pp3345 / gnome-with-patches Goto Github PK

I'm getting a crash with Wayland session only with the latest updates. I'm running:

mutter-3.36.2-101.fc32.pp3345.x86_64
gnome-shell-3.36.2-200.fc32.pp3345.x86_64

This usually happens when I interact with Firefox. I'm not sure which of the newly introduced changes causes it, but my earlier backport of !1241 worked fine. So that MR doesn't appear to be the problem.

Here's the relevant stack trace snippet of the crash, might be of help:

Mai 21 21:43:22 stardust systemd-coredump[24007]: Process 22381 (gnome-shell) of user 1000 dumped core.
                                                  
                                                  Stack trace of thread 22381:
                                                  #0  0x00007f764946653c wl_resource_post_event (libwayland-server.so.0 + 0x853c)
                                                  #1  0x00007f764a7a888a g_closure_invoke (libgobject-2.0.so.0 + 0x1388a)
                                                  #2  0x00007f764a7bb463 signal_emit_unlocked_R.isra.0 (libgobject-2.0.so.0 + 0x26463)
                                                  #3  0x00007f764a7c1ad9 g_signal_emit_valist (libgobject-2.0.so.0 + 0x2cad9)
                                                  #4  0x00007f764a7c1c43 g_signal_emit (libgobject-2.0.so.0 + 0x2cc43)
                                                  #5  0x00007f7648cc2af0 ffi_call_unix64 (libffi.so.6 + 0x6af0)
                                                  #6  0x00007f7648cc22ab ffi_call (libffi.so.6 + 0x62ab)
                                                  #7  0x00007f764946bcd2 wl_closure_invoke (libwayland-server.so.0 + 0xdcd2)
                                                  #8  0x00007f7649467132 wl_client_connection_data (libwayland-server.so.0 + 0x9132)
                                                  #9  0x00007f7649469bea wl_event_loop_dispatch (libwayland-server.so.0 + 0xbbea)
                                                  #10 0x00007f7649ae901b wayland_event_source_dispatch (libmutter-6.so.0 + 0x11b01b)
                                                  #11 0x00007f764a6bc7cf g_main_context_dispatch (libglib-2.0.so.0 + 0x527cf)
                                                  #12 0x00007f764a6bcb58 g_main_context_iterate.constprop.0 (libglib-2.0.so.0 + 0x52b58)
                                                  #13 0x00007f764a6bce73 g_main_loop_run (libglib-2.0.so.0 + 0x52e73)
                                                  #14 0x00007f7649a9b3e0 meta_run (libmutter-6.so.0 + 0xcd3e0)
                                                  #15 0x0000558795baac9c main (gnome-shell + 0x2c9c)
                                                  #16 0x00007f7649829042 __libc_start_main (libc.so.6 + 0x27042)
                                                  #17 0x0000558795baae7e _start (gnome-shell + 0x2e7e)

The crash occurs randomly, but the stack trace always is similar.

Nonetheless, thanks for keeping the COPR updated! We should be able to figure this one out. :)

I'm curious if there are any plans to release a patched Gnome 40 -- I've loved using it on Gnome 3.38 and earlier, and would love to add it to my F34 install 😄

The "edge resistance" feature makes window movements over screen boundaries feel sluggish and delayed. GNOME upstream disabled it recently (unless you press CTRL) due to that. Not sure if it makes sense to include it, but I like the change.

https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1268

https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/96

https://gitlab.gnome.org/GNOME/mutter/merge_requests/526

Have you tested it ?

Thanks

!867 "Use regions for Cogl clipping and culling" has some serious performance drawbacks for the icon spring animation as described in https://gitlab.gnome.org/GNOME/mutter/issues/971.
So please do not apply it on the mutter build.

The gjs package was updated in Fedora 31 and an update in the PPA is needed as well. Otherwise, Fedora's main version will override the PPA.

After the recent updates last week I do experience some crashes to login screen/gdm after locking the screen. (F30; intel; wayland; 3840x2160 hi-dpi 200% scaling; one external monitor, laptop-lid closed in docking-station)
Sometimes the crash to gdm happens immedeately after locking the screen and gdm login screen is displayed or this happens sometimes after some minutes of inactivity when locked (screen powered off due power saving). Can reproduce on two more or less identical systems.

Switching back to default Fedora packages fixes the issue.

It seems the related issue can be found upstream: https://gitlab.gnome.org/GNOME/mutter/issues/735

Would be cool if any patches fixing this regression could be incorporated in here as it's a pain working with unpatched g-s/mutter in my setup.

Thanks for your hard work on making some g-s 3.32 users happy!

