Re: RA with MV attributes

paul c wrote:
> David wrote:
> > paul c wrote:
> >
> >>David wrote:
> >>
> >>
> >>>...
> >>>Why can't we say that *by definition* the set of attributes of a
> >>>relation is part of its state not its type?
> >>>...
> >>
> >>We can say anything we want "by definition", the question is where does
> >>saying it get us? What conclusions does it lead to and so forth.
> >
> >
> > Marshall doesn't seem prepared to accept this definition on it face,
> > because he has assumed that the set of attributes of a relation is
> > surely its type.
> >
> >
> >>I
> >>like your suggestions because they seem to have something behind them,
> >>but don't ask me what, exactly. I wish you would distinguish whether
> >>you are talking about physical implementations or user concepts.
> >
> >
> > I guess this relates to the following discussion
> >
> > http://www.ai.mit.edu/docs/articles/good-news/subsection3.2.1.html
> >
> > I tend to think that both implementation and interface should be
> > reasonably simple, and it's bad to strongly favor one at the expense of
> > the other. I often look at what the implementation is telling me in
> > order to design a reasonable interface. There is nothing better than
> > getting your cake and eating it too.
> >
>
> That's a bit too blurry for me. What is simple as compared to what is
> too simple depends very much on the objective, as I think somebody said
> "as simple as possible, but no simpler!".

IIRC that was Einstein

> I do think that once one has a definition it is important to proceed to
> positioning it in the scheme of things, for example is this a physical
> approach for storing two relations in one or is it an attempt to promote
> three-valued logic or is it something else?

I'm interested in both its logical and physical aspects. It seems best initially to focus mainly on the logical.

I wouldn't call it a 3vl.

> I find this thread
> interesting because I think rva's have not been explored very thoroughly
> (perhaps they should be discounted eventually, but I don't think anybody
> has yet made a good case, or at least one I could understand, for this,
> yet).

What is rva? Did you mean mva?

> The "advantage" of avoiding redundancy in general is also a blurred one
> for me. If the domain we're interested in is what cars certain people
> own, I don't find it all redundant that two people may happen to own the
> same kind of car. Whereas avoiding physical redundancy as far as
> computer machinery is concerned is a very worthy goal.

It seems worth thinking about redundancy from the logical point of view - in terms of what it means for updates and schema changes. There are pros and cons of MV attributes that require analysis.

For example, adding an MV attribute to an existing relation is trivial - if every existing tuple maps the new attribute to the empty set then the addition of the attribute hasn't changed the information in the DB and it won't affect any existing queries. Received on Thu Jan 18 2007 - 06:28:47 CET