W3C home > Mailing lists > Public > www-webont-wg@w3.org > March 2002

TEST categorize test suites

From: Jeremy Carroll <jjc@hplb.hpl.hp.com>
Date: Wed, 13 Mar 2002 10:45:27 -0000
To: <www-webont-wg@w3.org>
Message-ID: <JAEBJCLMIFLKLOJGMELDOEFJCDAA.jjc@hplb.hpl.hp.com>

> ACTION Jeremy Carroll: look up different test suites and categorize

Summary:
========

Categories found:
A: entailment
B: class satisfiable
C: consistency between class and instance of class.
(Both postive and negative.)

B is a special case of C.

=================================

I have had a fairly brief look at Ian's [1] and Deb's (jtp) [2] test cases.

Most of the tests could be expressed as some class is or is not satisfiable.

The jtp cardinality tests were simple tests that systematically exercised a
feature (cardinality).

Ian's tests covered a variety of objectives.
Some appeared to be taken from published work, which looks like a *good
thing*.

Some were performance tests, which appeared designed to stress test the
software.
I suggest such tests are not in scope for the test cases part of OWL.

Some important tests, which I think will help differentiate between
implementation thoroughness were ones that required infinite models.
I believe that Ian's approach depends on all counter-examples being finite,
but clearly code that tries building the model to show satisfiability would
have trouble.

A few of the jtp tests did not seem to be to be easily transformable into
satisfiable tests e.g. (forgive the lisp)


        (max-cardinality child Person 4)
        (min-cardinality child Person 1)
        (Instance-of  fred  Person)
        (max-cardinality child  fred  7)
        (min-cardinality child  fred  6)


which gives a contradiction, but not because there are unsatisfiable
classes.


So, in summary, I think we have seen example tests that are in scope and
relevant of the following three forms:

- entailment (from Dan)
- class satisfiable
- consistency between class and instance of class.

Both postive and negative.

I note that class satisfiable can be expressed as the consistency between
the class and a blank instance.

Other than that I do not believe that we should expect all tests of any one
of these forms to be easily expressible in another one of these forms.
(Given enough logical machinery I suspect such mappings are possible, but
IMO test cases should attempt to assume only a small amount of machinery.)


I am imagining that John will identify a much broader range of test types
that will complement this list.

More examples of test suties would be welcome.


Jeremy

[1] http://www.cs.man.ac.uk/~horrocks/Private/DAML/Tests/
[2]
http://ksl.stanford.edu/projects/DAML/chimaera-jtp-cardinality-test1.daml
Received on Wednesday, 13 March 2002 05:47:59 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:57:48 GMT