Hi,
I installed Fedora 33 beta this morning and first thing what I did is added your repo of-cause :) And just want to report that performance is noticeably worth than on Fedora 32. Stutter on overview and selecting windows in there. Super fresh system and nvidia proprietary drivers installation so not much could interfere. Just letting you know and again huge thanks for maintaining such valuable repo 👍

Any chance of getting these added?

https://gitlab.gnome.org/GNOME/mutter/merge_requests/1003
https://gitlab.gnome.org/GNOME/mutter/merge_requests/1004
https://gitlab.gnome.org/GNOME/mutter/merge_requests/1017
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/936
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/948
https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/949

@pp3345 Thanks a lot for this great COPR repo and the time you take to keep it up to data, well done!

I have seen that you recently updated Mutter. Is there any reason why you did not update Gnome-shell with it and especially cherrypicked https://gitlab.gnome.org/GNOME/gnome-shell/merge_requests/763?
This will make life a bit better again :)

Thank you very much for providing this COPR! It made gnome-shell much more responsive so I can keep it a bit longer before generating new waste.

Since a while ago, I'm seeing a massive mouse cursor lag during Alt+Tab. This applies for different (if not all) package versions of your copr on Fedora 30.

Steps to reproduce:

1. Press Alt key down and keep it pressed.
2. Press and release Tab key
3. Move your mouse cursor over the application switcher overlay¹
4. Watch the cursor while you move it

What happens:
As long as the mouse cursor is over the application switcher overlay¹, the mouse cursor moves very slowly. As soon as it leaves the overlay, the mouse cursor moves at normal speed.

What should happen:
The mouse cursor should move the same speed no matter which gnome-shell part it is hovering.

¹: I don't know what this is called exactly, but I mean this list of icons that appear in a gnome-shell-style overlay before my applications when pressing Alt+Tab. It contains all the running applications.

Additional info:

• different versions of mutter, gnome-shell from this repo for Fedora 30
• the issue exists for quite some time
• I have not seen a build from this repository which has not been affected by this issue

MR is here: https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1441
That patch should improve the performance drastically with HiDPI screens with lower-powered Intel GPU's. Right now, the session is pretty laggy (20-30 fps if 3-4 window is opened maybe?).

With the last packaging (and therfore in the copr), mutter was updatet but not gnome-shell. The new mutter seems to be incompatible with the old gnome-shell resulting in an instantly crashing desktop!

I'll try to provide as much information as I can, but as this is the first issue I've ever submitted, try to bear with me.

I ran sudo dnf --refresh upgrade today and it installed the newest mutter patch from @updates (I have had the gnome-with-patches enabled for a long time). After restarting the machine, only the mouse cursor shows up, the rest of the screen stays black. The issue goes away when I disable the gnome-with-patches repo and downgrade mutter and gnome-shell via cli.

Steps to reproduce:

$ sudo dnf copr enable pp3345/gnome-with-patches
$ sudo dnf upgrade
$ reboot

Steps to remedy the issue:

$ sudo dnf copr disable pp3345/gnome-with-patches
$ sudo dnf downgrade gnome-shell mutter
$ reboot

Info about the upgraded packages:

    Upgrade  mutter-3.32.2-4.fc30.x86_64              @updates
    Upgraded mutter-3.32.2-3.10.fc30.pp3345.x86_64    @@System

Essential system info:

OS: Fedora release 30 (Thirty) x86_64 
Host: 20BTS33S01 ThinkPad X1 Carbon 3rd 
Kernel: 5.2.17-200.fc30.x86_64 
Resolution: 2560x1440 
GNOME 3.32.2 
CPU: Intel i5-5300U (4) @ 2.900GHz 
GPU: Intel HD Graphics 5500 

https://gitlab.gnome.org/GNOME/gjs/merge_requests/390 was closed, as it was not the completely correct solution. Instead, https://gitlab.gnome.org/GNOME/gjs/merge_requests/394 was implemented (and is also already merged into master).

Hi,
I am still in the process of understanding of what happened, and I am not even sure gnome-with-patches can do anything about it.

I updated my system yesterday, and after that Gnome Shell wouldn't start anymore. It would stall with a mouse cursor on the screen and a cursor blinking. The system was up, so I could switch to tty and login.

I think the problem was that mutter got an update from the official repository.

  Upgrade  glibmm24-2.60.1-1.fc30.x86_64                         @updates
    Upgraded glibmm24-2.60.0-1.fc30.x86_64                         @@System
    Upgrade  gnome-software-3.32.4-3.fc30.x86_64                   @updates
    Upgraded gnome-software-3.32.4-2.fc30.x86_64                   @@System
    Upgrade  mutter-3.32.2-4.fc30.x86_64                           @updates
    Upgraded mutter-3.32.2-3.10.fc30.pp3345.x86_64                 @@System

What fixed the issue was disabling the gnome-with-patches COPR and then run dnf distro-sync. (it pulled gnome-shell from the fedora repo).

What could be done to prevent this from happening in the future?

FYI, here are some related logs, and here are some more

To keep the amount of breakages to the absolute minimum, and streamline the maintainability of my Silverblue installation, I'd prefer to only install necessary RPMs since rpm-ostree override replace expects a path to an RPM, which might make testing the GNOME with patches difficult when there's a multitude of RPMs involved.

