Re: comments on TAG 0.4

[david_marston@us.ibm.com]

I saw this in the dialog about "testable" TAs:
SJG>#A1 might not be subject to testing (even if it is testable), say.
SJG>I would say that a test assertion is a subclass of an assertion.

I think this gets back to whether TAs are only for conformance or have
more general uses. I think the latter posture is best. Each TA is one
behavior or property of the IUT. Some of these are "measurable" in a
simple boolean way, some are "measurable" at varying degrees of
precision (e.g., the red color), some are "measurable" with a degree
of subjectivity (web page accessibility, perhaps?), and some are
entirely subjective (like "understandable") with the current science,
but may be objectively testable in the future. Orthogonally, SHOULD
statements are just as testable as MUST statements, at whatever point
along the measurability scale, and I thought that you guys had settled
that issue.

If you were to say that some assertions are not **test** assertions,
you've just caused some scope creep for your TC. I don't think this is
necessary. Continue to treat everything as a TA, but have a mandatory
property that distinguishes the subclasses. Perhaps "spec" TAs are
derived from MUST and SHOULD statements, while "instance" TAs just
obtain information about the IUT. Or call them "final" and
"intermediate" TAs.

My notion of a non-testable assertion would be: something that is both
very subjective on the measurability scale and subject to implementor
discretion anyway. This may be of interest to someone building tools
that work with TAs, but not in scope for TAG.
.................David Marston