623852 - [OS X] [Hardware acceleration off] Websites flickering, partially dis- and reappearing, tabs missing

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[nikarlo]

User-Agent:       Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:2.0b9pre) Gecko/20110106 Firefox/4.0b9pre
Build Identifier: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.6; rv:2.0b9pre) Gecko/20110106 Firefox/4.0b9pre

Tabs are white (empty) when scrolling and when I select one of them sometimes doesn't work and still to last selected tab.
The same also with a new profile.

Reproducible: Always

Comment 1

[Henrik Skupin [:whimboo][⌚️UTC+2]]

Please run Firefox in Safe Mode (http://support.mozilla.com/en-US/kb/Safe Mode) to make sure no extension is causing this issue. Further please try to disable hardware acceleration in the advanced pane of the preferences. Would be nice if you could post the graphic card information from the Help | Troubleshooting page. Thanks.

Comment 2

[nikarlo]

In safe mode is the same.
This is my graph info:
Adapter Description0x22400,0x20400 Vendor ID0000 Device ID0000 Adapter RAM Adapter DriversDriver VersionDriver DateDirect2D EnabledfalseDirectWrite EnabledfalseWebGL RendererNVIDIA Corporation -- NVIDIA GeForce 7600 GS OpenGL Engine -- 2.1 NVIDIA-1.6.26GPU Accelerated Windows0/1

Machine is a osx86 with retail snow leopard

Comment 3

[Markus Stange [:mstange]]

I'm changing the summary to the one from bug 622901, which is the same issue.

Comment 6

[Markus Stange [:mstange]]

Bug 624118 comment 0 has good steps to reproduce and a screenshot.

Comment 8

[nikarlo]

How I can help you?
I must try to compile myself from a trunk?

Comment 9

[Mehdi Mulani [:mmm] (I don't check this)]

Ran into this problem and tested it on nightlies.
Showed up in 2011-01-03 but not in 2011-01-02.

A similar issue, a line between the title bar and tab bar flickered showed up in 2011-01-02 though.

Comment 10

[Markus Stange [:mstange]]

(In reply to comment #9)
> A similar issue, a line between the title bar and tab bar flickered showed up
> in 2011-01-02 though.

That's bug 621762.

Comment 11

[Markus Stange [:mstange]]

I've bisected the checkins in the range, this is the cause of the regression:
http://hg.mozilla.org/mozilla-central/rev/b7d1bb50e829

Comment 12

[Markus Stange [:mstange]]

The bug originates in PushGroupAndCopyBackground (called from BasicThebesLayer::Paint): the call to cairo_paint fails. After that the context gets marked as being in an error state and everything falls apart.

The problem seems to be that we're trying to use a surface as source which can't be used in that way. Specifically, the surface "s" is the one we've constructed in -[ChildView drawRect:inContext:] (called "targetSurface" there) - in other words, it's the window buffer. However, the cairo quartz backend can only use offscreen buffers ("bitmap contexts") as source surfaces.
I don't know if there's a way at all to read back the pixels we've drawn into the window.

Comment 14

[Robert O'Callahan (:roc) (email my personal email if necessary)]

I see.

I guess we'll need to predict in advance whether we'll need PushGroupAndCopyBackground and if we will, use a manual backbuffer.

Comment 15

[Robert O'Callahan (:roc) (email my personal email if necessary)]

This obviously blocks.

Comment 16

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Actually the thing to do is allow PushGroupAndCopyBackground to fall back to just PushGroup if we can't copy the background. That's the conservative fix for the regression here.

Comment 17

[adamprocter]

I pasted a lot of info into Bug 623373 https://bugzilla.mozilla.org/show_bug.cgi?id=623373 - but yes both are the same bug

Comment 18

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Attachment: fix

This should fix it, although I don't have my Mac at home so I can't test it right now.

I wonder why this doesn't cause reftests to fail with acceleration off.

Comment 19

[Robert O'Callahan (:roc) (email my personal email if necessary)]

tryserver build will appear at http://ftp.mozilla.org/pub/mozilla.org/firefox/tryserver-builds/[e-mail]-cabf467b8766

Comment 20

[nikarlo]

(In reply to comment #19)
> tryserver build will appear at
> http://ftp.mozilla.org/pub/mozilla.org/firefox/tryserver-builds/[e-mail]-cabf467b8766

I can't find your directory

Comment 21

[Markus Stange [:mstange]]

(In reply to comment #18)
> This should fix it, although I don't have my Mac at home so I can't test it
> right now.

Yes, this fixes it.

+        if (s->GetAllowUseAsSource() &&
+            (s->GetContentType() == gfxASurface::CONTENT_COLOR ||
+             s->GetOpaqueRect().Contains(GetRoundOutDeviceClipExtents(this)))) {

I think instead of s->GetAllowUseAsSource() you need (s->GetAllowUseAsSource() || s->GetType() == gfxASurface::SurfaceTypeTee), otherwise the tee surface case further down won't ever be reached.

> I wonder why this doesn't cause reftests to fail with acceleration off.

Probably because reftests draw into a canvas, not into the window.

I don't think the gfxQuartzSurface.cpp hunk was supposed to be part of the patch.

Comment 22

[Markus Stange [:mstange]]

Comment on attachment 502455 [details] [diff] [review]
fix

r=me with the comments addressed

Comment 25

[Alfonso Martinez]

I think that this bug might be also what I'm seeing as the regression range is also from 2011-01-02 to 2011-01-03:

Loading a dialog in CKEditor for the second time doesn't show up until the page is clicked. Load http://ckeditor.com/demo and click for example in the smiley dialog, the first time behaves more or less correctly, but the second one will fail.

This fails in my macbook and works correctly in XP

Comment 27

[Robert O'Callahan (:roc) (email my personal email if necessary)]

http://hg.mozilla.org/mozilla-central/rev/4c3416fdc7b0

Comment 29

[nikarlo]

I still have the same problem with the latest release of Minefield 4.0 b10 pre

Comment 30

[Robert O'Callahan (:roc) (email my personal email if necessary)]

What changeset ID in about:config?

Comment 31

[nikarlo]

(In reply to comment #30)
> What changeset ID in about:config?

3d4620449437

Comment 32

[Robert O'Callahan (:roc) (email my personal email if necessary)]

That changeset doesn't have this patch. Try the next nightly build.

Comment 34

[James Rome]

On beta 9, I also see tabs disappear when I delete one.

Comment 35

[Graham Anderson]

I think Bug 626211 may be a dupe of this. This is how Twitter.com appears in FF4 B9: http://screencast.com/t/D88TCC1hTJme

Fixed in subsequent Minefield builds for me.

Comment 37

[nikarlo]

(In reply to comment #32)
> That changeset doesn't have this patch. Try the next nightly build.

is the same also on build 20110117033243

Comment 38

[Robert O'Callahan (:roc) (email my personal email if necessary)]

What's the build ID in about:buildconfig?

Comment 39

[nikarlo]

Now with 66addc5c30ca works very good!
Thanks to all!