# Re: atomic

Date: Fri, 02 Nov 2007 18:19:04 -0300
Message-ID: <472b944c\$0\$14867\$9a566e8b_at_news.aliant.net>

paul c wrote:

>> paul c wrote:

>
> ...
>
>>>
>>> For example?
>>
>>
>> The set of known salaries for an employee.

>
>
> If we mean salaries for one employee, eg., a single attribute relations
> like:
>
> KnownSalariesForJoe
> ___________________
>
> {10,000, 20,000},
>
> how could we make sense of:
>
> KnownSalariesForJoe
> ___________________
>
> {}
>
> ?
>
> eg., if the set of known salaries for "Joe" is empty, a dbms that
> supports RVA's can record this second relation, but couldn't we just as
> well have an empty value for it:
>
> KnownSalariesForJoe
> ___________________
>
>
> so I wonder: why allow the predicate to require a set in the first place?
>
> One argument for why might be that the predicate has been agreed to by
> all users (after a million dollars was spent assiduously gathering
> requirements). But when it comes to unnecessary requirements I don't
> care many how many users agree!
>
> If we mean "known salaries for each employee", I think that phrase
> presumes that we can identify each employee which means that Codd's
> "normalization" method in his 1970 paper applies, which means in turn (I
> think) that we don't need RVA's for this relation, so for me the
> question of an empty RVA doesn't "need" to come up.

We never "need" RVA's.

> (I should say that I'd guess Bob B among others here might guess that
> I'm talking towards my usual recurring them, to do with RVA's and
> recursion, the possible importance of which I admit I haven't been able
> to express in concise terms.)

R = { A int, B typeof R } -- works
R = { 1, valueof(R) } -- blowed up real good Received on Fri Nov 02 2007 - 22:19:04 CET

Original text of this message