Wayland: crash when configuring new outputs with kanshi

The issue:

• Qtile version: current master (c50cc12e)
• Running on Wayland
• No relevant logs found in qtile.log

I’m experiencing pretty reliable crashes (but not always) when my outputs (2 monitors at home, 3 at work) are being reconfigured by kanshi. This happens only when new outputs are added (HDMI cable connected, or USB DisplayLink dock connected, or wlopm --on \* runs). Interestingly, the crashes don’t seem to happen when removing outputs, like when a cable is disconnected and only my laptop screen remains active.

My qtile.log doesn’t contain any information related to this crash, and because I’m running qtile from my .bashrc, I don’t have any stdout or stderr output either (but will add this to the ticket soon if it contains anything interesting). As discussed in https://github.com/qtile/qtile/issues/3298 there don’t seem to be many options to debug these kind of crashes, so I’m currently hoping for the stderr output to contain a traceback.

Required:

• I have searched past issues to see if this bug has already been reported.

[arjan-s]

Sure, no problem! I can pretty reliably reproduce the crash with the following steps:

1. disable kanshi
2. use default config
3. unplug HDMI display
4. reload config with Ctrl-mod-r
5. replug HDMI display
6. reload config with Ctrl-mod-r (reloading repeatedly works best for reproducing the crash)
7. goto 3 if not crashed yet

This yields different debug outputs (I did it twice):

[3695518.300]  -> [e-mail](175, 0, 0, 0, 0)
[3695518.461]  -> [e-mail](1274633)
[3695518.467]  -> [e-mail]_id(33)
[3696897.455]  -> [e-mail](34, "wl_output", 4)
[3696897.510]  -> [e-mail](34, "wl_output", 4)
[3696903.083] [e-mail](wl_buffer@29, 0, 0)
[3696903.102] [e-mail](0, 0, 2147483647, 2147483647)
[3696903.105] [e-mail]()
[3696903.151]  -> [e-mail]()
Segmentation fault (core dumped)

[3858952.821]  -> [e-mail]_id(27)
[3858952.832] [e-mail](new id wl_callback@32)
[3858952.845] [e-mail]()
[3858955.369]  -> [e-mail]_remove(36)
[3858955.392]  -> [e-mail]_remove(36)
[3860253.358]  -> [e-mail](1439368)
[3860253.381]  -> [e-mail]_id(32)
[3861634.186]  -> [e-mail](37, "wl_output", 4)
[3861634.273]  -> [e-mail](37, "wl_output", 4)
Segmentation fault (core dumped)

If you want, I can upload the complete debug logs somewhere.

Additionally, here is another crash log with my regular config and without debugging:

00:00:12.114 [ERROR] [backend/drm/atomic.c:34] connector eDP-1: Atomic commit failed (pageflip): Device or resource busy

and another:

Errors from xkbcomp are not fatal to the X server
00:04:01.762 [ERROR] [xwayland/xwm.c:1556] xcb error: op ChangeProperty (no minor), code Window (no extension), sequence 289, value 4194305
00:04:01.762 [ERROR] [xwayland/xwm.c:1556] xcb error: op ChangeProperty (no minor), code Window (no extension), sequence 290, value 4194305
00:04:35.881 [ERROR] [xwayland/xwm.c:1556] xcb error: op ChangeProperty (no minor), code Window (no extension), sequence 363, value 4194308
00:04:35.881 [ERROR] [xwayland/xwm.c:1556] xcb error: op ChangeProperty (no minor), code Window (no extension), sequence 365, value 4194308
(EE) failed to read Wayland events: Broken pipe

With my regular config (and kanshi running) the crash is almost 100% reproducible, probably because the screens_reconfigured hook gets called several times after (un)plugging an external display. Why it isn’t just called once is a mystery to me. Probably has to do with the config getting loaded twice too, which is another mystery.

[m-col]

I think this will be fixed by https://github.com/qtile/qtile/pull/3812