- From: Gérard Talbot <www-style@gtalbot.org>
- Date: Fri, 01 Sep 2017 16:52:56 -0400
- To: W3C www-style mailing list <www-style@w3.org>
> 2. Test Metadata > https://github.com/w3c/csswg-drafts/issues/1730 It seems to me that CSSWG people are focusing on the wrong things, otherwise over-focusing on minor things. - Some tests are wrong, incorrect, bad; among them, some could be rehabilitated, some can not - Some tests are imprecise - Many tests are not streamlined, compact, minimized: eg unneedlessly over-contained, unneedlessly using abs. or rel positioning, with over-defined rules or extraneous declarations, with css 0-resets, etc - A minority of tests are too long and would benefit from being split into smaller tests. Then, such smaller tests would become easier to read, examine, understand. As many specifications become more complex and more sophisticated, minimization of tests should be a requirement and should become increasingly important - Many tests are missing a clear description of what the tests themselves are trying to test, of what they are claiming to be testing, are targeting, verifying exactly. Title text may not be helpful or descriptive, filename may not be best or too general, useful comments in the code might be absent - A huge majority of tests do not reuse already created and available reference files: this increases test suites complexity, increase unneedlessly number of files and duplicates efforts, etc. - A minority of tests can never fail - A minority of tests can generate a false positive or a false negative; sometimes, a browser's own flaws (bugs or lack of support) make this impossible to overcome for a reftest but sometimes it is possible to overcome - etc... Removing or maintaining the rel=help links for tests as a requirement will not improve nor compensate adequately any of the above. Requiring a rel=help link in all tests will not necessarly give a reliable indication of (sufficient or insufficient) test coverage. Requiring a rel=help link in all tests will not necessarly reveal if tests are needed for some sections or if there are already many tests in some other sections. The rel=help link for tests may be as useful as the ISBN bar code and the Dewey classification code are for a librarian and to a library/bookstore... that is if the ISBN bar code and the Dewey classification code are correct and precise to begin with. My point is: a missing or not, a good (correct, adequate) or bad (incorrect, wrong) Dewey classification code still does not (and will never) indicate at all if a book is a good (intrinsic value, worthiness), reliable, trustworthy, relevant book on a subject. Gérard
Received on Friday, 1 September 2017 20:53:22 UTC