Re: Guessing?

On May 28, 12:46 am, Bob Badour <bbad..._at_pei.sympatico.ca> wrote:
> paul c wrote:
> > Bob Badour wrote:
>
> >> paul c wrote:
>
> >>> David BL wrote:
>
> >>>> It seems to me that every base relvar will in practice have some
> >>>> defined intensional definition outside the RM formalism and
> >>>> inaccessible to the DBMS.
>
> >>> A practitioner who (knowingly) tolerates or suggests that is likely
> >>> either a sucker or a charlatan. (Bob B called Codd's example a
> >>> straw-man.)
>
> >> Goedel tells us there will always necessarily be some external
> >> intension. However, POOD asks us to minimize it and to render as much
> >> of the system as possible amenable to calculation.
> >> ...
>
> > I think I can dig that. In other words, if one has decided to depend on
> > a system, as far as that use is concerned it's pointless to worry about
> > information the system is incapable of providing?
>
> We are talking about design. If one has decided to design a system to
> depend on, it makes no sense to exclude necessary information from the
> system. If the system is missing necessary components, how can one
> depend on it?
>
> A tractor doesn't need information about who owns what property or
> information about crop rotations to be dependable and useful for the
> farmer. As long as the tractor reliably pulls things where the farmer
> wants it to pull them the farmer can depend on it.
>
> Similarly, a dbms doesn't need information about Congress for an
> organization to depend on it for legislative compliance.
>
> Goedel tells us our formalisms can be complete or consistent but not
> both. We choose "consistent and incomplete" because inconsistent leads
> to all sorts of bad things like unreliable, unpredictable, unstable etc.
> And because inconsistency doesn't even guarantee completeness.
>
> Codd's example assumes the insert must be disjoint absent the
> information about disjointedness and then concludes the dbms cannot
> update the view. The assumption is incorrect and hence the conclusion is
> unsound.

However it would be easy to construct a similar example where disjointness of the two base relvars isn't present and so won't be available as an integrity constraint to ensure the DBMS balks on the update to the derived union relvar. In reality the update must not be applied because of missing information. The "symmetry" solution (which is to apply the update to both base relvars) tends to records facts that weren't there in the first place. This follows from the idea that base relvars have intensional definitions that are unknown to the DBMS.

The symmetry solution pretends

    p(x) or q(x) => p(x) and q(x)

which is false. Received on Wed May 28 2008 - 05:11:51 CEST