Re: Implications of Relation-Valued Attributes

From: Paul Vernon <paul.vernon_at_ukk.ibmm.comm>
Date: Wed, 2 Jul 2003 17:44:21 +0100
Message-ID: <bdv29p$14p8$>

"Marshall Spight" <> wrote in message news:jpsMa.1260$
> Hi all,
> I've been thinking about relation-valued attributes lately.

Some points about RAVs are that

  1. The relational model is orthognal to data types (except for say Boolean plus some others if you want a catalog...)

From 1) it follows that any attribute types whatsoever can be used.

2) The notion of "atomic" or "scalar" has no absolute meaning*. It depends on what level (of abstraction) you are looking at. By defintion every attribute value is atomic, but this does not rule out say relation valued attributes.

  • this is taken directly from Chris Date's article (from "What the First Normal Formal Really Means" 3) If we allow RAVs and have recurisve grouping operations on them, then there is a danger of opening the rabbit hole of structure differing from data. The TTM takes the position that as long as we don't have such recursive operators we are safe. OTOH I think I would quite like such recursive operations, but they probably should be outside the relational algebra itself, in the same way that navigating within say an array valued attribute value is outside of the algebra.

Paul Vernon
Business Intelligence, IBM Global Services Received on Wed Jul 02 2003 - 18:44:21 CEST

Original text of this message