Re: set-valued values

paul c wrote:
> paul c wrote:
> > Another maybe crazy question - if instead of 'atomic values' (whatever
> > that means) a relational engine (note for David, I've avoided using the
> > term 'DBMS' !) expressed only values made up of sets, would the presence
> > of the empty set in both true and false extensions create any problems?
> > (I'm thinking that the relational requirement of attribute names means
> > there is no problem, eg., the presence of empty sets is just an artifact
> > of the mechanism that can usually be safely ignored.)
> >
> > As for representation, sometimes such values can't be represented
> > without access to other 'attributes', eg., values that are internal to
> > an engine. My attitude (no reasoning involved I'm afraid to say) is
> > that it's okay to give the builtin result 'true' in such cases. That
> > way, the engine can proceed to manipulate the expression if further
> > requests of made of it, concerning that result.
> >
> >
> > p
>
> One reason I asked this question is because Codd started with sets, then
> veered away from them in a way, I'm guessing because of such possible
> problems, whereas the singleton value has a logical complement that is
> easily recognized if one knows its domain. It's always puzzled me why
> he avoided sets when it came to the down and dirty of implementations.

I heard a story about this from someone who heard it from someone whom he considered to be close enough to the situation to know. This is distant enough and without any second source that I consider it apocryphal (but definitely curious if anyone has heard a similar tale).

The story is that because the efforts underway within IBM to work on a DBMS based on the relational model were starting with simple domains (which makes some sense from a practical standpoint), folks at IBM wanted Codd to zero in on that and leave "nonsimple domains" out of the picture. So the theory morphed a bit to align with practical considerations, thereby sending the IT profession into a few decades of... (ah, some of you already know my opinion so I'll skip it here).

Since there is no good mathematical reason to leave out set-valued attributes, such a story would explain why they were discounted up front. Again, I certainly do not know this to be true, but the story was passed along to me a few years back. Has anyone else heard any story along these lines?

--dawn Received on Fri Dec 08 2006 - 04:27:57 CET