Re: By The Dawn's Normal Light

Marshall Spight wrote:

>>A value is atomic (with respect to a system) iff that value has no
>>internal structure from the point of view of that system.

I don't see the problem really, don't we all understand and agree what internal structure means? I think the key point of the definition is that atomicity is always with respect to a system rather than an absolute concept.

> Does a date have internal structure?

To the relational engine, no. It has no understanding of a date value other than its type and how the DBMS represents it internally. It doesn't know about month, days, or years. It doesn't know that they come in any order. It doesn't even know whether two distinct representations are actually the same date.

To the type engine or whatever it's called (I think I've just invented this terminology), the date does have internal structure. It can extract things like the day, month or year.

> OTOH its internal representation may simply be a long. But
> if we can't decide whether it has internal structure without
> reference to its implementation, then "atomic" is not a
> logical concept, but an implementation concept, and hence
> has no part to play in conversations about what happens
> at the logical level.

Well I guess when we say values are atomic we mean "atomic w.r.t. the relational engine". For a standard RDBMS this is almost a tautology: all values are by definition atomic. But the problem arises when we think about extending the standard relational model with things like nested relations that are accessible to the relational engine. Then our values cease to become atomic at the logical level. (in the sense mentioned at the beginning of this paragraph).

> We've had these debates over and over, and no concensus
> has emerged. I really don't think there's anything useful
> to be learned from thinking about the phrase "atomic values."

I hate to be so contrary, but I think there is use in the concept at the logical level: specifically when considering extending the standard relational model to include "nested relations" (in the sense where they aren't hidden inside the type engine).

I agree though, that if you're just thinking about the standard relational model, then the concept is pretty irrelevant at the logical level.

Paul. Received on Sun Oct 24 2004 - 17:47:14 CEST