Re: How to normalize this?

Op maandag 6 mei 2013 17:18:38 UTC+2 schreef Jan Hidders het volgende:
> On 2013-05-06 12:26:10 +0000, Erwin said:
>
>
>
> > Op maandag 6 mei 2013 00:52:44 UTC+2 schreef Jan Hidders het volgende:
>
> >> On 2013-05-05 19:32:35 +0000, Erwin said:
>
> >>
>
> >>> That one does have the inter-component rule that R1 JOIN R2 === R1
>
> >>> JOIN>> > R3 (as I pointed out), no ?
>
> >>
>
> >> Only if you want to be information-equivalent but conventional>>
>
> >> normalization does not require that. It only aims at obtaining a>>
>
> >> lossless decomposition.
>
> >
>
> > Well there's the whole contradiction isn't it.
>
> >
>
> > How can deliberate admission of "information differences" coexist with
>
> > "aims of being lossless" ?
>
> >
>
> > If "information differences" are deliberately allowed, this can mean
>
> > either or both of two things : something may be added, something may be
>
> > lost.
>
>
>
> Well, "lossless" means nothing is lost, but, yes, something is allowed
>
> to be added, i.e., the new schema might be able to represent
>
> information that the old one could not.
>
>
>
> > I contend that the "lossless" in conventional normalization can be
>
> > upheld only for such a perverse and narrow meaning of the term that the
>
> > claim (of having "lossless" decompositions) becomes essentially
>
> > meaningless/futile/irrelevant/...
>
>
>
> That's an .. er .. interesting claim. :-)
>
>
>
> > Lossless then seems to mean purely (exaggerating a little bit) that
>
> > "all the original attributes are still in the design, somewhere,
>
> > somehow". Relation schemas as mere sets of attributes, and
>
> > normalization as manipulation/set-algebraic computation on sets of sets
>
> > of attributes, and the only restriction being that the union of the
>
> > final sets of attributes must be equal to the original set of
>
> > attributes (/ union of sets of attributes).
>
>
>
> Nope.That's a necessary condition, but certainly not a sufficient condition.
>
>
>
> -- Jan Hidders

Yes.

That's what's escaping me all the time.

"Lossless" seems to apply to a set of some stuff, all relating to relation schema design.

But "lossless" does not seem to apply to a set of "all" stuff, all relating to relation schema design. I say this because when I point out the introduction of constraints needed to preserve information equivalence between database schemas (aka "extensional equivalence" or "isomorphism/bijective mapping between allowed database values of the designs concerned"), then this is swiped off the table with "normalization procedure doesn't require information equivalence").

So.

Can you characterise for me precisely what that "set of 'some' stuff" is to which the "lossless" claim applies ? And what stuff it is precisely that "doesn't count" ? Received on Mon May 06 2013 - 19:47:07 CEST