- From: Gérard Talbot <www-style@gtalbot.org>
- Date: Sun, 22 May 2011 17:02:02 -0700
- To: "Linss, Peter" <peter.linss@hp.com>
- Cc: "timeless" <timeless@gmail.com>, "www-style@w3c.org" <www-style@w3c.org>
Le Dim 22 mai 2011 14:59, Linss, Peter a écrit : > > On May 22, 2011, at 1:31 PM, Gérard Talbot wrote: > >> >> Le Sam 21 mai 2011 23:29, timeless a écrit : >>> On Sun, May 22, 2011 at 9:13 AM, Peter Linss <peter.linss@hp.com> >>> wrote: >>> >>>> or perhaps an "only run untested mode", >>> >>> definitely >> >> >> I agree with and support this "only run untested mode" proposal. > > Ok, I'll add it to my todo list. > >>> >>>> if I add that mode I'll have the untested link activate it. I'm open >>>> to >>>> suggestions for improvements in the harness UI as well. >>> >>> my other problem is testing things which have a reference rendering... >>> >>> e.g. >>> http://test.csswg.org/harness/testcase?s=CSS21_DEV&f=html4&g=144&r=15&o=1 >> >> When I follow this link, I end up at >> >> http://test.csswg.org/suites/css2.1/nightly-unstable/html4/min-height-applies-to-002.htm > > A side effect of the intelligent test sequencing is that when sharing (or > revisiting) harness URLs that are index based (the 'r' value in the query > string above), you get whatever test happens to be at that index for the > engine you're using. You'll get a different test depending on your engine > and the current result data, so we're likely not looking at the same test. > > The index based URLs are a carry-over from old code (and yes, I know, bad > design, violates URL integrity, etc). I'm working on a fix that uses the > test names in the URL instead of the index. This way the URLs will always > get you to the same test, it's just the *next* test that will be different > based on engines and results... (this one's been bugging me for a while, > I'm also looking at using mod_rewrite to remove the query entirely.) I was > afraid that re-computing test indices would be a performance hit, but it > doesn't seem too bad (actually lets me speed up other code to compensate). > >>> >>> there are links to the reference, but they cause my browser to lose >>> its scroll position, so i can't view the top 1/3 of a test, click the >>> reference, check for flash, view the top 1/3, press <space> (shows >>> 2/3), click the test part, press space (shows 2/3), click the >>> reference (shows 2/3), check for flash. possibilities: >>> * being able to view things side by side >>> * abusing transparency to have a view w/ >>> -- test above reference >>> -- reference above test >> >> >> I understand this scroll position issue wrt reftests and agree that it >> is >> annoying: it makes the tester click (the scrollbar thumb) and/or scroll >> more. > > Understood. In general, tests that require scrolling are in violation of > our guidelines [1], so this isn't a problem that's high on my list. The > best solution that comes to my mind is using javascript in the harness to > swap the tests and references rather than a reload, but a design criteria > so far has been to not rely on script in the client. Another fix could be > to capture the scroll position and send it to the other page. scrollTop property during a scroll event on a element like a div is retrievable in almost all browsers: http://www.gtalbot.org/BugzillaSection/DocumentAllDHTMLproperties.html I have not tried on a <object> acting as an iframe though. > [1] http://www.w3.org/Style/CSS/Test/guidelines.html#the-long 1- The height of the <object> varies from one browser to another: that too can influence the amount of scrolling and the possibility of having to scroll to reach the [PASS][FAIL][Can not say][Skipped] buttons at the bottom. In Konqueror 4.6.3, the height of the <object data='...' type='text/html'> is considerably taller than the height of the same <object data='...' type='text/html'> in Firefox 4.0.1 2- .test { height: 70%; width: 100%; text-align: center; display: table-row; } .test p { height: inherit; display: table-cell; } object { height: 98%; (...) } a) "height: inherit" on a percentage value in nested cases may also be buggy in some browsers. And it is redefined in <link href='test_gecko.css' type='text/css' rel='stylesheet'> and <link href='test_webkit.css' type='text/css' rel='stylesheet'> from 70% to 90% b) Add to this: " CSS 2.1 does not define how the height of table cells and table rows is calculated when their height is specified using percentage values. " http://www.w3.org/TR/CSS21/tables.html#height-layout 3- There are a bunch redeclarations in the 3 stylesheets and sometimes more than once. They can be easily identified in some browsers: in webkit-based browsers, the declaration are striken and can be toggled by checking/unchecking a checkbox. Sometimes, they are needed though... but, after a quick look, sometimes not needed. 4- http://test.csswg.org/test_presto.css is defined for Opera browsers <link href='test_presto.css' type='text/css' rel='stylesheet'> but is 404 not found. 5- Since <h2 class="title"> has clear: both (btw, just clear: left; would suffice to clear the img), then <h3 class="testname"> does not need to be cleared with clear: both. Peter, I want you to know that overall I am extremely happy with the CSS 2.1 test suite as it is right now. Do not perceive my comments as a criticism. I think overall, the nr 1 problem with the test suite is the testcases themselves and the web authors' perspective[1]: there is now just too many issues right now with many testcases. I am for adding new testcases, better testcases to make the testsuite more complete, comprehensive. [1]: it would take me some time to elaborare on this... which I do not have right now. best regards, Gérard -- CSS 2.1 Test suite RC6, March 23rd 2011 http://test.csswg.org/suites/css2.1/20110323/html4/toc.html Contributions to CSS 2.1 test suite http://www.gtalbot.org/BrowserBugsSection/css21testsuite/ Web authors' contributions to CSS 2.1 test suite http://www.gtalbot.org/BrowserBugsSection/css21testsuite/web-authors-contributions-css21-testsuite.html
Received on Monday, 23 May 2011 00:02:35 UTC