RE: Manual Rewriting and Passing Entailments

On September 16, Jeremy Carroll writes:
> Peter wrote:
> > However, I think the WG may need to split out the (at least)
> > three possible uses of the tests in order to make progress on
> > this discussion:
> >
> > 1) Political expediency dictating that {all, many} of the tests
> > should be passable by {all, many} of the implementations.
> >
> > 2) Useful test cases for implementors, intended to exercise
> > everything from the trivial to the complex
> > 3)
> > Torture tests
> >
> I am pretty sure that Dave was coming solidly from the (2) case - he has
> found some of our tests useful, and he found some others of our tests useful
> after a little rework - he was just offering these back into our collective
> implementors' tool-kit. Personally I think we should only reject them if we
> believe everyone else is going to implement the comprehension axioms. Seems
> unlikely, OWL implementors who aren't chasing completeness, may well wish to
> avoid thinking about comprehension, or like Dave, think about it and decide
> that it is not appropriate in their environment.
> I am sure he is surprised at the level of debate this offer has excited.

And gratified I hope :-)

> (Type 3) is my favourite - why do we have such a complicated language if
> there aren't implementations !! :) ).

Depends on what you mean by torture tests. If you mean tests designed
to test/break implementations by introducing ever larger amounts of
non-determinism, then I agree that we should have such tests, but
suggest that they are less useful in general than smaller tests
designed to exercise tricky aspects of the semantics and/or of known
algorithms (e.g., cases where termination is problematical for tableau
algorithms). As an implementor, these latter kinds of test (which I
take to be Type 2) are much more useful, and hard to invent myself.

As far as torture tests are concerned, it was always clear that the
complexity of the consistency problem for any but the most trivial
language permits the creation of moderate sized tests that cannot be
solved in reasonable time by any implementation. The assumption is
that such pathological cases will rarely be encountered in practice.


> Jeremy

Received on Saturday, 20 September 2003 06:08:16 UTC