Bug 191688 – [GTK] Several compositing tests failing since r238090 "Make compositing updates incremental"

• Component: WebKitGTK
• Status: RESOLVED FIXED

[Miguel Gomez]

Several compositing related tests started failing since r238090:

compositing/backing/no-backing-for-clip-overlap.html [ Failure ]
compositing/backing/no-backing-for-clip.html [ Failure ]
compositing/backing/no-backing-for-perspective.html [ Failure ]
compositing/layer-creation/overlap-animation.html [ Failure ]
compositing/layer-creation/overlap-transforms.html [ Failure ]

Curiously, all of this tests produce almost the expected outcome except that some of the layers have the drawsContent flag set to 1 when it's not expected. These are the diffs:

compositing/backing/no-backing-for-clip-overlap.html:
https://build.webkit.org/results/GTK%20Linux%2064-bit%20Release%20(Tests)/r238220%20(8857)/compositing/backing/no-backing-for-clip-overlap-diff.txt

compositing/backing/no-backing-for-clip.html:
https://build.webkit.org/results/GTK%20Linux%2064-bit%20Release%20(Tests)/r238220%20(8857)/compositing/backing/no-backing-for-clip-diff.txt

compositing/backing/no-backing-for-perspective.html:
https://build.webkit.org/results/GTK%20Linux%2064-bit%20Release%20(Tests)/r238220%20(8857)/compositing/backing/no-backing-for-perspective-diff.txt

compositing/layer-creation/overlap-animation.html:
https://build.webkit.org/results/GTK%20Linux%2064-bit%20Release%20(Tests)/r238220%20(8857)/compositing/layer-creation/overlap-animation-diff.txt

compositing/layer-creation/overlap-transforms.html:
https://build.webkit.org/results/GTK%20Linux%2064-bit%20Release%20(Tests)/r238220%20(8857)/compositing/layer-creation/overlap-transforms-diff.txt

[Miguel Gomez]

compositing/overflow/clip-descendents.html [ Failure ]

This seems to be related to r230890 as well, but the diff is different from the other ones:

--- /home/slave/webkitgtk/gtk-linux-64-release-tests/build/layout-test-results/compositing/repaint/absolute-painted-into-composited-ancestor-expected.txt
+++ /home/slave/webkitgtk/gtk-linux-64-release-tests/build/layout-test-results/compositing/repaint/absolute-painted-into-composited-ancestor-actual.txt
@@ -1,5 +1,4 @@
 (repaint rects
-  (rect 0 0 200 200)
   (rect 300 0 200 200)
 )
 (GraphicsLayer

[Miguel Gomez]

And (In reply to Miguel Gomez from comment #1)
> compositing/overflow/clip-descendents.html [ Failure ]
> 
> This seems to be related to r230890 as well, but the diff is different from
> the other ones:

Sorry, this is failing due to the drawsContent flag as well, this is the diff

--- /home/slave/webkitgtk/gtk-linux-64-release-tests/build/layout-test-results/compositing/overflow/clip-descendents-expected.txt
+++ /home/slave/webkitgtk/gtk-linux-64-release-tests/build/layout-test-results/compositing/overflow/clip-descendents-actual.txt
@@ -36,6 +36,7 @@
           (offsetFromRenderer width=20 height=10)
           (position 48.00 230.00)
           (bounds 60.00 70.00)
+          (drawsContent 1)
           (children 1
             (GraphicsLayer
               (offsetFromRenderer width=20 height=10)
@@ -55,6 +56,7 @@
           (offsetFromRenderer width=20 height=10)
           (position 240.00 230.00)
           (bounds 60.00 70.00)
+          (drawsContent 1)
           (children 1
             (GraphicsLayer
               (offsetFromRenderer width=20 height=10)

[Miguel Gomez]

compositing/repaint/absolute-painted-into-composited-ancestor.html [ Failure ]

This is the one that happens also due to r238090 but is not related to the drawsContent flag. This is the diff

--- /home/slave/webkitgtk/gtk-linux-64-release-tests/build/layout-test-results/compositing/repaint/absolute-painted-into-composited-ancestor-expected.txt
+++ /home/slave/webkitgtk/gtk-linux-64-release-tests/build/layout-test-results/compositing/repaint/absolute-painted-into-composited-ancestor-actual.txt
@@ -1,5 +1,4 @@
 (repaint rects
-  (rect 0 0 200 200)
   (rect 300 0 200 200)
 )
 (GraphicsLayer

[Simon Fraser (smfr)]

(In reply to Miguel Gomez from comment #1)
> compositing/overflow/clip-descendents.html [ Failure ]
> 
> This seems to be related to r230890 as well, but the diff is different from
> the other ones:
> 
> ---
> /home/slave/webkitgtk/gtk-linux-64-release-tests/build/layout-test-results/
> compositing/repaint/absolute-painted-into-composited-ancestor-expected.txt
> +++
> /home/slave/webkitgtk/gtk-linux-64-release-tests/build/layout-test-results/
> compositing/repaint/absolute-painted-into-composited-ancestor-actual.txt
> @@ -1,5 +1,4 @@
>  (repaint rects
> -  (rect 0 0 200 200)
>    (rect 300 0 200 200)
>  )
>  (GraphicsLayer

I changed repaint stuff in r237942, but that should have added repaints, not removed them.

[Michael Catanzaro]

(In reply to Miguel Gomez from comment #0)
> Curiously, all of this tests produce almost the expected outcome except that
> some of the layers have the drawsContent flag set to 1 when it's not
> expected.

Probably unrelated, but reminds me of bug #187268, where we keep backing store attached when it shouldn't be....

[Diego Pino]

These tests are passing since 275045@main ([GTK] Enable `forceCompositingMode` by default).

https://results.webkit.org/?platform=GTK&suite=layout-tests&test=compositing%2Fbacking%2Fno-backing-for-clip-overlap.html

After updating test expectations in 279290@main, there are no more tests referring to this bug. Closing bug.