Regression: monitor scheduled config reloads not applied, internal screen stays off after disconnecting external screen

[aruhier]

Hi,
I use kanshi on my laptop with an external monitor. Kanshi has 2 profiles: when it's docked to my monitor, the internal screen of the laptop is disabled, otherwise the internal screen is enabled.

Example of kanshi config to reproduce the problem:

profile {
    output eDP-1 disable
    output DP-1 enable
}

profile {
    output eDP-1 enable
}

Since 8726a73, this doesn't work and after disconnecting the monitor, eDP-1 stays off. If I manually trigger kanshictl reload then it works, or if I plug back the screen it's in a weird state that turn on both screens and disconnecting it quickly after allows me to recover the screen. Kanshi detects the disconnection and tries to load the correct profile, but the screen stays off.

I bisected it to 8726a73. Going back to b940b0d makes it work again.

Edit: fix-monitor-config-event.patch fixes the problem, but is probably not a good solution.

Update: retried with Hyprland built on 5e441ca (2026-05-11), the problem is still present.

[Vladimir-csp]

I've seen this behavior, but intermittently, and only with specific laptop/dock, couldn't catch to diagnose.

[aruhier]

On my laptop (thinkpad x1 carbon gen 12), it happens with 2 different screens in USB-C (DP Alt).

[Vladimir-csp]

With 0.55.x it happens almost every time now, both with a Wavlink dock and a direct Type-c monitor. All hopes on #14547 it seems.

[aruhier]

@vaxerski, as you wrote the config refactor commit (8726a73), do you know where it could have added a race condition (I guess) in the monitor config/state?

[aruhier]

I used an LLM to come up with this patch: fix-monitor-config-event.patch

I tested it, it fixes the bug on main. The patch makes sense to me, after 8726a73 the reload is not applied if nothing is to be rendered on the screen, and this fixes it by queuing the reload in the main event loop.

However, I don't know if this is the best solution nor if it should be done outside of the scheduleReload() method.

[aruhier]

I changed the category to bug/config.

[aruhier]

I bump this issue, as the same thing happened on my other laptop, Fedora 44 ARM, so it seems to be easily reproducible.

[aruhier]

On my laptop on fedora, I run Hyprland 0.55.0 stable, the problem is present (on hyprlang at least, I didn't test with lua yet on it).

Edit: I updated the laptop running on git to 5e441ca, I confirm that the problem is still there. Applying the patch I linked before mitigates it.

[SafwanElmadani]

I'm on 0.55.0 as well and I have the same issue with Kanshi. The provided patch fixed it for me.

[aruhier]

Thanks! I don't want to create a PR for this patch because unless @vaxerski is ok with it, I'm really not sure that using the global event queue for this is a clean solution. For example, I don't know if it's safe if the monitor gets destroyed before the event is processed by the event queue.

[vaxerski]

Make a MR if you have the code, we can discuss there

[aruhier]

Thanks, I opened #14529.