Output mapping breaks (partial horizontal coverage) after tablet power-cycle on KDE Wayland with fractional scaling; daemon restart always cures it

Steps to Reproduce

On KDE Plasma 6 (Wayland), with my display at 3840×2160 @ 125% fractional scaling (logical resolution 3072×1728), the tablet maps correctly to the full screen only until the tablet is power-cycled. After turning the tablet off and on again (2.4 GHz wireless dongle — dongle itself stays plugged in), the pen's horizontal travel covers only part of the screen (roughly half), while vertical coverage remains full. Running systemctl --user restart opentabletdriver immediately and reliably restores full coverage — 100% reproducible in both directions.

The behavior matches the FAQ's note that OTD does not receive monitor-layout updates dynamically — it appears the daemon caches/stales the display geometry at (re)detection time and never refreshes it.

Steps to reproduce

1. Boot PC with OTD daemon enabled, tablet powered on. Mapping is correct (full screen coverage).
2. Power the tablet off and back on (physical power switch; wireless dongle remains connected).
3. Move the pen: horizontal travel now covers only ~half the screen width; vertical is unaffected.
4. Run systemctl --user restart opentabletdriver (no settings changed) → full coverage restored instantly.
5. Power-cycle again → broken again. Restart daemon → fixed again. Deterministic.

Environment

Item	Value
OS	Arch Linux (rolling, up to date as of 2026-08-10)
OTD version	0.6.7 (distro package 0.6.7-2)
Desktop	KDE Plasma 6.7.4, Wayland (KWin)
Display	3840×2160 @ 125% fractional scaling → logical 3072×1728 at position (0,0)
Outputs	One active monitor; a second virtual output exists but is disabled (cannot be removed; required by user's workflow)
Tablet	XP-Pen Deco LW, connected via 2.4 GHz wireless dongle
USB ID	28bd:0935 (lsusb: "XP-Pen Deco LW")
Kernel device names	Hanvon Ugee Deco LW (pen), Hanvon Ugee Deco LW Keyboard, Hanvon Ugee Deco LW Mouse — three input nodes, present regardless of tablet power state
OTD output mode	Artist Mode
OTD output mapping	Virtual Display: 3072×1728, X 1536, Y 864 (centered)
OTD tablet area	Full: 254×152.4 mm, X 127, Y 76.2, rotation 0

Suggested fix direction

On tablet (re)detection, re-query the current display layout and re-apply the saved output mapping (i.e., treat device re-arrival like a daemon start with respect to display geometry), and/or subscribe to compositor layout changes (e.g., xdg-output / wl_output updates) so mappings can't go stale.

Workaround in use

udev rule restarting the user daemon on dongle add events (boot/replug), plus a global keyboard shortcut running systemctl --user restart opentabletdriver after each tablet power-cycle. Full runbook available on request.

P.S. I've used AI to help me re-write this in an orderly fashion.

Expected Behavior

Expected behavior

OTD re-detects the tablet on re-pair and re-applies the configured output mapping against current display geometry, without requiring a daemon restart.

Observed Behavior

Additional observations

1. Fractional scaling interacts with the symptom: at 125% scaling, coverage is ~50% of screen width; temporarily setting scale to 100% improves it to ~75% (3840 native vs 3072 logical mismatch). So the stale geometry OTD caches appears to be in physical/native pixels rather than logical, or reflects the virtual-desktop span rather than the single output.
2. Dual tablet detection: OTD's window title shows "XP-Pen Deco L, XP-Pen Deco LW" — the dongle's interfaces can enumerate as either entry across power-cycles. Output settings were saved for both entries, so this is not the cause of the mapping loss (verified: the bug persists even with identical correct settings on both).
3. Why this can't be worked around in user space: in Artist Mode the daemon exclusively grabs (EVIOCGRAB) the physical tablet's input nodes, so no user-session process can observe the pen waking up (e.g., libinput debug-events sees nothing from the physical device). The input nodes also persist while the tablet is off (device count never changes), and the tablet exposes no /sys/class/power_supply/ entry. Therefore no udev/input/power signal exists for the power-cycle event; the only viable user-side workaround is a daemon restart triggered manually or on dongle replug/boot.
4. KWin-side causes were excluded: KDE's Drawing Tablet KCM correctly defers to the userspace driver; the old org.kde.KWin.InputDevice D-Bus mapping overrides no longer exist in Plasma 6; mapping the virtual device via KWin was tested both ways.
5. Possibly related to known issues around stale display geometry / absolute-mode mappings breaking when monitor or scaling settings change (e.g. #4105 and the Linux FAQ note about restarting OTD after monitor configuration changes).

Tablet Device

XP-Pen Deco LW

Diagnostic Information

Diagnostics

Happy to attach daemon logs captured across a power-cycle (journalctl --user -u opentabletdriver), settings.json, and any other output you need, just tell me what to collect.

Acknowledgements

• I have installed OpenTabletDriver by following its official installation instructions.
• I have checked the OpenTabletDriver Wiki and respective FAQ pages, and my issue was not covered or actually fixed.
• I have searched the existing issues and this new issue is not a duplicate of any.
• I have written a concise and meaningful title.
• I am on the latest version of OpenTabletDriver.

Start by reproducing the power-cycle sequence and comparing journalctl --user -u opentabletdriver logs before and after the failure. Trace tablet re-detection, saved output mapping, and display geometry handling; the fix is complete when the configured 3072×1728 mapping remains full-width after power-cycling without restarting the daemon.