Bug 259858 – [GTK] Blank webview with 2.41.6 in NewsFlash

• Component: WebKitGTK
• Status: RESOLVED FIXED

[Jan]

Created attachment 467205 [details]
epiphany & newsflash webkit://gpu output (both are identical)

The flatpak builds of NewsFlash with org.gnome.Sdk//master show a blank webview on my system with AMD GPU. Local builds work as expected.
I'm on fedora 38 which ships webkitgtk6.0 2.40.5 while the gnome master sdk ships 2.41.6.

The symptoms are pretty similar to #228268 in that the content of the webview is completely white (even in dark mode) but text and images are visible during a drag action. The inspector UI is completely unaffected by the bug.

The devel flatpak of epiphany which makes use of the exact same gnome sdk with the exact same webkit binary works as expected.

The culprit for NewsFlash on webkitgtk6.0 2.41.6 seems to be the hardware acceleration policy. If I force it to software rendering with "Never" content appears as expected.
That makes me wonder why epiphany works because "/org/gnome/epiphany/web/hardware-acceleration-policy" is set to "always". Or is the flaptak making use of its own sandboxed version of the gsettings database?

A CI build of NewsFlash that hits the bug on my pc can be found here: https://gitlab.com/news_flash/news_flash_gtk/-/jobs/4810676070/artifacts/raw/io.gitlab.news_flash.NewsFlash.Devel-3.0.0-e3738cd5.flatpak

[Jan]

Created attachment 467206 [details]
blank webview with drag content visible

[Michael Catanzaro]

(In reply to Jan from comment #0)
> That makes me wonder why epiphany works because
> "/org/gnome/epiphany/web/hardware-acceleration-policy" is set to "always".
> Or is the flaptak making use of its own sandboxed version of the gsettings
> database?

It is; flatpak application generally cannot read host settings. But I doubt that's relevant to this bug. Unless you've disabled hardware acceleration, it will be enabled. A quick way to test is to use the environment variable WEBKIT_DISABLE_COMPOSITING_MODE=1 (e.g. flatpak run --env=WEBKIT_DISABLE_COMPOSITING_MODE=1 ...)

> A CI build of NewsFlash that hits the bug on my pc can be found here:
> https://gitlab.com/news_flash/news_flash_gtk/-/jobs/4810676070/artifacts/raw/
> io.gitlab.news_flash.NewsFlash.Devel-3.0.0-e3738cd5.flatpak

I will imminently update the master runtime to WebKitGTK 2.41.90. Please try that and report back whether it makes any difference. There have been so many graphics-related fixes recently that it probably doesn't make sense to investigate until we know whether it still occurs with all the latest fixes.

[Michael Catanzaro]

Please also test *without* WEBKIT_DISABLE_COMPOSITING_MODE and *with* WEBKIT_DISABLE_DMABUF_RENDERER=1

[Jan]

Had a chance to test with 2.41.90:

- view is still white with HardwareAccelerationPolicy::Always
- WEBKIT_DISABLE_COMPOSITING_MODE=1 makes things appear on the screen again
- and so does WEBKIT_DISABLE_DMABUF_RENDERER=1
- both variables together work as well, but I'd assume disabling compositing makes dmabuf obsolete?

[Carlos Garcia Campos]

WEBKIT_DISABLE_DMABUF_RENDERER only makes sense when accedlerated compositing is not disabled. What's the output of ls /dev/dri? I'm not sure I understand what change made it fail, was the upgrade from 2.40.5 to 2.40.6 or a flatpak sdk upgfrade? or both?

[Jan]

ls /dev/dri
by-path  card1	renderD128

I can't really use the gnome 44 platform with my master builds since I already rely on new features from gtk 4.12 and libadwaita 1.4.

But I tried running the stable version of NewsFlash with the master platform and could reproduce the issue there as well

flatpak run --runtime=org.gnome.Platform//master io.gitlab.news_flash.NewsFlash

will also render a completely white webview

The only change is going from org.gnome.Platform//44 -> org.gnome.Platform//master

The 'surf' autopkgtests are failing in Debian and Ubuntu, it seems it started after the 2.41.4 -> 2.41.5 update.

Testing with 2.41.91 on an ubuntu mantic GNOME or xfce env confirms the problem, WEBKIT_DISABLE_COMPOSITING_MODE=1 workaround the issue here but WEBKIT_DISABLE_DMABUF_RENDERER=1 doesn't

And when the display is empty those messages are printed

Cannot get default EGL display: EGL_BAD_PARAMETER
Cannot create EGL context: invalid display (last error: EGL_SUCCESS)

is there more information or debugging that I could provide to help there? the issue is preventing the new webkitgtk version to land in the current Ubuntu serie

[Michael Catanzaro]

Carlos thinks the problems with surf are unrelated. Please report a separate bug for surf.

[Michael Catanzaro]

So, back to the Newsflash problem. Presumably MiniBrowser is broken too, right? Can you check to be sure, please:

$ flatpak run --command=/usr/libexec/webkitgtk-6.0/MiniBrowser io.gitlab.news_flash.NewsFlash

At least we know the problem here is caused by dmabuf renderer, so at least we have a workaround that we can deploy if necessary.

[Michael Catanzaro]

(In reply to Jan from comment #6)
> But I tried running the stable version of NewsFlash with the master platform
> and could reproduce the issue there as well
> 
> flatpak run --runtime=org.gnome.Platform//master
> io.gitlab.news_flash.NewsFlash
> 
> will also render a completely white webview

I installed NewsFlash and used that command to run it using the master runtime. But unfortunately, it works perfectly fine for me. I also have an AMD GPU.

[Michael Catanzaro]

(In reply to seb128 from comment #7) 
> Testing with 2.41.91 on an ubuntu mantic GNOME or xfce env confirms the
> problem, WEBKIT_DISABLE_COMPOSITING_MODE=1 workaround the issue here but
> WEBKIT_DISABLE_DMABUF_RENDERER=1 doesn't

Oh, and this is another indication that you have a different issue. The problem Jan is hitting with NewsFlash is resolved by WEBKIT_DISABLE_DMABUF_RENDERER=1.

[Michael Catanzaro]

So since you can reproduce and I cannot, using the same flatpak and same runtime, my first thought was this would be a hardware-specific issue. But your GL renderer is:

AMD Radeon RX 580 Series (polaris10, LLVM 16.0.6, DRM 3.52, 6.4.6-200.fc38.x86_64)

while mine is:

AMD Radeon RX 570 Series (polaris10, LLVM 16.0.6, DRM 3.54, 6.5.0-0.rc7.20230821gitf7757129e3de.50.fc39.x86_64)

So yeah, we have basically the same polaris10 graphics hardware. Seems pretty unlikely that RX 580 would be broken while RX 570 is fine.

Next guess: maybe, if we're lucky, the problem was miraculously fixed in the runtime sometime between your previous comment on August 13 and today? I noticed there were some regressions in mesa recently. It's at least possible.

[Jan]

It is indeed a miracle. I've been  running with "WEBKIT_DISABLE_DMABUF_RENDERER=1" for quite some time now. So I tested the mini browser without and it worked. Tested NewsFlash without and it worked as well.

The only caveat is I experience some flickering on load and rare freezes where the old article doesn't disappear until I force a redraw by resizing. With "WEBKIT_DISABLE_DMABUF_RENDERER=1" the flickers are gone. But that deserves a new issue.

(In reply to Michael Catanzaro from comment #10)
> Carlos thinks the problems with surf are unrelated. Please report a separate
> bug for surf.

ack, the new bug is https://bugs.webkit.org/show_bug.cgi?id=261268