Bug 277109 – REGRESSION (281042@main): [ macOS ] 2x tiled-drawing/scrolling/overflow/overflow-scrolled-* is flaky failing.

• Component: Scrolling
• Status: RESOLVED FIXED

[Anfernee Viduya]

tiled-drawing/scrolling/overflow/overflow-scrolled-down-tile-coverage.html 
tiled-drawing/scrolling/overflow/overflow-scrolled-up-tile-coverage.html 
are flaky failing. 

HISTORY: 
https://results.webkit.org/?platform=mac&style=debug&style=release&suite=layout-tests&suite=layout-tests&test=tiled-drawing%2Fscrolling%2Foverflow%2Foverflow-scrolled-down-tile-coverage.html&test=tiled-drawing%2Fscrolling%2Foverflow%2Foverflow-scrolled-up-tile-coverage.html&version_name=Sonoma&version_name=Ventura

DIFF:
tiled-drawing/scrolling/overflow/overflow-scrolled-down-tile-coverage.html 
@@ -44,12 +44,12 @@
                   (bounds 385.00 3000.00)
                   (drawsContent 1)
                   (visible rect 0.00, 1800.00 385.00 x 512.00)
-                  (coverage rect 0.00, 1544.00 897.00 x 1024.00)
+                  (coverage rect 0.00, 1287.99 385.00 x 1024.01)
                   (intersects coverage rect 1)
                   (contentsScale 1.00)
-                  (tile cache coverage 0, 1536 385 x 1464)
+                  (tile cache coverage 0, 1024 385 x 1536)
                   (tile size 512 x 512)
-                  (top left tile 0, 3 tiles grid 1 x 3)
+                  (top left tile 0, 2 tiles grid 1 x 3)
                   (in window 1)
                 )
               )
tiled-drawing/scrolling/overflow/overflow-scrolled-up-tile-coverage.html
@@ -44,12 +44,12 @@
                   (bounds 385.00 3000.00)
                   (drawsContent 1)
                   (visible rect 0.00, 200.00 385.00 x 512.00)
-                  (coverage rect 0.00, 0.00 897.00 x 1024.00)
+                  (coverage rect 0.00, 200.00 385.00 x 1024.01)
                   (intersects coverage rect 1)
                   (contentsScale 1.00)
-                  (tile cache coverage 0, 0 385 x 1024)
+                  (tile cache coverage 0, 0 385 x 1536)
                   (tile size 512 x 512)
-                  (top left tile 0, 0 tiles grid 1 x 2)
+                  (top left tile 0, 0 tiles grid 1 x 3)
                   (in window 1)
                 )
               )


LINK: 
https://build.webkit.org/results/Apple-Sonoma-Debug-WK2-Tests/281341@main%20(2070)/tiled-drawing/scrolling/overflow/overflow-scrolled-down-tile-coverage-diff.txt
https://build.webkit.org/results/Apple-Sonoma-Debug-WK2-Tests/281341@main%20(2070)/tiled-drawing/scrolling/overflow/overflow-scrolled-up-tile-coverage-diff.txt

DESCRIPTION: 
In 281042@main changes were made that were around WebCore::GraphicsLayer::setTileCoverage

[Radar WebKit Bug Importer]

<rdar://problem/132528380>

[Anfernee Viduya]

REPRODUCIBILITY: 
I was able to reproduce this issue on ToT using command. 

run-webkit-tests 281196@main tiled-drawing/scrolling/overflow/overflow-scrolled-up-tile-coverage.html --iterations=10 --debug

run-webkit-tests 281196@main tiled-drawing/scrolling/overflow/overflow-scrolled-down-tile-coverage.html --iterations=10 --debug

[Anfernee Viduya]

Pull request: https://github.com/WebKit/WebKit/pull/31252

[EWS]

Test gardening commit 281407@main (6ead4f05a6e9): <https://commits.webkit.org/281407@main>

Reviewed commits have been landed. Closing PR #31252 and removing active labels.

[Anfernee Viduya]

Pull request: https://github.com/WebKit/WebKit/pull/32072

[EWS]

Test gardening commit 282142@main (1e16f12559d4): <https://commits.webkit.org/282142@main>

Reviewed commits have been landed. Closing PR #32072 and removing active labels.

[Matt Woodrow]

It looks like there’s at least two issues here.

One is that the requestAnimationFrame intervals can happen faster than 16ms, and `HistoricalVelocityData::velocityForNewData` ignores updates faster than, so we can sometimes end up with 0 velocity (for both tests).

The second is that the caller of `TileController::adjustTileCoverageWithScrollingVelocity` recomputes `m_velocity.lastUpdateTime` using `MonotonicTime::now`, and then this function computes the delta with another call. This results in a tiny, but not quite zero time delta. The future rect gets moved by this amount, which I don't think is correct. The 'down' test fails frequently, because this movement value varies a fair bit (racily, depending on the time delta) and affects the position of the final rect.

[Matt Woodrow]

Pull request: https://github.com/WebKit/WebKit/pull/45143

[EWS]

Committed 294876@main (ec36eefa5d9f): <https://commits.webkit.org/294876@main>

Reviewed commits have been landed. Closing PR #45143 and removing active labels.