1287066 - Specific styling causes graphic corruption when I drag/scroll fake scrollbar

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[arni2033]

Attachment: testcase 1 - graphics corruption in blue scrollbar.zip

>>>   My Info:   Win7_64, Nightly 49, 32bit, ID 20160526082509
STR_1:
1. Download and extract attached "testcase 1" in a folder with a short full name
2. Open "2016.04.20 00-40-17 Реклама ВКонтакте.htm" in Firefox
3. Hover mouse over the blue scrollbar (selector ".wcommunity_scrollbar_inner")
4. Hold left mouse button
5. Slowly move mouse down by ~100px without releasing left mouse button

AR:  Scrollbar thumb poorly redraws: displays many transparent lines
ER:  Scrollbar should be rendered in a normal way - solid blue (or lightblue)


STR_2:
1. Download and extract attached "testcase 1" in a folder with a short full name
2. Open "2016.04.20 00-40-17 Реклама ВКонтакте.htm" in Firefox
3. Hover mouse over the scrollable area with blue scrollbar (selector ".wcommunity_scrollbar_inner")
4. Rotate mouse wheel down several times, then up several times

AR:  Scrollbar doesn't redraw correctly: parts of scrollbar thumb are drawn on empty place
ER:  Scrollbar thumb should be drawn only on it's actual position

Notes:
 1) It's better reproducible on HiDPI, e.g. pref layout.css.devPixelsPerPx set to 1.25
 2) The page shows some prompts after a while which temporarily break the testcase (remove graphic
    corruption). Probably this bug only manifests with specific layerization, I don't know.

This is regression from bug 1258558. Regression range:
> https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=826afa1625f49803d32acfaa5db5add1bede9d34&tochange=1cc3d0a27d30363a6e3e3727d67fc05ef758b826

Comment 2

[Astley Chen (inactive)]

Per reporter's comment:

This is regression from bug 1258558. Regression range:
> https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=826afa1625f49803d32acfaa5db5add1bede9d34&tochange=1cc3d0a27d30363a6e3e3727d67fc05ef758b826

Comment 3

[Benoit Girard (:BenWa)]

I can't reproduce on Mac Nightly 50.0a1 (2016-07-13). Trying more similar configs...

Comment 4

[Benoit Girard (:BenWa)]

I see the scrollbar thumb fall out of sync in this testcase and filed bug 1287147 for it.

Comment 5

[Benoit Girard (:BenWa)]

Can you please include your about:support information? In particular I want to see what rendering options you're using on Windows. I can't reproduce this on windows either. It must be specific to your rendering options.

Comment 6

[arni2033]

Attachment: Troubleshooting Information 2016.07.15 20-38-31.xht

Comment 7

[arni2033]

I just double-checked the regression range - no mistake here.
Some more notes:
3) It happens during opacity transition, so in STR_2 it's better: {1} not to hover scrollbar itself
  {2} do several "scroll sessions", 1 second each   {3} make 2s pauses between "scroll sessions"
4) It also happens with HWA disabled
5) STR_1 isn't reproducible on normal DPI (layout.css.devPixelsPerPx set to 1.0)
   Actually, DPI level in my OS is set to 1.25, maybe this explains Note 1.
> screencast: https://dl.dropboxusercontent.com/s/jfkexheysnk7p97/bug%201287066%20comment%207.webm

Comment 8

[Benoit Girard (:BenWa)]

Have you been able to reproduce this using the latest nightly build? I've been trying different versions with APZ+D3D11+D2D and I'm still not able to reproduce.

The regression range would explain this kind of bug. But being able to reproduce would really help in understanding why there's a mismatch between empty region equality.

Comment 9

[Benoit Girard (:BenWa)]

Ok, I can reproduce on Jeff' machine but not mine. I'm unsure what the environment difference is.

