Re: Why is "group by" obligatory in SQL?

From: paul c <>
Date: Fri, 24 Jul 2009 16:51:20 GMT
Message-ID: <cElam.37969$PH1.11087_at_edtnps82>

Cimode wrote:
> Snipped..

>> I suspect that what a 'TRDBMS' is, isn't yet fully known.  Lots of open
>> questions, here're just a few:
>> - Codd espoused logical data independence, yet by assuming some
>> relations couldn't be updated in certain ways, he allowed a kind of
>> contradiction, or at least a kind of dead-end.

> That is unclear. Unfortunately we can't ask Codd for that.
> ...

Right, it becomes our problem It is very clear that he allowed 'inserts' wrt some relations but not 'deletes', and vice-versa for other relations. That is a big loophole if you ask me, like when you ask one of the locals for directions to the next town and he answers "you can't get there from here". Contradicts what Codd called relational closure.

>> - there is normalization theory to do with avoiding redundancy, but
>> nothing comparable to do with avoiding ambiguity.

> Ambiguity is subjective.
> ...

I would say not. Assuming there are good reasons for logical data independence, then examination of the possibilities might lead one to assume that POOD is a logical consequence of data indendence. Whereas I think most people would say POOD is a principle, not a consequence which to me seems a reaction that is closer to laziness than subjectivity.

>> - where is the constraint theory?  will normalization turn out to be
>> just a small part of this?

> I would not state exactly that way. I know for a fact that solving
> the problem of constraint specialization representation simplifies
> normalization up to a point where it is not normalization anymore, at
> least not in the traditional sense of a cumbersome process.

I would say that the 'constraint problem' hasn't even been defined. Date took some baby steps by trying to classify different kinds of constraints. But sometimes such baby steps obscure the forest. Normalization is very much about information structure and I think it would be more practical to try to formalize the scope of constraints. If that could be done, then perhaps the constant question of 'where' constraints should be applied, eg., application or dbms would be clearer. (I don't believe the question is as obvious as many RT advocates suggest, there is more to it than trying to make sure the dbms is authoritative.)

> I have been working on that for a full decade and made fundamental
> discoveries that not only applies to ra but also to computing. Just
> no time to publish a book about it.

>> - is the ultimate interface a program that it is in effect a relation?

> Precisely. My point is that *only* a computing model can bring a
> response to such question and that such question is orthogonal to RM.
Received on Fri Jul 24 2009 - 18:51:20 CEST

Original text of this message