Bug 207038

<rdar://problem/59058757>

This test seems extremely flaky. This test is slowing down EWS ios-wk2 queue and sometimes causes EWS to blame random patches for this test failure. https://ews-build.webkit.org/#/builders/24/builds/9793 - "Found 1 pre-existing test failure: fast/hidpi/image-srcset-svg-canvas-2x.html" https://ews-build.webkit.org/#/builders/24/builds/9786 - "Found flaky test: fast/hidpi/image-srcset-svg-canvas-2x.html" https://ews-build.webkit.org/#/builders/24/builds/9779 - "Found 1 new test failure: fast/hidpi/image-srcset-svg-canvas-2x.html" We should consider updating test expectations meanwhile while it is being investigated.

I am working on reproducing this failure with test lists.

This is not easily reproducing with a test list

Marked as failing in https://trac.webkit.org/changeset/255985/webkit Sense this will end up skipping the test on EWS I expect another test to begin failing in its place.

The next test in line did end up failing. Now fast/hidpi/image-srcset-svg-canvas.html is failing on iOS-ews

Created attachment 390086 [details] Test list from bot This is the list of tests that run before the failing test on the test runner. most likely one of these tests is setting a bad state

I am able to reproduce the failure by using this command on the news bot: run-webkit-tests --no-build --no-show-results --no-new-test-results --clobber-old-results --release --ios-simulator --child-processes 1 compositing/webgl/webgl-reflection.html fast/hidpi/image-srcset-svg2.html fast/hidpi/image-srcset-svg-canvas.html remove either of the first two tests and it will no longer fail. So something in the interaction of these three tests is causing this cascading failure: compositing/webgl/webgl-reflection.html fast/hidpi/image-srcset-svg2.html fast/hidpi/image-srcset-svg-canvas.html

This is the first news run that shows the original test failure from fast/hidpi/image-srcset-relative-svg-canvas-2x.html https://ews-build.webkit.org/#/builders/24/builds/9513

(In reply to Truitt Savell from comment #6) > The next test in line did end up failing. Now fast/hidpi/image-srcset-svg-canvas.html is failing on iOS-ews Filed https://bugs.webkit.org/show_bug.cgi?id=207851 to track that.

Changed expectations to force a skip of compositing/webgl/webgl-reflection.html on EWS instead of the two test fast/hidpi/image-srcset-relative-svg-canvas-2x.html and fast/hidpi/image-srcset-svg-canvas-2x.html. https://trac.webkit.org/changeset/257091/webkit

*** Bug 207851 has been marked as a duplicate of this bug. ***

Created attachment 394772 [details] Patch

With disabling the canvas accelerated drawing, the iOS EWS passed the all the tests from the first time. I will resubmit the patch one more time and see how the bots behave.

Created attachment 394779 [details] Patch

(In reply to Said Abou-Hallawa from comment #14) > I will resubmit the patch one more time and see how the bots behave. Thanks for the patch. I have pressed the 'Rebuild' button 5 times on one of the successful ios-wk2 build (https://ews-build.webkit.org/#/builders/24/builds/14055). Let's see the results in following ones: https://ews-build.webkit.org/#/buildrequests/411396 https://ews-build.webkit.org/#/buildrequests/411397 https://ews-build.webkit.org/#/buildrequests/411398 https://ews-build.webkit.org/#/buildrequests/411399 https://ews-build.webkit.org/#/buildrequests/411400

Created attachment 394795 [details] Patch

Created attachment 394804 [details] Patch

Created attachment 394805 [details] Patch

Created attachment 394806 [details] Patch

Created attachment 394807 [details] Patch

Created attachment 394808 [details] Patch

I have again pressed 'Rebuild' button 6 times on one of the recent successful ios-wk2 build (https://ews-build.webkit.org/#/builders/24/builds/14091) for patch 394808. Let's see result in following builds: https://ews-build.webkit.org/#/builders/24/builds/14094 https://ews-build.webkit.org/#/builders/24/builds/14095 https://ews-build.webkit.org/#/builders/24/builds/14096 https://ews-build.webkit.org/#/builders/24/builds/14097 https://ews-build.webkit.org/#/builders/24/builds/14098 https://ews-build.webkit.org/#/builders/24/builds/14099

iOS EWS bots succeeded in processing the latest patch from the first time with the 6 uploads I did and with the 6 retries that Aakash did.

Created attachment 394820 [details] Patch

Comment on attachment 394820 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=394820&action=review > LayoutTests/ChangeLog:9 > + Disable the canvas scaling anti-aliasing by applying the CSS property > + "image-rendering: pixelated;" to the <img> and the <canvas> elements. Did you figure out what the state leaking form compositing/webgl/webgl-reflection.html is? With the root cause not fixed, it seems very likely that other tests will be affected.

Comment on attachment 394820 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=394820&action=review >> LayoutTests/ChangeLog:9 >> + "image-rendering: pixelated;" to the <img> and the <canvas> elements. > > Did you figure out what the state leaking form compositing/webgl/webgl-reflection.html is? With the root cause not fixed, it seems very likely that other tests will be affected. No I have not. But I looked at the canvas drawing code carefully and I did not see anything that can be leaked from a previous page. Most likely the leaking in underlaying simulator frameworks. I think we should avoid testing the antialiasing in all the hidpi tests. So I think this patch is worthy landing. We can keep the bug open and continue investigating but let's land this patch and see the results. And hopefully this will shorten iOS EWS bots processing time.

Comment on attachment 394820 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=394820&action=review >>> LayoutTests/ChangeLog:9 >>> + "image-rendering: pixelated;" to the <img> and the <canvas> elements. >> >> Did you figure out what the state leaking form compositing/webgl/webgl-reflection.html is? With the root cause not fixed, it seems very likely that other tests will be affected. > > No I have not. But I looked at the canvas drawing code carefully and I did not see anything that can be leaked from a previous page. Most likely the leaking in underlaying simulator frameworks. > > I think we should avoid testing the antialiasing in all the hidpi tests. So I think this patch is worthy landing. We can keep the bug open and continue investigating but let's land this patch and see the results. And hopefully this will shorten iOS EWS bots processing time. I am convinced.

Committed r259174: <https://trac.webkit.org/changeset/259174> All reviewed patches have been landed. Closing bug and clearing flags on attachment 394820 [details].