Re: safety and external predicates

On Thu, 14 Aug 2008 21:06:23 +0100
Axel Polleres <axel.polleres@deri.org> wrote:

> Michael Kifer wrote:
> > 
> > On Thu, 14 Aug 2008 16:28:54 +0100
> > Axel Polleres <axel.polleres@deri.org> wrote:
> > 
> >> Chris Welty wrote:
> >>> I still am not convinced that safeness is anything more than an academic 
> >>> requirement for CORE.   
> >> Even if my background is academic, let me try to roll up my previous 
> >> arguments made so far from what I'd consider an implementation-oriented 
> >> viewpoint of CORE.
> > 
> > <skipping a very informative survey :-)>
> > 
> > 
> > Axel,
> > 
> > As far as I understand from the overall message, Chris' question was whether
> > decidability is an academic issue. Whether it is PR or LP, people do not care.
> > 
> > Safety is an important issue for implementing special cases, but this was not
> > the question.
> 
> Micheal,
> 
> There was one overall concern in Chris' mail before the two questions:
> 
>  >> I still am not convinced that safeness is anything more than an
>  >> academic requirement for CORE.
> 
> I replied to that one, and hope to have made clear that safeness is not 
> an academic issue, since termination or implementability of CORE on 
> existing systems certainly are non-academic systems.

No, it is not an academic issue.
I just thought that you gave a very long (but admittedly informative) reply to
a secondary issue in the email in question.


> As for decidability, let me repeat/refine the question which was also in 
> my first reply:
> 
>  > 1) Decidability: is is important that RIF-Core have
>  > decidable reasoning? That is, any compliant RIF-Core reasoner
>  > (implementation) will be guaranteed to terminate on any rule-set?
> 
> Decidability of WHAT exact problem are we after?
> 
> - Decidability/Termination of ground entailment of RIF conditions?
> 
>     my personal answer: yes, should IMO both be required for CORE.
> 
> - Decidability/Termination of non-ground queries of RIF conditions?
> 
>     my personal answer: yes, should IMO both be required for CORE.

I don't have an overarching preference, but I am leaning towards NOT requiring
decidability for non-ground queries (and not requiring that every query has a
finite # of answers).

As a separate, non-normative doc, we could specify several cases where
decidability is assured.
I prefer that to ramming just one of those special cases down the core spec.

In *that* document you can have binding patterns and whatnot. This is because
binding patterns will apply only to systems that happen to use a compatible
evaluation strategy.


michael

Received on Thursday, 14 August 2008 20:18:52 UTC