Date: Mon, 05 Nov 2007 15:42:43 GMT
>> The relvars and keys example might be a little more interesting than >> mine. I'd say this is because D&D emphasize that an empty header can be >> a key. But if it is, then there would be only one tuple for a given >> relation that has an empty RVA, no other tuples [about that relation's >> keys].
> Yes, but this is only a special case of the constraint that no key may
> be a proper subset of another key for the same relation. It is not some
> strange nullological phenomenon.
>> I suppose >> if RVA's weren't allowed, a dbms that didn't support RVA's would have to >> presume the empty header as a key when no other set of attributes was >> specified.
> I see no need for presumption. Without RVAs, an AMI would probably be in
> order, and by not associating any attributes with it, one would have an
> explicitly empty key, no?
I'd see it exactly as an application presumption if the catalog had a value in the "AMIS" table that wasn't in the "KEYS" table (unless the dbms didn't support empty headers as keys.) The single table/relvar allowing RVA's seems neater to me, more exact than using two relvars.
Bob B's example of pizza prices seems more exact too. Some pricing apps might be simpler because of it. Received on Mon Nov 05 2007 - 16:42:43 CET