RE: Error in (approved) test 0009

On Mon, 2007-08-06 at 19:44 +0200, Hausenblas, Michael wrote:
> Dan,
> 
> Sooo many question for a simple-minded person, as I am :)
> 
> Ok. One step at a time.
> 
> > If you changed the content of the test, the status should 
> > revert to unapproved, right? i.e. a new decision should be 
> > made to approve the new contents.
> 
> I'm not a W3C-Test-Professional, but down-to-earth: When we simply 
> *fix* a broken TC -- which intention and semantics has been approved --
> we do not actually *change* but (ehm what was the word again? ah!) *fix*
> it ;)
>
> So, if you prefer it, I'd happily propose to review this one again,
> from a efficiency POV I'd keep it approved. 

I would need to sync up with Ralph Swick and/or Steven Pemberton
before I ask you to re-do some work, but...

I don't know how to approve "intention and semantics"; I only know how
to approve sequences of bytes.

In the GRDDL WG, sometimes the test suite editors were given
discretion to change some bytes after a test case was approved...
I was a little uncomfortable with that, but as long as it was stuff that
was really orthogonal to GRDDL (things like
license text in XML comments, as I recall), I let it go.
But in the case of 0009, the change is clearly substantive
and not editorial. The TF records show a decision to approve
a SPARQL query that said ... WHERE { X P Y }; it's quite a stretch
to say that decision applies to a SPARQL query ... WHERE { Y P X }.
Perhaps you can get away with that once or twice, but if it happens
dozens of times, the evidence that the test suite was reviewed
and approved by the TF gets much less compelling.

> > Hmm... I'm assuming this test suite is organized roughly like 
> > the RDF Core, OWL, SPARQL, and GRDDL test suites, where 
> > approval of a test is always traceable to a recorded 
> > decision. Is that the way this test suite is organized? 
> 
> Roughly, yes. We're still in a kind of learning phase - so any
> improvement suggestions welcome :)
> 
> However, I'm very happy you asked this; till now I assumed
> either everybody would understand the system -- or nobody would
> really care ;)
> 
> So, here is the idea:
> 
> Every section in [1] has a title 'Review and Approval YYYY-MM-DD'
> with YYYY-MM-DD representing the date the review took place (in case
> of a telecon, or when it was closed in case of a mailing list review).

Ah... I had missed that... so test cases are grouped by date. Hmm...
what if a test case is discussed multiple times? Does it show up
in multiple places? I suppose I can see how this works.

Hmm... do test cases get URI names? What's the full URI of
test 0009?

Ah... per the manifest, it seems to be:

http://www.w3.org/2006/07/SWD/RDFa/testsuite/xhtml1-testcases/rdfa
-xhtml1-test-manifest.rdf#Test0009


I don't see links in the manifest to the approval decision records.

You have explained quite a bit about how this works, but I'm still
not quite sure I understand. It's possible that you just have to
be at the teleconferences to understand the process. That's fine,
for now, as long as it's eventually written up in a way that's clear
enough for consumers of the tests.

-- 
Dan Connolly, W3C http://www.w3.org/People/Connolly/

Received on Monday, 6 August 2007 18:13:12 UTC