Re: The Theoretical Foundations of the Relational Model

On 24 Jun 2002 15:30:41 +0200, hidders_at_hcoss.uia.ac.be (Jan.Hidders) wrote:
>> Find me the first person who lives on Elm Street, deterministically, in a
>> relational system. It matters to me. Yes, sure, you can add some
>> additional ordering field if that is a requirement, but if the requirement
>> is very common (and it is), then the statement that order does not matter,
>> was at best misleading.
>
>Nobody said that order isn't important. What was said was that the order of
>propositions in the extension of a predicate doesn't matter. So if the order
>is important for you then you have to encode this somehow in your predicates
>by adding extra roles/columns. That keeps your reasoning within conventional
>logic (even in higher order logics the extension of predicates is unordered)
>and prevents a bias towards a certain way of storing the information. Why
>not keep things simple if there is an easy way to do so?

What we have here is a sign that in doing something with the database, we may be doing something which is not entirely equivalent to predicate logic. In some cases, the "easy" way to do stuff would be to specify a logical or physical order as part of the data model (of views, perhaps of tables). Since order doesn't matter to the predicate logic, this could be allowed without loss. Of course, order is allowed in result sets. Why should a result set be able to have some property (order, nonuniqueness of members) that base tables and relations do not?

>Ah, if that is what you mean then I agree that object identity is still a
>problem in the relational model. It boils down to the requirement that
>values in the domain must have a representation, i.e., we do not allow
>abstract object identifiers. In logic such restrictions are usually not made
>so the relational model has to justify this by itself. The biggest problems
>I see with this are
>
>- representability: if your query result contains object identifiers how are
> these represented in the result
>
>- object id creation: which object creation operations are natural

That's a good statement of the situation.

>I don't see an easy solution to this right now, but I also don't think that
>OOP data models are going to help us much here. Note that to some extent
>the object identity problem is already solved in Entity-Relationship data
>models and Object-Role data models, which are better data models anyway.
>However, these don't really solve the problem of how to represent and
>manipulate data as elegantly as the relational model.

I don't claim to have any definitive answers, either. I'd love to find a seminar on this, see it attacked from different directions over twelve weeks.

Joshua Stern Received on Mon Jun 24 2002 - 17:20:50 CEST