Re: more closed-world chatter

On May 6, 11:21 pm, Jon Heggland <jon.heggl..._at_idi.ntnu.no> wrote:
> Marshall wrote:
> > So, there are a variety of possible answers. Bob
> > for example considers this within the context of
> > a type theory that has subtyping, whereas Jon
> > proposes a type mismatch. Both answers have
> > their merits. Another question: what is the type
> > of that attribute in the result relation? Possible
> > answers include the top of the type lattice and
>
> The most specific common supertype, I'd say (parroting D&D, as usual:).
> Which might be the top, of course, but not necessarily.

Ah, yes; that's what I should have said. We could consider defining the answer to be the least upper bound of the type lattice or the greatest lower bound.

> > the bottom; I am inclined to prefer the bottom.

Restating: I am inclined to prefer the greatest lower bound.

> Are you familiar with TTM's IM prescription 13: Join etc. with
> inheritance? Are you able to reconcile your preference with that?

Is that the one that says the type of an attribute in the result of a join is the most common specific supertype? If so, yes I am familiar, and no, I can't reconcile it. I don't see any reason to prefer a type that's more general than necessary. I also wrote up a brief analysis in another post, in which I dropped attribute types and added equivalent constraints and worked it out from there; it also came up with the greatest upper bound.

Or just generally speaking, we want to propagate as much information from the operands into the result as possible, which means the result should have the tightest bounds that are sound. This is the least upper bound: there will be no values in the attribute of the result that belong to any type more general than the least upper bound.

Marshall Received on Mon May 07 2007 - 09:25:30 CEST