[xslt-conformance] Re: Rendition Instance stylesheets from Carmelo

[David Marston/Cambridge/IBM <david_marston@us.ibm.com>]

A few notes:
You can't use @id to uniquely identify test cases across all submitted
suites. You need to form a master id by concatenating the suitename and
the test @id assigned by the submitter. This would ensure that there is
no clash between, say, LotusIBM-numbering01 and Microsoft-numbering01.

This rendition stylesheet assumes that we will ship an XML file that
holds the results of our test-case review. It hadn't occurred to me
that we would do so, but I can see how that opens up an interesting
opportunity. We could post revised editions of our evaluation file
every so often, for test labs to download. The hope is that more tests
would migrate to "s1" status over time, like when we get replies from
the WG about gray areas. The whole TC should discuss this idea.

The matching of discretionary items needs a file containing the
questionnaire answers for the processor under test. Then the items in
the test case must match the items in the questionnaire, item for item,
with the same behavior. One test case can use more than one discretionary
choice. Picking the raise-error case as a default isn't fair to all
test subjects. Not all discretionary items even offer that behavior as
a choice. If we don't have a test case that matches on all discretionary
items, we just have to forego testing that aspect of the processor. In
other words, we may apply 4728 tests to Processor A and 4691 tests to
Processor B due to exclusions. That may mean that Processor B is not as
well tested as Processor A. That's tough luck until we can fill in all
the gaps, which I think is in the distant future.

Based on the above discoveries, I decided not to review the stylesheets
in detail, because addressing the above could cause substantial changes.
.................David Marston