Description: imported/w3c/web-platform-tests/css/css-text/text-transform/text-transform-capitalize-007.html The first failure I saw on the dashboard after a period of consistent passing was on 12/26/2021 at 245578@main. The regression range looks to be approximately 245578@main - 245569@main. 245569@main has to do with webkit-font-kerning CSS property, so perhaps that is it. History: https://results.webkit.org/?limit=50000&suite=layout-tests&test=imported%2Fw3c%2Fweb-platform-tests%2Fcss%2Fcss-text%2Ftext-transform%2Ftext-transform-capitalize-007.html&platform=mac&flavor=wk1&recent=false Image Diff: https://build.webkit.org/results/Apple-Monterey-Release-WK1-Tests/253769@main%20(5092)/imported/w3c/web-platform-tests/css/css-text/text-transform/text-transform-capitalize-007-diffs.html
<rdar://problem/99158592>
I have marked this test as a flaky image failure while this issue is investigated.
Test gardening commit 253797@main (20c9698d4800): <https://commits.webkit.org/253797@main> Reviewed commits have been landed. Closing PR #3679 and removing active labels.
Description: imported/w3c/web-platform-tests/css/css-text/text-transform/text-transform-capitalize-007.html ADDING: imported/w3c/web-platform-tests/css/css-text/text-transform/text-transform-capitalize-003.html imported/w3c/web-platform-tests/css/css-text/text-transform/text-transform-capitalize-014.html The best regression range I have been able to gather is 245578@main-245564@main. I have marked these tests as flaky image failures while this issue is investigated.
Test gardening commit 253799@main (b7671788c0d2): <https://commits.webkit.org/253799@main> Reviewed commits have been landed. Closing PR #3683 and removing active labels.
(In reply to Karl Rackler from comment #4) > The best regression range I have been able to gather is > 245578@main-245564@main. The link for this is: https://github.com/WebKit/WebKit/compare/42888f9d55e6863c9e1a8110cc164f81a63c1158...00261fe1276bda1900e7d3af2447e861c48542d2
The failure is because legacy and modern line layout produce different line breaking for the test content. While modern line layout produces the correct rendering (matching other browsers), legacy let the content overflow. So if anything this a progression :). However the reason why it is flaky is because: normally we run both .html and -expected.html with IFC, but in some cases the large text content comes in in multiple chunks and when that happens, we reach the cap on the "dynamic change allowed for IFC due to lack of partial layout" setting (see LineLayout::shouldSwitchToLegacyOnInvalidation). So when the content comes in in one large chunk, we run modern (on both .html and -expected) but when it comes in in chunks, we run one of the files with legacy -> fail. This gets fixed when we introduce partial layout to IFC (and remove that dynamic change cap).