Re: By The Dawn's Normal Light

From: erk <eric.kaun_at_pnc.com>
Date: 27 Oct 2004 10:06:59 -0700
Message-ID: <1098896819.678723.282990_at_c13g2000cwb.googlegroups.com>


> There are three ways a value might be "non atomic"
>
> The first is that the values might contain some substructure within
> their representation.

So "atomic" is about representation (implementation) rather than logical definition? If so, then relational has nothing to say about it.

> In this case, I take the same view you do, that distinguishing
between the
> relational engine and the type engine is useful, at least at the
conceptual
> level.

It's also useful for at least one other practical reason: users can define their own types and have the RDBMS manage relations containing values of those types. Extensibility and expressiveness.

> The second is that the values might themselves be tuples.

I believe relations are the only allowable compound types in relational theory; a tuple, while useful in programming against a relational engine, is a relation for all intents and purposes.

> The third is that the values might be sets.

What about lists and bags and such? No reason to single out sets from other "collections". And in fact no reason to distinguish it from other types.

> If I got it right, it is this third case that distinguishes the RDM
from
> the Nelson-Pick model

RDBMSs don't understand sets any more than they do lists.

  • erk
Received on Wed Oct 27 2004 - 19:06:59 CEST

Original text of this message