Bug 286507 – [ iOS ] compositing/geometry/fixed-position-composited-page-scale-smaller-than-viewport.html is flakily failing.

• Component: Compositing
• Status: RESOLVED FIXED

[Yasmin Karimi]

compositing/geometry/fixed-position-composited-page-scale-smaller-than-viewport.html

This test is flakily failing on iOS. 

HISTORY:

https://results.webkit.org/?suite=layout-tests&test=compositing%2Fgeometry%2Ffixed-position-composited-page-scale-smaller-than-viewport.html

TEXT DIFF:

@@ -5,11 +5,12 @@
 
 PASS rectsBeforeStyleChange.length is 1
 PASS rectsBeforeStyleChange.length is rectsAfterStyleChange.length
-PASS rectsBeforeStyleChange[0].top is rectsAfterStyleChange[0].top
-PASS rectsBeforeStyleChange[0].left is rectsAfterStyleChange[0].left
+FAIL rectsBeforeStyleChange[0].top should be 47. Was 46.99999713897705.
+FAIL rectsBeforeStyleChange[0].left should be 74. Was 74.0000057220459.
 PASS rectsBeforeStyleChange[0].width is rectsAfterStyleChange[0].width
 PASS rectsBeforeStyleChange[0].height is rectsAfterStyleChange[0].height
 PASS successfullyParsed is true
+Some tests failed.
 
 TEST COMPLETE

DIFF URL:

https://build.webkit.org/results/Apple-iPadOS-18-Simulator-Release-WK2-Tests/289317@main%20(266)/editing/selection/ios/selection-moves-between-composited-layers-pretty-diff.html

REPRODUCIBILITY:

I was able to reproduce this issue on the iOS 18.2 simulator at ToT using the following command:

run-webkit-tests --ios-simulator compositing/geometry/fixed-position-composited-page-scale-smaller-than-viewport.html --iterations 15

[Radar WebKit Bug Importer]

<rdar://problem/143593144>

[EWS]

Test gardening commit 289373@main (99cc0d3dc30a): <https://commits.webkit.org/289373@main>

Reviewed commits have been landed. Closing PR #39523 and removing active labels.

[Ben Schwartz]

I suspect this regressed at 282170@main -- that was a pretty big structural change that touches code paths related to this test. I believe that's most likely causing a race condition in a layout tests setting.

[Jonathan Bedard]

This test appears to be an ImageDiff test, and I don't understand the original analysis from Yasmin.

Today, compositing/geometry/fixed-position-composited-page-scale-smaller-than-viewport.html only fails with ImageDiff, if you look at https://results.webkit.org/?suite=layout-tests&suite=layout-tests&test=compositing%2Fgeometry%2Ffixed-position-composited-page-scale-smaller-than-viewport.html. When you look into the diffs, it seems like we're not focusing correctly.

[Jonathan Bedard]

Removing the "Regression" tag because I think this is about taking a snapshot too quickly. I have a speculative fix, which I will upload, but I haven't gotten a chance to validate because I don't have an iOS Simulator handy right this moment.

[Jonathan Bedard]

Pull request: https://github.com/WebKit/WebKit/pull/55258

[EWS]

Committed 305376@main (6c3e4e8af893): <https://commits.webkit.org/305376@main>

Reviewed commits have been landed. Closing PR #55258 and removing active labels.