923961 - B2G Desktop OOP on mac doesn't display anything from child processes

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Gregor Wagner [:gwagner]]

It would really nice to have b2g oop again.
It seems to work but we don't render content from child processes.
For the homescreen we only get the background image. I can still click on the homescreen and open apps.
The browser for example renders the UI but any content is only white.
The clock app for example just shows a white screen. If I go to the window manager I see the correct screenshot of the app.

Comment 1

[Gregor Wagner [:gwagner]]

Blake, you looked at this a little bit. Can you add any findings here?

Comment 2

[Blake Kaplan (:mrbkap) (inactive)]

I didn't really look into the painting stuff very much.

Comment 3

[Gregor Wagner [:gwagner]]

Milan, can we get an estimation here when someone can take a look?
oop b2g-desktop has been broken for months. Debugging our koi+ bugs takes much longer because of this.

Comment 4

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

I appreciate the difficulty without the desktop tools, but people have been trigger happy on making graphics bugs koi+, so this is going to wait until those are cleared.  I'd say December at the earliest.

Comment 5

[Preeti Raghunath(:Preeti)]

Based on comment 4 moving to 1.3

Comment 6

[Gregor Wagner [:gwagner]]

(In reply to Milan Sreckovic [:milan] from comment #4)
> I appreciate the difficulty without the desktop tools, but people have been
> trigger happy on making graphics bugs koi+, so this is going to wait until
> those are cleared.  I'd say December at the earliest.

Thanks for the reply. So this means we are not gonna have it for any 1.3 features.
It's important to know for our planning.

Comment 7

[Andreas Gal :gal]

Is this a regression?

Comment 8

[Gregor Wagner [:gwagner]]

(In reply to Andreas Gal :gal from comment #7)
> Is this a regression?

Yes, but unclear when it regressed.
B2g-desktop OOP was completely broken for months now. We had to fix many bugs to get it working again and this seems the last piece to make it work on mac again.
jgriffin is also ready to turn on tests for it so we don't end up regressing it again.

Comment 9

[Andreas Gal :gal]

Lets identify the bug. This can't be that hard. Can you describe the simplest possible test case to me?

Comment 10

[Blake Kaplan (:mrbkap) (inactive)]

mhenretty and I will look at this tomorrow.

Comment 11

[Gregor Wagner [:gwagner]]

(In reply to Andreas Gal :gal from comment #9)
> Lets identify the bug. This can't be that hard. Can you describe the
> simplest possible test case to me?

1) Enable oop by switching http://hg.mozilla.org/mozilla-central/file/19fd3388c372/b2g/app/b2g.js#l254
2) make a b2g-desktop build and start with a fresh profile.
3a) and we have a new regression: bug 930836
3b) otherwise you would just see a white screen whenever a content process is visible (like homescreen).
The homescreen still works because random clicking opens applications and if I hit the browser, I see the browser UI.

Trying to find the regression for bug 930836 now.

Comment 12

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

This regression is exactly why we need the gfx code fixed. It's blocking turning on a whole lot of other tests, and until we have those tests running other things are going to keep regressing.

This is actively preventing people from doing B2G development. It should definitely be a higher priority than any 1.3 work.

Comment 13

[Blake Kaplan (:mrbkap) (inactive)]

Attachment: Force b2g-desktop to composite on OSX.

bent, mhenretty and I debugged this over the past couple of days. With this patch as well as:

diff --git a/b2g/app/b2g.js b/b2g/app/b2g.js
index bc73180..11b4466 100644
--- a/b2g/app/b2g.js
+++ b/b2g/app/b2g.js
@@ -251,8 +251,8 @@ pref("ui.dragThresholdY", 25);
 // Layers Acceleration.  We can only have nice things on gonk, because
 // they're not maintained anywhere else.
 #ifndef MOZ_WIDGET_GONK
-pref("dom.ipc.tabs.disabled", true);
-pref("layers.offmainthreadcomposition.enabled", false);
+pref("dom.ipc.tabs.disabled", false);
+pref("layers.offmainthreadcomposition.enabled", true);
 pref("layers.offmainthreadcomposition.async-animations", false);
 pref("layers.async-video.enabled", false);
 #else

we paint child processes and things seem to work. Looking at the bug that added this call, it seems like it should be safe to disable it.

The basic problem is that we refuse to do OMTC on transparent windows on Mac.

Comment 14

[Markus Stange [:mstange]]

But the window should not be transparent, should it? nsContainerFrame::SyncWindowProperties should call nsChildView::SetTransparencyMode(eTransparencyOpaque), which should forward it to nsCocoaWindow::SetTransparencyMode(eTransparencyOpaque), which should call [mWindow setOpaque:NO]. Is that call not happening? Or does it happen after we've already made the decision not to do OMTC?

Comment 15

[Markus Stange [:mstange]]

Comment on attachment 823733 [details] [diff] [review]
Force b2g-desktop to composite on OSX.

I guess for a quick fix to unblock things this is ok, but in that case please file a bug for finding the answer to comment 14 and reference the bug number in the code.

Comment 16

[Andreas Gal :gal]

In related news, why does it bother the OMTC code that the window is transparent?

Comment 17

[Markus Stange [:mstange]]

It bothers it if the window has a shadow (or at least it only has a reason to be bothered in the shadow case, not sure if the code reflects that) because OS X only draws a correct shadow for transparent windows when we're drawing without HWA into the normal window pixel buffer and not into an OpenGL context, and we only support OMTC with GL.

Comment 18

[Gregor Wagner [:gwagner]]

\o/ Thanks everyone!
https://hg.mozilla.org/integration/b2g-inbound/rev/cf582248b52e

Comment 19

[Blake Kaplan (:mrbkap) (inactive)]

I have the followup bug asked for in comment 15 sitting in a tab at home just waiting for me to click on "submit." I'll do that tonight.

Comment 20

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/cf582248b52e

Comment 21

[Blake Kaplan (:mrbkap) (inactive)]

(In reply to Blake Kaplan (:mrbkap) from comment #19)
> I have the followup bug asked for in comment 15 sitting in a tab at home
> just waiting for me to click on "submit." I'll do that tonight.

This is now bug 933106.

Comment 22

[Jason Smith [:jsmith]]

Already in 1.3, clearing nom.