- From: Jeremy Carroll <jjc@hplb.hpl.hp.com>
- Date: Wed, 07 Jan 2004 18:23:07 +0000
- To: david_marston@us.ibm.com
- Cc: Sandro Hawke <sandro@w3.org>, www-qa-wg@w3.org
> > How did you demonstrate the existence of more than one conformant (and > presumably interoperable) implementations? Sandro's explanaton was clear I think. I'll add one point. The tests give one or two (sometimes three) documents and specify that these documents have some property. If we have many different products, including ones that the WG might not have thought of, and for each of these the implementor signs off as "yes my implementation treats this set of files (with one, two or three members) as having this property, in an appropriate way, and we have sufficiently pertinent tests (maybe in terms of coverage, maybe only in terms of coverage of the difficult bits) then that does show interoperability. To go with Sandro's <em> example Data: <html> <body><p>I <em>like</em> this.</p> </body> </html> Test: "like" is emphasised We might have a voice2html converter which is tested by a human reading out the text emphasising "like" and low-and-behold the test data is produced. We might have a voice browser that reads "like" with emphasis, and we might have a standard browser that shows "like" in italics By expressing the test abstractly enough we can demonstrate interoperability between these different products. (Obviously this example is not credible). I am not sure how much variety of product we actually had running the OWL and RDF core tests, but I am pretty sure that there were some significant differences in the way they executed the tests - I think that's been a strength rather than a weakness. Jeremy
Received on Wednesday, 7 January 2004 13:35:06 UTC