tag-discuss message
[Date Prev]
| [Thread Prev]
| [Thread Next]
| [Date Next]
--
[Date Index]
| [Thread Index]
| [List Home]
Subject: Reasons to wait on convening a TC
- From: david_marston@us.ibm.com
- To: tag-discuss@lists.oasis-open.org
- Date: Mon, 27 Nov 2006 22:51:56 -0500
I agree with some of the more outspoken
participants on this list that we are converging on what could/should be
done, and what does not have to be done, but I think we have more converging
yet to do, and I will comment on some issues below. Just as important,
we need better information about whether specs such as SBVR [1] are good
enough. Perhaps a special profile of SBVR for test assertions (TAs) will
be all that's needed.
There are several people on this list
who worked on the OASIS Conformance TC[2] and/or the W3C QA documents[3],
and I would like to see remarks from more people!
Points where I think we are well
converged:
1. The TC, if one is needed and convened,
would specify a way to state TAs, given the definition [4] cited by Jacques
from prior work.
2. The main target user of a specification
for TAs would be a document editor for specs of the type commonly issued
by OASIS, W3C, OMG, etc.
3. If a spec is published on this topic,
it should provide a skeletal structure for a generic TA, as a fill-in-the-blanks
guide. Another way of stating this is that the spec would provide a model
of a well-formed TA.
4. In some way, TAs are normative "statements"
that establish testable constraints of a spec. Current specs often contain
many solid, formal, testable sentences in their prose, so TAs need to present
significant added value.
Points where we might be converging
on the conclusion that these are NOT IN SCOPE:
1. Defining test case metadata. Several
developments are occurring in this area.
2. Addressing execution of tests is
only in scope for the notion of testing a given TA.
3. Notwithstanding (2), there is a useful
parallel between the preconditions or setup for a test case and the contingencies
of a TA. If the two can be tied together, it supports automated generation
of test metadata from the TAs.
4. Profiles/levels/modules/discretionaries
have already been done (see Spec Guidelines and ViS at [3]), so TA work
should use this prior work as a foundation rather than reinvent the Dimensions
of Variability.
Points where convergence seems likely,
if not already achieved:
1. There needs to be synchronization
between the TAs and the prose of the spec, such as by one deriving from
the other.
2. Testing can be automated with a test
harness driven by platform-independent test case metadata. TAs should be
cited by the metadata. The TA spec might make a few other mild assumptions
about the test harness.
3. There is no need to try to limit
the number of TAs for any given spec. Assume that automation will be involved
if the number is large.
The question of representing TAs in
XML is still open, apparently. I was astounded when Dave Pawson wrote on
11/27/2006:
>I really
don't see that mandating XML as the way in which tests are expressed
>is useful? For some it will be that the XML will be an unecessary wrapper
>for simple text.
Perhaps Dave and others are
unaware of how well XML has helped with test case metadata for XSLT/XPath/XQuery
testing. (The XQuery test suite [5] has a catalog of thousands of test
cases.) There are two kinds of benefits: filtering of the collection down
to relevant subsets, and re-purposing the data by XSLT transformations.
Keep in mind that XSLT can produce a text file as output. For test cases,
this means that the test case metadata can be transformed into a platform-specific
script for sequential execution of a set of cases. For TAs, I think that
well-designed markup might support "joins" of two or more TAs
that eventually give the specs of what individual test cases do. In some
cases, the specs could be further transformed into actual inputs for testing.
Even if we are unsure of the state of
the art in automated reasoning over a set of premises, we can assume that
someday it will be possible to process a set of TAs by software and obtain
further value. The further value might be about test coverage or test cases
needed, or it may be about consistency in specs. If we design TAs to be
amenable to machine processing (e.g., by expressing them in XML), then
there is hope that early efforts at TAs will become more valuable in the
future as new tools emerge.
In addition to the workload imposed
on the TC that would try to develop this specification, we must keep in
mind that attempts to compose conformant TAs would increase the workload
on the various TCs and WGs that attempt to use it. That's why my earliest
remark in this forum was that the TAs have to contribute enough value to
be "worth it" for the WG that uses them. I think such value can
only be realized if the TAs are usable in automated testing.
.................David Marston
[1] http://www.omg.org/docs/dtc/06-03-02.pdf
[2] http://www.oasis-open.org/committees/documents.php?wg_abbrev=ioc
[3] http://www.w3.org/QA/Library/
[4] Test
Assertion: a Test Assertion is a statement of behavior for an implementation:
action or condition that can be measured or tested. It is derived from
the specification’s requirements and bridges the gap between the narrative
of the specification and the test cases. Each test assertion is an independent,
complete, testable statement on how to verify that an implementation
satisfies a conformance requirement.
[5] http://www.w3.org/XML/Query/test-suite/
[Date Prev]
| [Thread Prev]
| [Thread Next]
| [Date Next]
--
[Date Index]
| [Thread Index]
| [List Home]