W3C home > Mailing lists > Public > www-qa-wg@w3.org > January 2003

Re: Open issues on SpecGL (the AR-001 to -028 list)

From: David Marston/Cambridge/IBM <david_marston@us.ibm.com>
Date: Tue, 21 Jan 2003 23:59:17 -0500
To: www-qa-wg@w3.org
Message-ID: <OF499199F9.AE8F8228-ON85256CB6.001722C4-85256CB6.001B8C37@lotus.com>





>- AR-015,AR-020(a) (limited technique of justification for DoV usage);
>proposal: adopt the more generic approach...

Agreed. Also, ensure that Ck 8.1 doesn't repeat 3.5. Under 3.5, we
require that use of discretionary items (one or more) be justified in
general. Guideline 8 checkpoints should refine: why this particular
discretion grant, and why not tied in with others if it's not.

>- AR-014 (other sources of definition for conformance terms)

AR asked why the definitions in SpecGL are special. They probably
are, but they could still be cited just as RFC2119 is cited.

>- AR-022: we need a rationale for CP 8.3

Interaction with other DoV (e.g., discretion only applies if foo
module is implemented) should be explained somewhere. Is this the
CP that requires it? Also, verbiage may inadvertantly imply that
only one choice may be implemented when that was not the intent.
If multiple choices are allowed, it affects tailoring of the
test suite.

>- AR-023: need rationale and maybe rewording

This is about Ck 8.4, viz.:
> To fulfill this checkpoint, the specification MUST state that
> given identical conditions, the effect of a discretionary choice is
> consistent within a single implementation.

There are some sub-goals here, each with a different rationale.
First of all, discretionary choices should be consolidated up to the
"design philosophy" level: developers should be choosing a policy
rather than making micro-choices about whether situation A is an
error but B is not. Also, the spec MUST address whether a choice is
made by the implementer and compiled in (and the doc for the
product specifies which choice was made), or the implementer is
allowed to pass through the decision to the user. At some stage of
setup, someone has to make the choice, else we wouldn't know which
test case to apply. In particular, the end user should get errors in
a predictable fashion, which (I think) is the aspect that the above
verbiage addresses.

>- AR-010 (GL 3 and 10 merging): ...
>Otherwise, I think the issue boils down to: does it make sense in
>specGL to differentiate the process of defining a policy with the way
>to put it in the spec?

Well, GL3 lives on, so I think we said that GL10 is about writing
down *all* the decisions made on GL1 to 9. GL3 is tested by looking
for statements in the Conformance section that present the overall
conformance policy and, where applicable, the policy for each class
of product. Maybe the problem is that GL3 is no more desrving of a
forward reference to the Conformance section than any other GL
between 2 and 9. I think GL3 is also where specs like Infoset are
required to say that their conformance assessment technique is
indirect.
.................David Marston
Received on Wednesday, 22 January 2003 00:01:21 GMT

This archive was generated by hypermail 2.2.0 + w3c-0.30 : Thursday, 9 June 2005 12:13:12 GMT