Re: GRDDL Test Documents - moving all the tests, including approved tests?

On Wed, 21 Mar 2007, Dan Connolly wrote:

> I would rather not use that part of the OWL test process.
>
> The URI of a test is an integral part of the test, especially
> for EARL reporting purposes. If you move the test in URI space,
> you modify it substantially, such that at least *my* approval
> of it is void, and I think the WG's approval of it should be
> void.

I don't see how simply moving the URI of a test (consistently) voids the 
approval, especially if the change is to the URI *only*.  I.e., consider 
if as part of the test we calculate a checksum of some kind for the source 
document.  If we move the URI, but the checksum is the same, is it really useful to 
consider it a different test? Do we really have to resort to an 
owl:sameAs to explicitely state that two tests with the same content but 
different identifiers are the same for testing purposes?

I can see the argument for keeping a consistent base URI for the 
'networked' tests as the test subject would be fetching the source document 
from the same location.  And assuming http://www.w3.org/2001/sw/grddl-wg/td remains 
the permanent repository for the GRDDL test cases (with a lifespan which 
spans beyond that of the WG), then yes, I'd agree we should maintain the 
absolute URI's for those tests.

However, for tests that can be run locally (and this is the majority), I 
don't see the value in *not* allowing the tester to resolve a relative URI against a base of their 
choice (their testing environment) for the sake of reporting results.  It would seem counter to the
direction we are already going in providing the convenience of running such tests locally.

Making a determination of an appropriate base URI for each test is a 
simple matter of adding a localized xml:base and I don't see this as a 
procedural nightmare once we have some consensus on a naming 
convention for the tests that accomodates the fact that some can be run 
locally and others cannot.

A suggested convention:

- For networked tests, resolve all relative URIs (in input and output documents) against http://www.w3.org/2001/sw/grddl-wg/td/
- For localized tests, resolve all relative URIs (in input and output documents) against a base URI appropriate for the environment where the test was performed

I think the combination of zipping up the tests 
*from* a single location (*the* permanent test repository) for delivery 
with the test document, the use of localized xml:base to match a WG decision on a URI naming 
convention for the tests, and the use of relative URIs consistently (in 
inputs and outputs) will solve the headache of a delivery mechanism for publication as well as the 
management of cross references.

The only uncertainty *I* have is with regards to the value of including 
the networked tests in the zip delivered with the test specification. 
Probably a minor consideration, but I'm curious about other opinions on this.

> Changing the CVS last-modified date of the input/output files
> in a test such that they're after the date of the decision
> also makes the records much harder to audit.
>
> Perhaps re-approving tests in bulk is not a big obstacle,
> but I do see it as a consideration.

I don't see bulk-reapproval as much of an obstacle, but I also don't think 
it is neccessary as having an explicit naming convention goes more 
directly to the heart of the issue.

Chimezie Ogbuji
Lead Systems Analyst
Thoracic and Cardiovascular Surgery
Cleveland Clinic Foundation
9500 Euclid Avenue/ W26
Cleveland, Ohio 44195
Office: (216)444-8593
ogbujic@ccf.org

Received on Wednesday, 21 March 2007 14:49:51 UTC