1071686 - Reproducible APZC failure and 300ms scroll delay

Status: RESOLVED FIXED

(Core :: Panning and Zooming, defect)

Description

[Benoit Girard (:BenWa)]

STR:
1) Shift+reload https://bugzilla.mozilla.org/show_bug.cgi?id=1071156 (I believe it works there because the CC box is scrollable)
2) Position the mouse to the right of the CC box, outside.
3) Fling scroll the page down.

If you have trouble reproducing try to minimize the time between the page load and the fling.

Comment 1

[Kartikaya Gupta (email:[e-mail])]

The APZ failure box only flashes on for a few frames for me. I think that might just be because there's a period of time between when we dispatch the repaint request with the displayport and the repaint happens and re-layerizes the content to scroll. During that time the APZ is still updating the transform on the scroll info layer but it doesn't have content so the red box flashes.

Comment 2

[Kartikaya Gupta (email:[e-mail])]

(Also just for completeness, you need to enable the FPS counter pref to see the red square)

Comment 3

[Kartikaya Gupta (email:[e-mail])]

Attachment: Layer dumps during red square

Here are the relevant layer dumps. The first one in the file is the steady state before scrolling starts. In this case there is a scrollinfo layer with scrollid=5 for the main content with no APZ transform, so no red square. It also doesn't have a displayport (presumably because the initial-displayport code at [1] is ifdef'd to b2g). In the last layer dump in the file, the scrollable content repaint has taken effect, so there is a thebes layer for it, and the APZ transform is applied to that. All the layer dumps in between show the red square because the APZ transform is applied to the container layer but the repaint and relayerization hasn't happened yet.

I suspect we can fix this by expanding the ifdef'd code to run on OS X as well.

[1] http://mxr.mozilla.org/mozilla-central/source/layout/base/nsLayoutUtils.cpp?rev=1c9e801f66aa#2772

Comment 4

[Kartikaya Gupta (email:[e-mail])]

I verified that removing the #ifdef (so the code is used on all platforms) fixes this problem. Markus has a patch on bug 1009786 that includes this change, so I'll make this block on that. We can dupe it over once that lands.

Comment 5

[Kartikaya Gupta (email:[e-mail])]

Attachment: Patch

In the interests of getting this fixed sooner (and reduce the risk in bug 1009786), I'd like to land this part separately from bug 1009786. Any objections?

Comment 6

[Markus Stange [:mstange]]

Comment on attachment 8504755 [details] [diff] [review]
Patch

No objections, but you'll probably also need the reftest fuzziness annotations from attachment 8494800 [details] [diff] [review].

Comment 7

[Kartikaya Gupta (email:[e-mail])]

I did a try push to check [1]; so far the reftests on desktop platforms are green. Maybe the fuzziness was coming from a different part of your patch?

[1] https://tbpl.mozilla.org/?tree=Try&rev=30d9b096ea77

Comment 8

[Markus Stange [:mstange]]

Oh, right, your patch doesn't cause them because it has no effect on platforms without APZ.

Comment 9

[Kartikaya Gupta (email:[e-mail])]

https://hg.mozilla.org/integration/fx-team/rev/1a5153ae014f

Comment 10

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/1a5153ae014f