Re: separation of church and state?

On Oct 6, 9:06 am, paul c <toledobythe..._at_ooyah.ac> wrote:
> I finally sprung for CJ Date's "Writings, 2000-2006" and skimming it,
> noticed this point in chapter 10. It reminded me of the recent posts
> about avoiding books that start with silly sentences even though this
> quote is from page 174:
>
> "Ordering, by contrast, is not part of the relational algebra; nor can
> it be, because its result isn't a relation. This doesn't mean you can't
> have an ORDER BY operator, of course - it just means that operator isn't
> part of the algebra as such, and it can't be used in an expression
> that's nested inside some other (relational) expression, or more
> generally in any context where the result is indeed required to be a
> relation. That's why you can't use ORDER BY in a view definition, for
> example."
>
> It seems a little doctrinaire to me. I can agree that the "result isn't
> a relation" but on the other hand a user could see such a result without
> knowing that "ORDER BY" was involved and not be faulted for taking it to
> be a relation. For that matter, in some apps, users take it for granted
> that all results are arbitrarily ordered and that those results can be
> used to produce other results.

I'm not happy with Date's treatment of this issue. It doesn't distinguish
among the different kinds of order for one thing. (Preorder, partial order, total order.) For another, it seems to rely on some intuition about what ordering is, but he doesn't spell that out and I don't think it's even all that good.

It seems he's assuming that 1) there's only one kind of order, 2) that an ordered set is isomorphic to a sequence, and 3) that it doesn't matter
what one orders on. In fact, 1) there are at least three different kinds
of order, 2) is only true for finite totally ordered sets and 3) a total
order applied to a set of attributes that is not a superkey becomes a preorder for the relation. I would expect a thorough treatment of order within the relational model to address all of these issues.

> By analogy of separating the logical from physical implementation, if
> you want to declare a separation of church and state, I'd think you'd
> need to mention both. Not to tout SQL but I took the above to mean that
> if a table were declared with an "INDEX", it shouldn't be allowed to
> participate in expressions of the relational algebra, which seems
> extreme and somewhat useless to me.

To me this is a separate issue. On the one hand we really want to be able to separate logical results from their performance characteristics. On the other hand, it's important that we have some mechanism for those writing queries to be able to have some model of performance, or anyway some way of considering performance when necessary. Virtually all languages don't do the first one very well; SQL is an exception here. I can't think of an analog in another language to being able to add an index and change the performance of a query independently from its semantics. That's a pretty cool feature.

Marshall Received on Sun Oct 07 2007 - 01:34:21 CEST