Comment 10

[arni2033]

Ok, that's great. BTW if I set layers.prefer-d3d9 -> true, the bug disappears, but only with HWA

Comment 11

[Sylvestre Ledru [:Sylvestre]]

Benoit, do you have any hope that this bug is going to be fixed for 48? Thanks

Comment 12

[Milan Sreckovic [:milan] (needinfo for best results)]

We'll try to get the solution in time for 48, then we can decide if the uplift is too risky.

Comment 13

[Benoit Girard (:BenWa)]

We can 'likely' have a solution which is to backout the patch causing a regression and changing how the region builder works. I'd like to understand the issue a bit better before doing that.

Comment 14

[Benoit Girard (:BenWa)]

Found the culprit:
https://dxr.mozilla.org/mozilla-central/rev/4c05938a64a7fde3ac2d7f4493aee1c5f2ad8a0a/gfx/layers/Layers.h#870

The correctness here depends on IsEqualEdges and causes this bug otherwise.

Comment 15

[Benoit Girard (:BenWa)]

Attachment: Bug 1287066 - Restore Layer::SetVisibleRegion behavior to fix invalidation bug.

Review commit: https://reviewboard.mozilla.org/r/65674/diff/#index_header
See other reviews: https://reviewboard.mozilla.org/r/65674/

Comment 16

[Benoit Girard (:BenWa)]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=805217c214ea

Comment 17

[Benoit Girard (:BenWa)]

Comment on attachment 8773037 [details]
Bug 1287066 - Restore Layer::SetVisibleRegion behavior to fix invalidation bug.

Review request updated; see interdiff: https://reviewboard.mozilla.org/r/65674/diff/1-2/

Comment 18

[Benoit Girard (:BenWa)]

Comment on attachment 8773037 [details]
Bug 1287066 - Restore Layer::SetVisibleRegion behavior to fix invalidation bug.

Review request updated; see interdiff: https://reviewboard.mozilla.org/r/65674/diff/2-3/

Comment 19

[Markus Stange [:mstange]]

Comment on attachment 8773037 [details]
Bug 1287066 - Restore Layer::SetVisibleRegion behavior to fix invalidation bug.

https://reviewboard.mozilla.org/r/65674/#review62738

r+ if you change the commit message, and assuming you have tested that this fixes the bug.

Please also file a bug for finding out why this fixes things and add a comment that points to that bug.

Comment 20

[Benoit Girard (:BenWa)]

Comment on attachment 8773037 [details]
Bug 1287066 - Restore Layer::SetVisibleRegion behavior to fix invalidation bug.

Review request updated; see interdiff: https://reviewboard.mozilla.org/r/65674/diff/3-4/

Comment 21

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/67186c2d647e
Restore Layer::SetVisibleRegion behavior to fix invalidation bug. r=mstange

Comment 22

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/67186c2d647e

Comment 23

[Benoit Girard (:BenWa)]

Comment on attachment 8773037 [details]
Bug 1287066 - Restore Layer::SetVisibleRegion behavior to fix invalidation bug.

Approval Request Comment
[Feature/regressing bug #]: bug 1258558
[User impact if declined]: Corrupted rendering in rare cases (see this bug)
[Describe test coverage new/current, TreeHerder]: This impact general rendering which has extensive coverage in reftest.
[Risks and why]: Fairly low, we're disabling an optimization for a very specific case. This should be lower risk than backing out bug 1258558 as well.
[String/UUID change made/needed]: None

Comment 24

[Sylvestre Ledru [:Sylvestre]]

The next version is the RC, we do want this patch that late?

Comment 25

[Benoit Girard (:BenWa)]

Given that AFAIK this is the only test case that can reproduce this it's probably wise to take into aurora but not beta.

Comment 26

[Gerry Chang [:gchang]]

Comment on attachment 8773037 [details]
Bug 1287066 - Restore Layer::SetVisibleRegion behavior to fix invalidation bug.

This patch fixes a scrollbar regression but it's too late for 48 beta RC. Take it in aurora.

Comment 27

[Carsten Book [:Tomcat]]

 https://hg.mozilla.org/releases/mozilla-aurora/rev/517663b29d7f