Re: SEMANTICS - we need them NOW

I agree with Peter. What do you test with a test case when what you mean
(semantics) is not defined? Or ill-defined? I simply do not understand compliance
testing without knowing what the results should be. How do you know what
fails/succeeds if there is no well-defined target failure/success?

Now, it is also true that hypothetical test cases can help drive out semantics,
because they help drive out requirements, which help drive out the language and
the semantics.

Leo

"Peter F. Patel-Schneider" wrote:

> From: Dan Connolly <connolly@w3.org>
> Subject: Re: SEMANTICS - we need them NOW
> Date: 05 Aug 2002 13:20:43 -0500
>
> >
> > On Mon, 2002-08-05 at 10:32, Ian Horrocks wrote:
> > >
> > > Dear All,
> > >
> > > As you may have noticed from the discussion about testing, we are
> > > already in trouble within the WG because we have so far failed to
> > > resolve the central question of the semantics of the language.
> >
> > Huh? Discussion of test cases is one of the ways that we decide
> > the central questions of semantics.
>
> Agreed.   But note what you just said.
>
> > > In
> > > particular, it really makes little sense to continue with work on
> > > developing a test suite until we know the meaning of the language we
> > > are supposed to be testing.
> >
> > My experience says developing tests as we make the relevant
> > decisions is best.
>
> Again agreed.  But, again, note what you just said.
>
> Test cases without semantics can be very dangerous.  In particular, it is
> possible to have a collection of test cases that all appear to be fine, but
> that do not lead to a well-defined semantics.  (Think of the test cases that I
> put together leading to the paradoxes.)
>
> Small collections of test cases can be used to distinguish between
> different behaviours of a formalism, or even to drive the search for
> meaning for a formalism.  However, test cases in the absence of a meaning
> for a formalism are not useful.
>
> I feel that decisions on test cases (and other things as well, by the way)
> are being made without an overall view (or, at least, with multiple,
> diverging, overall views) of what OWL constructs are supposed to mean.  I
> fear that we are in danger of ending up with a collection of test cases
> that cannot be extended to a cohesive whole.
>
> > --
> > Dan Connolly, W3C http://www.w3.org/People/Connolly/
>
> Peter F. Patel-Schneider
> Bell Labs Research

--
_____________________________________________
Dr. Leo Obrst  The MITRE Corporation
mailto:lobrst@mitre.org Intelligent Information Management/Exploitation
Voice: 703-883-6770 7515 Colshire Drive, M/S W640
Fax: 703-883-1379       McLean, VA 22102-7508, USA

Received on Tuesday, 6 August 2002 17:08:44 UTC