Re: NULLs

On Jan 4, 2:50 pm, "Brian Selzer" <br..._at_selzer-software.com> wrote:
> "Hugo Kornelis" <h..._at_perFact.REMOVETHIS.info.INVALID> wrote in message
>
> news:60rqn35dm3og64dt42hgdisgm8rut5lqr2_at_4ax.com...
>
>
>
> > On Sat, 29 Dec 2007 10:34:06 -0800 (PST), Marshall wrote:
>
> >>On Dec 28, 2:59 pm, Hugo Kornelis
> >><h..._at_perFact.REMOVETHIS.info.INVALID> wrote:
>
> >>> 3VL is not a result of using NULL to represent missing information, but
> >>> a result of allowing missing information. IMO, there are only two
> >>> options: either you deal with missing information, and with the 3VL that
> >>> results from it -- or you somehow alter reality so that information is
> >>> never missing again, for any reason.
>
> >>Full normalization is a third option.
>
> > Hi Marshall,
>
> > As in "one table per elementary fact type"? Full agreement here, see my
> > message to Jim.
>
> Excuse me, but I don't think normalization has any bearing whatsoever. Even
> with a fully normalized schema it may be that there can be missing
> information.

I think we have to be careful on terminology here - if a model has no nulls in it then as far as the /logical model/ is concerned absolutely nothing is missing. All predicates are satisfied by full instantiations.There is a distinction between what it means for information to be missing at the conceptual layer and the logical layer, and I fear a lot of arguments about the issue on cdt are down to not specifying which layer is being discussed.

> Even if there is one table per elementary fact type, there can
> still be missing information: the difference is that instead of a null that
> indicates that there should be a value but it just hasn't been supplied, you
> have to decide whether the absence of a row indicates that the attribute
> does not apply or that it does apply but that a value just hasn't been
> supplied.
> Splitting the tables only alters the indicator from an explicit
> one to an implicit one, and information is lost in the process. Moreover,
> decomposition isn't enough, you must also alter the intended interpretation
> for those elementary fact tables so that each represents what /is known to
> be/ the case instead of what /is/ the case. Each table would then be closed
> with respect to what is known about the world, but this can lead to
> confusion: if it is true that it is known that 'X' is the case, then 'X' is
> the case, but if it is not true that it is known that 'X' is the case, then
> 'X' may or may not be the case. So you again are faced with 3VL.
>
> Using nulls that indicate only that there should be a value but it just
> hasn't been supplied interferes only with the Domain Closure Assumption,
> since it can no longer be determined with certainty which individuals exist,
> but using one table per elementary fact type instead plays havoc with both
> the Domain Closure and Closed World Assumptions, and introduces an extra
> level of indirection: the database no longer directly represents reality (as
> is the case when using an explicit indicator), but rather what is known to
> be true about reality.
>
> > I think I addressed the rest of your message in my reply to David. Let
> > me know if you think I failed to address any of your points.
>
> > Best, Hugo
Received on Fri Jan 04 2008 - 16:10:48 CET