Re: Jan's well-defined view updates definition

Mikito Harakiri wrote:
>
> The problem of inverting views and solving transformation equations seems to
> be more general than just view updates. For example, given
>
> Q * D = V
>
> one can be asked to find base relations D that look like V after being
> transfomed by Q. Here V is not a single view, but a set of views, of course.
> Once again, the way we treat view updates is highly unsymmetrical: we
> consider a single view, but many base relations.

That's not a problem, just let both be database instances defines as, for example, functions that map relation names to a relation.

> Adding symmetry and generalizing the problem is a typical
> mathematical approach. (Unlike computer science that jumps to
> definitions without trying to understand the problem;-)

As a typical mathematician, you are overgeneralising here. :-P

> There seems to be a way to consolidate our positions, though.

Consolidate or reconcile? Actually I think you misunderstand what we disagree about. I never claimed that my definition is *the* definition. At most I would claim that it is probably the most liberal one that still has a firm logical basis, doesn't introduce ad-hoc rules and therefore makes sense. But as I already said, other more restrictive definitions can certainly be useful and it is very interesting to compare them to see which restrictions are introduced and what the benefits of these restrictions are.

> Given a
> well-defined view update, if we can specify an additional view constraint
> equation that gives the unique solution, we both would be happy, right? In
> your Emp example there is such an equation (found by trial and error), but
> what general case is like?
>
> Anyway, could you please outline next step(s), assuming that we accepted
> well-definess and commutativity?

Well, the point of the notion of commutatively updatable is that it captures the intuition that the user can undo or repair his or her mistakes by adding a few more updates. Sounds familiar doesn't it? Sounds very much like the constant complement approach, which again is very similar to what you are proposing (the non-updated part in the extra views stays constant). Could they be one and the same?

• Jan Hidders