I'm a Wayland user, but I also occasionally use X11 to play games that require low input latency, and I'm mainly interested in testing https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/798. So, as the title says, what's the minimal set of packages required to be replaced in my case? For example, what xorg-x11-* packages need to be overlayed/replaced if any at all?

I have the following packages replaced in the base image of Silverblue 32 for now:

xorg-x11-server-Xorg 
xorg-x11-server-Xwayland
xorg-x11-server-common
gnome-shell
mutter

Tried mutter without !1309, with !1441 and also !168. With !168, it is felt more responsive experience, little but feelable.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/168

Built RPM:
mutter-3.38.1-101.fc33.pp3345.x86_64.zip

Hello,
I opened issue on gnome github and Daniel van Vugt answered there and recommend to apply two patches.
https://gitlab.gnome.org/GNOME/mutter/-/issues/1291
this two:
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1284.patch
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1269.patch
Honestly I don't know how to use it. Could you please implement this two patches into next build?
I will greatly appreciate it.
Thank you.

I just updated my system to Fedora 33 today (running Wayland), and I've noticed a rather annoying issue around browsers.

Specifically: whenever a browser (tested both Firefox 82 and Chrome 86) is maximized, I used to be able to move my mouse to the top border of the screen and select tabs without having to precisely point at the tab itself -- rather, just the top edge of the screen. With Fedora 33 and patches, an imperceptible "gap" is introduced that means the tabs are unclickable when my mouse is pushed all the way to the window border.

Perhaps an easier example on vanilla GNOME is the Activities bar. I can no longer click on any of the targets (notifications, calendar, etc) without clicking directly on the targets. If the mouse is right on the top edge, it fails to click them.

Here's a screenshot of things working as intended:

With my mouse pinned to the top edge of the screen, I can still click and switch to the Ars Technica tab. On the new patch set, putting the mouse in the same place would not select the Ars Technica tab, and clicking does nothing. I have to actually click in the middle of the tab target in order to successfully switch.

Disabling gnome-with-patches with no other changes does fix things, so I'm fairly confident it is one of the new patches. (Things were fine on F32 with the patchset.) Additionally, as mentioned before, I noticed the issue on both Chrome (XWayland) and Firefox (Wayland).

Hi Yussuf!
Was wondering if you could implement patches from this repository as well: https://copr.fedorainfracloud.org/coprs/whot/high-resolution-wheel-scrolling/ . Here is developer's blog: https://who-t.blogspot.com/ Problem is that the packages of his is conflicting with yours obviously. Ofcourse if it is possible and you have a time and desire.
Just trying to get gnome experience smoother :)
Thank you for your help again. With today's update gnome with nvidia is practically flawless. Just very very occasional lagging occur.
Regards.

I've got a strange side effect, on a Dell 7400 (full intel i7/8), with patches installed from the copr.

Zoom, (which uses its custom Qt 5.9 libs) runs on xwayland by default [QT_QPA_PLATFORM=xcb]
It causes random complete crashes of the shell: you're kicked to VT, and gdm has to be restarted manually.

One can make it crash consistently by going to the list of scheduled meeting and clicking on 'copy invitation link'
I've spent a bit of time on looking for a cause, and only reverting these patches avoids these crashes...

This journald entries seems to show the crash messages de journal.txt

BR

The results are simply insane and there are no regressions reported yet.

Hello!
I understand that F30 is near EOL, but i still using it and it will be great if you update rebuilded gnome-shell package with patches for the latest gnome-shell package version available in repo - 3.32.2-3.
Currently your CORP have 3.32.2-2.3.fc30.pp3345.x86_64 version and dnf wants to update to "updates" repo version( 3.32.2-3) from fedora repo.

Thanks in advance!

This should build agaisnt the latest Master now.

https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1003#note_754072

https://gitlab.gnome.org/GNOME/mutter/merge_requests/493

It looks like https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/168 has been revived and updated with master. I did a quick check to see how easy it would be to merge with the 3.36 branch, and it was harder than I had hoped. But I remember it bringing significant latency improvements in a previous Fedora release with gnome-with-patches, so I'm submitting it here for consideration.

The packages dated the 1st of May 2020 have an issue with Vulkan starting up, both with vkcube showing a black screen and wine games utilising DXVK on default gnome settings (Wayland) with these patches.

No issues with Xorg backend

System is an AMD Ryzen 7 1700 with a PowerColor Red Devil RX 5600 XT

Please add culling fixes, i.e. these patches:

https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1324
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1323

I'm on a system with slow Intel iGPU with a dual screen setup, I'm pretty sure it should make a big difference. :)

Hello, Thanks so much for your great repo. Just want to report that latest update broke dash to dock extension. Please take a look in it.
Thank you.

Does https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285 apply cleanly (or with a small diff) on 3.36? If yes, it might be worth considering to add it as the performance improvements for people with screens having different refresh rates is immense.

And, more interestingly, cannot switch to tty with any of the combination of the key.