Re: New verbiage for CP 8.4 <- AI-2003-0122-2 (was: "Draft minutes" of January 22, 2003 Teleconference)

The new verbiage isn't working too well for me. One part is that I
don't think all readers will take "factorize" to mean the same thing.

Perhaps the rationale could focus on the test suite....
1. Being a discretionary item means that different implementations can
   do different things.
2. Can still have test cases for each of those things.
3. Test lab needs to know which choices were made, etc., in order to
   know which conformance test cases apply.
4. Even if the user gets to make a choice (say, by API argument), the
   test lab still needs to know how the user does it, so they can
   test all available behaviors.
5. If spec doesn't want implementation to pass through choices to user,
   and the nature of the discretionary item makes it feasible, need
   explicit verbiage to say implementer must choose.
6. Test lab that wants to test multiple implementations needs to know
   that there is parity or substitutability across the alternatives,
   and when one discretionary item applies only if a particular
   choice was made on another item.
7. Discretionary items might only apply to certain modules, levels,
   or profiles, which should be stated in spec.

We also need something about bundling. Stated in a positive way, one
bit of discretion (e.g., how strictly to enforce schema typing) may
be a global choice for the implementer that is mentioned in many parts
of the spec where it has individual consequences. A spec that has the
scattered parts and does not bundle them into a single policy-level
discretionary item should fail this checkpoint. This "consistency"
test requires good wordsmithing.
.................David Marston

Received on Tuesday, 28 January 2003 12:56:14 UTC