Re: domain questionnaire

JRStern wrote:
>
> Well, Codd has mostly been quite, but Date has mostly been noisy,
> over the past ten years or so, ranting about how objects were evil,
> to make a long story short. And, frankly, he had many valid points.
> IMHO. In the last year or two, Date has been trying to make peace
> with objects, and trying to ignore or disown his own earlier
> statements. No, I don't have any specific pointers or citations
> handy.

Not neccessary, I know them and I completely agree with you. It's nice to hear such a balanced opinion. All too often people are in either camp A or B and think that the other side does not really understand what is going on.

> The one point that I think the object people got right (by accident,
> perhaps) all along, and that Date could not accept because it
> violates the basic dogma of relational theory, is that an object
> (read: attribute or tuple) might have an identity separate from its
> value. But I do not think that either side, object or relational,
> has yet described a theoretically-based way to merge object identity
> and relational algebra.

Actually Codd has done that in the "second version" of the relational model: RM/T. Date discusses it in his latest book. See:

  http://www.aw.com/product/0,2627,0201547325,00.html

But for "higher level" data models such as ORM there are also good formal descriptions of their semantics plus a good philosophy (inherited from NIAM) about why and how it models reality.

> Which is to say, that the issue of whether a database should reflect
> an external ontology or not (and if so, how, and by what theory), I
> think maps directly to this issue of how object identity must be
> handled, ...

I agree. I think it is reasonable to assume that if the database reflects an external ontology then every object should be somehow identifiable. (If you cannot identify the things you are talking about, then how do you know what you are talking about?) What is not so obvious is that this should neccessarily be done by means of the attributes of the object. I think it is quite acceptable to have objects that are identified by a combination of some attributes and relationships.

But if you are using the database to store objects as occurr in your OO program, then even that identifiability assumption may be too strong. Or is it? I am still not really sure about that.

-- 
  Jan Hidders