Bug 194798 – [ Mac ] REGRESSION (r237587) Layout Test compositing/visible-rect/animated-from-none.html is flaky text diff failure

• Status: RESOLVED
• Resolution: FIXED
• Component: Tools / Tests

[Shawn Roberts]

The following layout test is flaky on Mac

compositing/visible-rect/animated-from-none.html

Probable cause:

Test started to become flaky after R237924. Prior revision passes. 

reproduced locally with:

run-webkit-tests compositing/visible-rect/animated-from-none.html --iterations 5000 -f --exit-after-n-failures=1 --no-retry-failures

Flakiness Dashboard:

https://webkit-test-results.webkit.org/dashboards/flakiness_dashboard.html#showAllRuns=true&tests=compositing%2Fvisible-rect%2Fanimated-from-none.html

Diff:

--- /Volumes/Data/slave/mojave-release-tests-wk2/build/layout-test-results/compositing/visible-rect/animated-from-none-expected.txt
+++ /Volumes/Data/slave/mojave-release-tests-wk2/build/layout-test-results/compositing/visible-rect/animated-from-none-actual.txt
@@ -35,6 +35,7 @@
                   (position -100.00 0.00)
                   (bounds 200.00 200.00)
                   (contentsOpaque 1)
+                  (transform [1.00 0.00 0.00 0.00] [0.00 1.00 0.00 0.00] [0.00 0.00 1.00 0.00] [0.00 0.00 0.00 1.00])
                   (visible rect 100.00, 0.00 100.00 x 200.00)
                   (coverage rect 100.00, 0.00 500.00 x 200.00)
                   (intersects coverage rect 1)

[Radar WebKit Bug Importer]

<rdar://problem/48181898>

[Wenson Hsieh]

(In reply to Shawn Roberts from comment #0)
> The following layout test is flaky on Mac
> 
> compositing/visible-rect/animated-from-none.html
> 
> Probable cause:
> 
> Test started to become flaky after R237924. Prior revision passes. 

I haven't investigated this to any depth yet, but it seems somewhat unlikely that <https://trac.webkit.org/r237924> would cause a layout test on macOS to become flaky.

[Wenson Hsieh]

Is it possible that another change (which might've landed around the same time as r237924) caused this?

[Shawn Roberts]

It is definitely possible. Partnering with another bot watcher they thought perhaps the Rect it was referring to was it attempting to draw a rectangle and it was off. 

237925 has a similar reference but I don't think anything had been changed in that regard.

[Shawn Roberts]

https://trac.webkit.org/changeset/237925/webkit

[Simon Fraser (smfr)]

The dumping code is:
    if (m_transform && !m_transform->isIdentity()) {
        ts << indent << "(transform ";
        ts << "[" << m_transform->m11() << " " << m_transform->m12() << " " << m_transform->m13() << " " << m_transform->m14() << "] ";
        ts << "[" << m_transform->m21() << " " << m_transform->m22() << " " << m_transform->m23() << " " << m_transform->m24() << "] ";
        ts << "[" << m_transform->m31() << " " << m_transform->m32() << " " << m_transform->m33() << " " << m_transform->m34() << "] ";
        ts << "[" << m_transform->m41() << " " << m_transform->m42() << " " << m_transform->m43() << " " << m_transform->m44() << "])\n";
    }


So this transform must be slightly non-identity. Figure out why!

[Antoine Quint]

I think the reason is that the animation's progress can be a very small number larger than 0. I don't know whether that is a bug or a shortcoming of the test.

[Antoine Quint]

I think for the Web Animations engine the right thing to use is the "ready" promise and not an "animationstart" event where the event delivery is async and progress may not be 0 at the time it's delivered.

[Antoine Quint]

Created attachment 367047 [details]
Patch

[Dean Jackson]

Comment on attachment 367047 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=367047&action=review

> LayoutTests/ChangeLog:11
> +        than 1 and yield an animated transform that isn't quite the identity matrix.

Don't you mean 0?

[Antoine Quint]

Committed r244082: <https://trac.webkit.org/changeset/244082>