Re: cdt glossary - TABLE

Jan Hidders wrote:

>

That is not strictly true. A fully inverted index is a fine implementation
of a relation, and it does not have order. Every attribute is decomposed
into its own list. Thus there is no internal ordering to the rows, because it's not a row-wise representation. In fact, this would be true for any non-rowwise representation.

Part of the problem here is that people are not used to thinking about interface separately from implementation. Thus as soon as they are presented with an abstraction, they immedately conceptualize it in terms of the first implementation they can imagine. (Not saying you are doing this, Jan-- it's just something I've observed in a lot of people, and I think it really contributes the difficulty people have with thinking about unordered information.)

> This is similar to how
> denotations of a set such as {a, b, c} and {c, b, a} also introduce each
> a different order on the elements of the set. In the first denotation
> the order is a < b < c, in the second it is c < b < a. This order,
> however, is merely an aspect of the representation and not a property of
> the thing that is represented. In other words, the two different
> denotations actually represent the same thing.

Nicely put.

> For tables this means
> that the order in memory of the representation of the body is not really
> part of the body, or, put in another way, if we change that order in the
> representation then it would still represent the same body.

Yes.

Also: another attribute of any in-memory representation of a relation is that for each row, it will be possible to identify the specific chip it's on. Yet another attribute of the representation is total power consumption. Voltage. Position in space. etc. etc. None of these have anything to do with the value of the relation, though; along with order, we could change any and all of them, and still have the same relation value.

Marshall Received on Mon Jul 11 2005 - 04:03:41 CEST