something I expound is incorrect?
Perhaps just starting by telling me what I think to be true that you
think false, or vice versa, would be helpful. I would very much
appreciate knowing.

Are those who develop the products whose data model I prefer, such as
IBM U2, Intersystems Cache', jBASE, Revelation, OpenQM..., all also
incorrect in their understanding? What about the IBM DB2 9 developers?
Do they have the data model right?

I started with two issues I wanted to investigate, where all of those
using "relational databases" modeled data differently than those of us
with "legacy databases." I had worked with both approaches and was
coming to the conclusion that the old stuff did some things a lot
better than the new (although the new tools were all swoopier with
typically more features). 1) 1NF vs NF2 and 2) NULL handling, 3VL vs
2VL. Now that I see that the state of the research in the RM has many
people agreeing that relation-valued attributes are acceptable and 2VL
is preferred, I figured I was more in the majority than I had
previously known, although most SQL-DBMS products are not yet up to
speed.

I added in the issue of lists. Although RVAs are accepted by some,
lists are not yet as accepted in the theory other than treating them as
user-defined types (not incorporated into query languages, as best I
can tell). I'm still in the minority on that one, it seems, but I'm
guessing that the RM will catch up with MV on that one at some point
too.

While it is now clear to me that NF2, 2VL, and lists are desirable
(when I started asking about them, I thought I would be convinced to
abandon them by listening to the theory experts), issues about strong
vs duck vs weak typing are less clear to me, along with metadata
specifications and constraints in general. I lean toward the DBMS
doing minimal typing, with services that maintain the DBMS handling
constraints. I'm guessing I'm still in the minority on that one, but
surely this is not an argument where I stand alone as there have been
discussions about typing with every computer language. Java and
JavaScript each have their charm and their problems in this area.


I have changed my mind on a lot of things since posting here
originally. But so far you are right that I haven't seen the proof
that I should want to move from NF2 to 1NF (or the
form-formerly-known-as-1NF) nor from 2VL to 3VL. It even appears that
the industry is headed in my direction, don't you think?

Is there something you KNOW to be true and can prove that you think
that I do not believe to be true? If so, I would be happy to read
about it. I would prefer that I not be missing something that is
perfectly clear to everyone else. Thanks Walt. --dawn

walt

this ground has already been covered by others. I see no need to cover it
again.

neo

I may have missed the explanation above, if initially john has no cars,
tom has two cars, and bob has one; how does the MV solution later
handle each car having 0 to many manfacturers?

Because dbd's data model is very general, systematic, normalized, holy,
etc; one simply adds those attributes like this:

(new gm 'mfg)
(new honda 'mfg)

(set (. 'aston 'martin) mfgdBy honda)
(set (. 'aston 'martin) mfgdBy gm)