- From: Hiroyuki Ikezoe <hiikezoe@mozilla-japan.org>
- Date: Tue, 16 Jun 2015 11:04:06 +0900
- To: Brian Birtles <bbirtles@mozilla.com>, Michael Blain <mpb@chromium.org>, public-web-perf <public-web-perf@w3.org>
Hello all, Thanks for the test. The test is very helpful to me! Just a quick note about the tests running on Firefox with my current implementation of the Frame Timing API. (with removal vendor prefix) test_frame_on_frame_timing_buffer_full.html PASS 2/2 test_frame_on_frame_timing_buffer_full_not_called.html PASS 1/1 test_frame_timing_clear_frame_timings.html PASS 4/4 test_frame_timing_composite.html PASS 21/21 test_frame_timing_cross_origin_iframe.html PASS 3/3 test_frame_timing_css_animation.html PASS 5/5 test_frame_timing_iframe.html FAIL 9/25 test_frame_timing_offpage_iframe.html FAIL 5/7 test_frame_timing_render.html FAIL 9/21 test_frame_timing_set_buffer_size.html PASS 2/2 There might be some bugs in my implementation but most of failures are caused by timing issue as Brian noted. -- hiro On 2015年06月16日 10:45, Brian Birtles wrote: > On 2015/06/16 6:55, Michael Blain wrote: >> Hi all, >> Here are a few Web Platform tests for the Frame TIming API. >> >> I understand that Mozilla, Chrome and IE have all pretty much decided to >> gate Frame Timing on the Performance Observer spec, so some of the >> details of the calls will change, but this is what we've been testing >> Chrome's implementation against. >> >> Any feedback would be appreciated! > > Thanks for contributing these. > > Please bear in mind that we would like to use these tests on test > machines which are often under considerable load and have limited > resources. > > For example, to avoid intermittent failures, when writing animation > tests we normally assume a worst-case scenario where frames might be as > much as 100s apart. > > From glancing at these tests it seems we're assuming a 5ms difference > between triggering a render and the time recorded on the render entry. > Is that correct? If so it would be quite unsuitable for the kind of test > environment we (and I suspect others) use. I'd suggest these tests > should be written in such a way that they don't make any assumptions > about timeliness but just test, for example, the order/existence of > events, their relative times etc. > > On a further note, the description of startTime seems to have been > (incorrectly?) reverted.[1] > > The previous text that spelled out that it corresponds to the time > passed to rAF callbacks for that frame made much more sense.[2] I bring > this up here because the test appears to be following the updated text > but that seems undesirable.[3] > > Best regards, > > Brian > > [1] > https://github.com/w3c/frame-timing/commit/cf7fcf4985fb5e7475e0fe4c2b82d4b795d8e73e > > [2] As introduced here: https://github.com/w3c/frame-timing/pull/10 > [3] As discussed here: https://github.com/w3c/frame-timing/issues/9 > -- Hiroyuki Ikezoe <hiikezoe@mozilla-japan.org>
Received on Tuesday, 16 June 2015 13:38:50 UTC