Date: Wed, 07 Nov 2007 09:13:26 +0100
Quoth paul c:
> One thing that annoys me about them is how they allow two attributes
> with the same name to be thought of as "in the same outer relation", if
> you will.
What do you mean? That an RVA might have an attribute with the same name as an attribute in the enclosing relation? If you think of these as "in the same outer relation", it is your thinking that is faulty, I'd say. I don't see the difference between this and an attribute having a possrep with a component name that is the same as a different attribute. But perhaps I'm misinterpreting you.
> But I'd like to suggest that even though I throw the term RVA around,
> pretty loosely too, I'm not sure all these posts have been about RVA's
> but rather something else, even if that something else is not well
> defined in the context of the rest of RT. Perhaps "set-valued
> attributes" describes it better.
Definitely; all the examples I've bothered to read (and write) have been using unary "RVAs" without attribute names, so one might as well talk about SVAs (which don't seem to be in dire need of better definition). And I find this amusing, as I've gotten the impression that there has been established a tentative consensus here that SVAs and indeed all kinds of XVAs /except/ RVAs are fine and dandy, and don't violate 1NF; but that RVAs are dangerous, and lead to second-order logic, inconsistency, confusion and excessive renaming---apparently because the relational engine "understands" them. Pish and tosh, I say. Cumbersome, yes. Mostly useless, yes. Fundamentally different from other data types, no.
-- JonReceived on Wed Nov 07 2007 - 09:13:26 CET