Re: test manifest file

From: "Jeremy Carroll" <jjc@hplb.hpl.hp.com>
Subject: RE: test manifest file
Date: Mon, 11 Aug 2003 12:19:32 +0200

> I will try and explain my rationale, if it does not work for people, I am
> happy to change:
> 
> Every test of one file has two associated levels (often the same).
> One of these is the syntactic level of the file, the other is the 'level'
> (typically semantic) of the test. Really for the NotOWLFeatureTest both of
> these are redundant, and perhaps silence would be the best path.
> 
> Something that uses owl:foobar is normally syntactically not in OWL Lite or
> OWL DL (although I suppose we could have a test that was e.g.
> 
> owl:foobar rdf:type owl:Class .
> 
> )
> 
> Thus for all the tests we currently have the syntactic level of the file is
> given as OWL Full. I take Peter to not have difficulties with this.

Well, I'm still not sure what this level means then.  I have taken it that
the syntactic level of a file was the minimum OWL level at which the file
was syntactically valid.  I view all the NotOWLFeatureTest files as being
syntactically valid OWL Full, as even

	owl:foobar rdf:type rdfs:Class .
	owl:foobar owl:minCardinalityQ "ss" .

is a valid OWL Full ontology.  Admittedly it doesn't make much sense, but
this doesn't make it syntactically invalid.

> The level of the test is then a bit moot.
> I put "Lite" on the grounds that I would expect an OWL Lite consistency
> checker to recognise that this was not an OWL feature and to produce a
> warning message.

Again, what does Lite as a level mean here?  I've taken the level of
tests to be the most-constraining level at which the test makes sense.  

So for NotOWLFeatureTest, Full is incorrect, as the files are in OWL Full.
Lite, by itself is also incorrect, as the files are not just invalid in OWL
Lite, but also in OWL DL, and syntactic invalidity in OWL Lite does not
imply syntactic invalidity in OWL Full.

> The change that I would be happiest with is simply removing the level of the
> test.

I think that this would be incorrect.

> Jeremy

There are, I think, two defensible stances to take on the level property.
The first would be as above - that the level is the tightest one at which
the file is syntactically valid, or the test is valid.  The second stance
would be to use all levels instead of just the tightest one.  

The second stance is easier for automated tools to handle as they don't
need to understand in which direction the levels loosen for the particular
tests.  However, it would require changing many or most of the manifest
files to make the implicit levels explicit.

peter

Received on Thursday, 14 August 2003 10:49:09 UTC