Re: Notions of Type

erk wrote:
> David Cressey wrote:
> >
> > I'm surprised the PROJECT is such a problem. Maybe I should stay out of the
> > discussion, because this is a little over my head. But here goes, anyway:
> >
> > Why can't you define a "set of attributes" as a relation? I'm thinking
> > that an empty relation (one with no tuples) has exactly the same
> > information content as a "set of attributes". If you do that, why can't
> > you say,
> >
> > PROJECT <relation>, <empty relation> -> <relation>
> >
> >
> > Or have I violated some other aspect of the formalism?
>
> No, you can certainly do that. I was trying to be conservative in the
> type definition - a relation we'll be using only for its attribute
> definitions (its header, not its body) isn't really being "used" as a
> relation. For that reason, it seemed parsimonious to require only a set
> of attributes. One still needs to constrain the header of the second
> relation - it can't have attributes the first one doesn't.

Oh, but it can, because we're using a generalized operator that can do much more than just PROJECT.

Marshall Received on Fri Aug 18 2006 - 07:53:30 CEST