Re: Does Codd's view of a relational database differ from that ofDate&Darwin?[M.Gittens]

From: Alexandr Savinov <>
Date: Thu, 16 Jun 2005 15:47:09 +0200
Message-ID: <42b182de$>

Jon Heggland schrieb:
> In article <42b16163$>, says...

>>Jan Hidders schrieb:
>>>For a very short explanation and a link to Jeffrey Ullman's sheets:
>>I find the problems raised in UR model very similar to those motivating 
>>COM. However, the provided solution is hardly acceptable (it is 
>>typcially relational). In COM we have advantages of UR model but the 
>>solution is based on other principles.
>>In particular, the query
>>select office_id
>>from Offices
>>where = 'sally'


> I looked at Ullman's slides and found a strange thing. I quote:
> Suppose we have relations ED, EO, EP, and DM,
> connecting employees to departments, phones, and
> offices, respectively, and departments to managers.
> [...]
> Consider a query "find the offices of employees
> managed by Sally."
> (End quote)

What is strage here? It is precisely an example that I used.

> Note that the relations do not say that *employees* have managers, just
> that *departments* do. Is it the case that an employee in a given
> department is always managed by the manager of that department? It is
> possible, even probable, but the database (as presented) does not say!
> Thus, the query is really underspecified (or invalid, even), and the
> presented solution is based on an assumption that may be wrong. Just an
> observation, but I think it is worth noting.

This problem is obvious for me. In this case we have a kind of zigzag query (we move up and down in the hierarchy several times). Normal inference is where we propagate constraints down (to the universal relation) and then aggregate it up to the target.

There is also another problem. Sometimes there are alternative (not unique) paths along which we can propagate constraints. In this case the database also cannot help but it is not its fault. The model simply does not define what path to choose (path is used in terms of COM).

In UR it is a serious drawback bcause it is based on relational model (its spirit and fundamental assumptions), it is actually a complement to the RM. We need to look at the data differently in order to solve the problem.

Received on Thu Jun 16 2005 - 15:47:09 CEST

Original text of this message