Hot-plugging a disabled head sends an additional "phantom" mode through the zwlr_output_head_v1 protocol

[andriyDev]

Please fill out the following:

• Sway Version:

  • sway 1.10

• Debug Log:

  • https://gist.github.com/andriyDev/70cd4e4f2593e2ac3ac8b5234899c291

• Configuration File:

  • Used the default config (except switched to term=alacritty, menu=wofi)

• Description:

  • Connect to sway with any app that supports the zwlr_output_management_v1 protocol and just listen to the events (Here's my tiny Rust example, you need the wayland-client and wayland-protocols-wlr dependencies).
  • Connect an external monitor (I am using a monitor connected through HDMI)
  • Disable the monitor (I did this using wdisplays which itself uses the output management protocol).
  • Disconnect the monitor
  • Reconnect the monitor
  • See that the new head sends a mode event, but this mode is never populated (its size and refresh rate are never sent).
  • Bonus: Adjust any monitor (e.g., change the position of another monitor, re-enable the disabled monitor), and a finished event will be sent for the "phantom" mode.

• Other details:

  • I tried this on river and dwl which both use wlroots 0.18. They do not seem to report this phantom mode, but they also have different behavior: Sway will keep the monitor disabled after hotplugging. river and dwl on the other hand will re-enable the monitor on hotplug (I suspect they are just treating the monitor as a brand new monitor). This is also why I've filed this issue against Sway and not wlroots.

[andriyDev]

TL;DR even more reasons for why this is a Sway issue and not necessarily a wlroots issue.

Ok perhaps I'm misinterpreting the relationship with river and dwl. Something I didn't notice before (running two desktops at once and sharing a monitor): the "bonus" part seems to always re-enable the disabled monitor. That is, if you have two monitors eDP-1 and HDMI-A-1, apply the above steps to the HDMI-A-1 monitor, and then move eDP-1 (e.g., through wl-displays), HDMI-A-1 enables itself. So maybe river and dwl are doing something correctly that Sway isn't, resulting in this weird buggy behavior.

I assumed Sway was trying to be "smart" and keeping the monitor disabled since it was previously disabled, but maybe instead its a race condition of some kind. That could explain why after changing any other head, the "phantom" mode gets a finished event: the mode gets created accidentally which is sent before this phantom gets cleared server-side. Only later when another "done" batch is sent does the mode realize it doesn't exist and it "finishes" itself.

[emersion]

Do you have a WAYLAND_DEBUG=1 log of your client?

[andriyDev]

Here you go! https://gist.github.com/andriyDev/dfc662d8ea79df1f61b1efdfc6fa2bad