Re: Interesting article: In the Beginning: An RDBMS history

Jan Hidders wrote:
> dawn wrote:
> >
> > Outside of the issue of how coupled a logical model is to the physical
> > model, I see no harm in numbering the attributes provided you also name
> > them.
>
> It also makes the theory more complicated than necessary. What if we
> take the join of two relations, what is then the order of the
> attributes in the resulting table?

Yes, I responded to Marshall on this. I can see that this ordering is outside of relational operators. One way it is handled is to have only stored data include an ordering which is not present in any derived views.

> How do you define this such that a
> join is still commutative? This is rather crucial for quey
> optimization.

I can see that would be the case if using a set-based query language.

> And what about unions? Are they allowed already if the
> set of attributes is the same, or does their order also have to be the
> same? If the orders are allowed to differ, how is then the order in the
> result of the union defined in such a way that we keep the usual
> algebraic identies?

Yes, when doing set operations on an ordered set of attributes, everything is handled as if the ordering were not there. So, the ordering is irrelevant to SQL, but employed by other languages working with the data. I don't know that it is ideal to have this, but it doesn't seem to be a problem to model the stored data as actual relations (with attributes ordered) either. --dawn Received on Fri Apr 07 2006 - 16:26:21 CEST