1009722 - Skia content + OMTC Basic Layers + Linux = blank window

Status: RESOLVED WORKSFORME

(Core :: Graphics, defect)

Description

[Benoit Jacob [:bjacob] (mostly away)]

STR:
  1) Be on Linux
  2) Use Skia content
  3) Use OMTC basic layers
Result: entirely blank browser window. It's still clicky, though, and clicking at the right place brings up the tooltips (probably separate layers, as they have css transforms).

It works fine if you change any of these 3 factors:
 - If you try on Windows instead of Linux, it works (with Skia content and OMTC basic)
 - If you use Cairo content instead of Skia content, it works (With Linux and OMTC basic)
 - If you use old main-thread basic layers instead of OMTC basic layers, it works (on Linux with Skia content).

The xrender preference does not make a difference.

Comment 1

[George Wright (:gw280) (needinfo me!)]

For reference:

layers.acceleration.disabled = true
layers.offmainthreadcomposition.enabled = true
layers.offmainthreadcomposition.force-basic = true

Comment 2

[George Wright (:gw280) (needinfo me!)]

Matt theorised that it's because EndRemoteDrawing() in nsWindowGtk wasn't implemented. A quick hack to implement it with a system that calls Put() on mShmImage passing mGdkWindow as the target didn't show anything. (This is with xrender disabled)

Comment 3

[Frederic Plourde (:fred23)]

(In reply to Benoit Jacob [:bjacob] from comment #0)
> It works fine if you change any of these 3 factors:
>  - If you try on Windows instead of Linux, it works (with Skia content and
> OMTC basic)
>  - If you use Cairo content instead of Skia content, it works (With Linux
> and OMTC basic)
>  - If you use old main-thread basic layers instead of OMTC basic layers, it
> works (on Linux with Skia content).
> 
> The xrender preference does not make a difference.

Benoit, I guess this was using GTK3 builds, right ?

While working on Xlib surface removal for OMTC basic under GTK3, I got the exact same trouble and tracked this issue down to nsWindow::GetThebesSurface() not being implemented for GTK3 path (So BasicCompositor was not getting any surface from nsWindow::StartRemoteDrawing())

This was subsequently fixed by Bug 1013552

(I don't exactly know why this worked with 'cairo' backend and not 'skia' tho).
So it would be worth it to try to reproduce this bug again.

Comment 4

[Benoit Jacob [:bjacob] (mostly away)]

(In reply to Frederic Plourde (:fred23) from comment #3)
> (In reply to Benoit Jacob [:bjacob] from comment #0)
> > It works fine if you change any of these 3 factors:
> >  - If you try on Windows instead of Linux, it works (with Skia content and
> > OMTC basic)
> >  - If you use Cairo content instead of Skia content, it works (With Linux
> > and OMTC basic)
> >  - If you use old main-thread basic layers instead of OMTC basic layers, it
> > works (on Linux with Skia content).
> > 
> > The xrender preference does not make a difference.
> 
> Benoit, I guess this was using GTK3 builds, right ?

This was using whatever was the default when I wrote this comment. Was GTK3 already the default? I certainly didn't do anything specifically to opt in to GTK3.

Comment 5

[Frederic Plourde (:fred23)]

Ah, I don't think GTK3 was the default at this time either.
So there's something weird here...

Ok, let's keep that open, as I'll be having the required setup/build to test that very shortly.
So stay tuned for more news on that front.

thanks.

Comment 6

[George Wright (:gw280) (needinfo me!)]

No, this is on GTK2 builds. You can reproduce it by setting the prefs in comment 1 in nightly

Comment 7

[Séamus O'Connor]

When I upgraded to Developer Preview 40.0a2 on Linux w/ Fedora 21 it started with blank windows like this bug suggests. Both Beta and nightly worked fine. I tried changing various combinations of: 

layers.acceleration.disabled = true/false
layers.offmainthreadcomposition.enabled = true/false
layers.offmainthreadcomposition.force-basic = true/false

Every time FF DP restarted it showed blank windows. Going to about:support showed that I had a non-default setting for: gfx.xrender.enabled set to false. Changing to true and restarting fixed it. Was my issue related or unrelated to this bug?

Comment 8

[Pawel Widera]

Attachment: screenshot of the blank window

Firefox 40 run on GNU/Linux renders a blank window with skia content backend.

Comment 9

[Pawel Widera]

I'm confirming this bug. After upgrade from Firefox 39 to 40 on Linux Mint, it renders a blank window (see attached screenshot).

I had default settings ("true") for "gfx.renderer.enabled" and "layers.offmainthreadcomposition.enabled", and modified "gfx.content.azure.backends" set to "skia,cairo" (see attached about:support info).

After changing the content backend to cairo, the Firefox window is render correctly.

Comment 10

[Pawel Widera]

Attachment: about:support info

Comment 11

[Shmerl]

With GTK3 enabled in Firefox 46, Skia, OMTC and full acceleration work fine for me using these settings in about:config

layers.acceleration.force-enabled = true
gfx.canvas.azure.backends = skia
gfx.content.azure.backends = skia
gfx.canvas.azure.accelerated = true

about:support shows:

GPU Accelerated Windows: 2/2 OpenGL (OMTC)
AzureCanvasBackend: skia
AzureContentBackend: skia
AzureSkiaAccelerated: 1

OS: current Debian testing x86_64 / KDE Plasma 5. GPU: Nvidia GTX 680, closed driver 361.42.

Comment 12

[George Wright (:gw280) (needinfo me!)]

(In reply to Pawel Widera from comment #9)
> I'm confirming this bug. After upgrade from Firefox 39 to 40 on Linux Mint,
> it renders a blank window (see attached screenshot).
> 
> I had default settings ("true") for "gfx.renderer.enabled" and
> "layers.offmainthreadcomposition.enabled", and modified
> "gfx.content.azure.backends" set to "skia,cairo" (see attached about:support
> info).
> 
> After changing the content backend to cairo, the Firefox window is render
> correctly.

Do you also see it on nightly?

Comment 13

[Mason Chang [Inactive] [:mchang]]

Is this still a problem?

Comment 14

[Pawel Widera]

I'm not sure how to test it in 48+. The config options have changed and
"layers.offmainthreadcomposition.enabled" is no longer there. How can I
enable the OMTC and acceleration?

I've tried Shmerl's settings (in comment 11)  but these do not enable
canvas acceleration. I end up with the following in about:support:
- AzureCanvasAccelerated: 0
- AzureCanvasBackend: skia
- AzureContentBackend: skia
- AzureFallbackCanvasBackend: none
- CairoUseXRender: 1

Comment 15

[Mason Chang [Inactive] [:mchang]]

(In reply to Pawel Widera from comment #14)
> I'm not sure how to test it in 48+. The config options have changed and
> "layers.offmainthreadcomposition.enabled" is no longer there. How can I
> enable the OMTC and acceleration?
> 
> I've tried Shmerl's settings (in comment 11)  but these do not enable
> canvas acceleration. I end up with the following in about:support:
> - AzureCanvasAccelerated: 0
> - AzureCanvasBackend: skia
> - AzureContentBackend: skia
> - AzureFallbackCanvasBackend: none
> - CairoUseXRender: 1

I don't think we have hardware acceleration on linux, but I'm not quite sure.

To check if acceleration is enabled, check the preference "layers.acceleration.disabled" is not set to true. Can you please also attach your whole about:support here please?

Comment 16

[Mason Chang [Inactive] [:mchang]]

Resolving as WFM. Please re-open if this is still an issue. By default, we should have OMTC with basic layers on linux with 48.