- From: Kirill Gavrylyuk <kirillg@microsoft.com>
- Date: Sun, 9 Mar 2003 22:10:27 -0800
- To: "pfawcett" <pfawcett@speakeasy.org>, <www-qa-wg@w3.org>
Great work indeed! Had couple questions/comments (since I missed the discussion, you might have covered it there): - tagging asserts/normative language by categories is a nice idea. We should probably revisit what do we understand by a test assert though to enable such tagging. Plus some of categories (like "undefined") don't have asserts associated with. >(cross platform stuff goes to ExTec.) - Was there a reason behind removing "platform independent" criteria form the guidelines doc? - system must allow for tests to be filtered based on metadata criteria. Sounds a bit absolute. How much of filtering capability would we like? For any attribute in the metadata? - Addition of test cases. For test management system, the previous draft required ability to add tests for any are of the spec. May sound trivial, it's a common flaw of the test management systems. - Ease of use. Looks like gone in this outline. May sound trivial and non-specific, but important for the conformance test cases. - Spec coverage. Is this part of the support for metadata? - Some of the "operational like" checkpoint removed (for good reason) could find their place at the Operational Guidelines. Shall we make such analysis, to make sure we are covered? Again, apologies for not being able to call in last week - was loaded over the top. -----Original Message----- From: pfawcett [mailto:pfawcett@speakeasy.org] Sent: Sunday, March 09, 2003 7:31 AM To: www-qa-wg@w3.org Subject: New Possible TestGL outline Howdy folks, This is a brief break down of the new structure/outline for TestGL that Patrick and I came up with after fridays meeting. We both think that this has a significantly better flow to it, addressing some of Lynne's major general concerns. We also tried to steer away from 'loaded language' like methodology, test framework and a few other terms we found in the document that caused difficulty as they meant different things to different folks. We also tried to condense things down a bit. At the beginning we thought we would be getting rid of a guideline but instead we ended up creating a new one. But the checkpoints are down to a reasonable 2-4 per guideline rather than up to 14 in some. Finally we made an effort to not duplicate effort (and thus potentially get out of sync) with other documents. Primarily this concerns the interaction of TestGL and OpsGL. Ops has a number of checkpoints concerning how test materials are contributed, what criteria must be followed and so on. So we made an effort not to re-include that here. Many of these do not have priorities assigned yet either. Finally we tried to steer clear of the "this is the right way to write a test suite" and instead focus on the "this is what a good test suite should be composed of". Lynne's new checkpoint for Issue 107 (that's just been posted would fit nicely in to the new G1 or G2. If nothing else it can server as a basis for discussion on Monday, is this a better organization than before and are we missing any thing. Thanks, Peter Outline of new guidelines: G1 - high level functional analysis of spec to determine strategy of test development. (was G2-G3) - combine 2.1 and 2.2 in to one checkpoint analyze the specification and determine how to structure test materials. determine what testing areas the specification is composed of. - 3.1 determine how to cover each area. Is only one approach going to be used or will there be more than one. - 1.10 develop user scenarios for the specification. (move 3.2 to ex tec or some such or as descriptive text.) G2 - deep analysis the spec and extract what needs to be tested and how (was G1) - extract assertions/normative language and tag according to category - Using categories provided by patrick In other words rather than having explicit checkpoints for each required, optional, depreciated, discretionary, under-defined, etc asserts, have a checkpoint that has all asserts or normative language extracted and then grouped by category. It's the same basic idea but it takes 4 checkpoints into one. - determine level of coverage (depth/breadth) and priority of coverage (what's most important). G3 - test management system (was part of G4) - Have a test management system - the system must support meta data like documentation, pass/fail criteria, coverage level, state of test, association back to asserts, and dimensions of variability. G4 - test suite/cases development (was G6) - prototype test framework (6.2) (ops deals with submission and criteria for acceptable submissions) G5 - test execution. (was part of G4) - The metadata from the management system must provide sufficient information to allow tests to be executed in a predicable and consistent way. - automation is encouraged (cross platform stuff goes to ExTec.) - system must allow for tests to be filtered based on metadata criteria. (This is where the DOV really enters in for a test suite rather than in the analysis part. In the analysis you want to identify them but here is where you really care) - the test execution process should save output from tests for analysis (pass/fail, what cases failed, and logs if relevant). G6 - result reporting (G5) - must support result reporting (5.1) - it should create reports as a unified package (for example like a web page (5.3)) - It must indicate what passed and failed. - It should be automated if possible - It should supported filtering and comparison of results G7 - conformance testing. (G7) - encourage vendors to use the test suite. (p1) - encourage vendors to publish results (p3)
Received on Monday, 10 March 2003 01:10:36 UTC