Bug 52720 – REGRESSION (r75987): Assertion failure in WebView::scrollBackingStore when scrolling page with composited content

• Status: RESOLVED
• Resolution: FIXED
• Component: New Bugs

[Adam Roben (:aroben)]

REGRESSION (r75987): Assertion failure in WebView::scrollBackingStore when scrolling page with composited content

[Adam Roben (:aroben)]

Created attachment 79416 [details]
Patch

[Adam Roben (:aroben)]

I just realized that I've probably affected the pixel results by adding layoutTestController.display() to these tests. That function puts a gray overlay over everything. Maybe that's OK, or maybe I should write a new dumpAsText test whose whole purpose is to trigger this assertion?

[Simon Fraser (smfr)]

Comment on attachment 79416 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=79416&action=review

> LayoutTests/ChangeLog:11
> +        This makes these tests trigger this bug: <http://webkit.org/b/52720> REGRESSION (r75987):
> +        Assertion failure in WebView::scrollBackingStore when scrolling page with composited content

But shouldn't keep a test that triggers the bug?

[Simon Fraser (smfr)]

... we keep ...

[Adam Roben (:aroben)]

Comment on attachment 79416 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=79416&action=review

>> LayoutTests/ChangeLog:11
>> +        Assertion failure in WebView::scrollBackingStore when scrolling page with composited content
> 
> But shouldn't keep a test that triggers the bug?

I guess I wasn't clear in the ChangeLog.

Changing the tests in this way makes them trigger the bug. Prior to this patch, the tests did not trigger the bug. After changing the tests, they do trigger the bug. Then I fixed the bug in WebKit. I'll try to come up with a better way of phrasing this. But as I mentioned above, perhaps I should be writing a new test instead of modifying these.

[Adam Roben (:aroben)]

(In reply to comment #2)
> I just realized that I've probably affected the pixel results by adding layoutTestController.display() to these tests. That function puts a gray overlay over everything.

Turns out this change didn't modify the pixel results. I guess the whole view gets repainted during the scroll, so the gray overlay doesn't show up in the final results.

I'll clarify the ChangeLog and post a new patch.

[Adam Roben (:aroben)]

Created attachment 79431 [details]
Teach WebView::scrollBackingStore about compositing mode

[Adam Roben (:aroben)]

Created attachment 79604 [details]
Teach WebView::scrollBackingStore about compositing mode

[Adam Roben (:aroben)]

Committed r76254: <http://trac.webkit.org/changeset/76254>