Re: An object-oriented network DBMS from relational DBMS point of view

From: Bob Badour <bbadour_at_pei.sympatico.ca>
Date: Mon, 12 Mar 2007 20:40:50 GMT
Message-ID: <mViJh.9596$PV3.94226@ursa-nb00s0.nbnet.nb.ca>

Marshall wrote:

> On Mar 12, 11:01 am, "Dmitry Shuklin" <shuk..._at_bk.ru> wrote:
>

>>Hello
>>
>>>Every relation, by definition, has a set of attributes
>>>whose values are unique, and which can therefore
>>>be used in relational expressions to uniquely identify
>>>a row. That's all you'd get out of a pointer. Further,
>>>by requiring the unnecessary pointer, you constrain
>>>implementations unnecessarily.
>>
>>But when I lose pointers I lose instance identity.

>
> Yes. One of the biggest advantages of excluding pointers
> from the model is that you lose instance identity, and all the
> complexity that comes with it. This is particularly desirable
> in a distributed context.

Instances of what, though? That is the question.

>>So It is impossible
>>or very dificult for some table to be a self row. It is not a deal for
>>me.
>>Also I lose possibility for some row to be contained by many tables.

Huh? What the hell does he suppose views do?

> These are just further consequences of the lack of instance identity.
>
> Also, I note that the things you're describing are features,
> independent
> of any problem context. Many problems admit to many different
> techniques
> for solving them. Simply heaping together features from OOPLs and
> the relational model will not get you good design. There are
> advantages
> in learning how best to use each techinque on its own merits. Once
> you have mastered that, you are in a much better position to consider
> how two models might be integrated, or even whether that would be
> a good idea.

One has to start at a very basic level asking why a feature exists in the first place. Received on Mon Mar 12 2007 - 15:40:50 CDT