Make the numeric keypad type digits on the Omarchy lock screen and polkit prompt
Issue #8552 was read in full. Its body attributes the bug to QTBUG-32687 and says stopping omarchy-fcitx5.service changed nothing for that reporter. A later comment traces the event order (modifiers then keymap, reported as Hyprland discussion 16371, which the old issue URL now redirects to), and another comment traces the second keymap to fcitx5's virtual keyboard and the XKB-800 round-trip, fixed by xkbcommon/libxkbcommon#1054 (merged 2026-07-16) and A/B tested with kb_options = "compose:caps". That commenter also notes a setup without fcitx5 that still fails. So the fcitx5 keymap flip is one confirmed trigger, not the only one, and the cause overstated it. Confirmed here: libxkbcommon is 1.13.2-1 installed and in Arch extra, upstream releases list 1.14.0-beta1 and beta2 only. The libxkbcommon 1.14 claim was softened because the thread only shows the keymaps should then match, not a test. The per-surface fixes, PR #10530 (polkit) and #12667 (lock screen), are open, and grep finds no KeypadModifier handling in /usr/share/omarchy on 4.0.4-1. The danger was wrong: /etc/pam.d/system-auth and /etc/pam.d/omarchy-lock-password both run pam_faillock deny=10 unlock_time=120, the tally is per user, so failed pkexec attempts count toward the same lockout and testing there is not safe. Not exercised: keypad behaviour, pkexec, locking.
Every password attempt counts toward one per-user faillock tally: /etc/pam.d/system-auth (used by polkit and sudo) and /etc/pam.d/omarchy-lock-password both run pam_faillock deny=10 unlock_time=120. Ten failures across the lock screen, pkexec and sudo combined lock the account for two minutes everywhere, including a TTY login. Test with pkexec true rather than at the lock screen, but stop after two or three failures, and check the count with faillock --user $USER.
Symptom
The numeric keypad works in the terminal but types nothing (or moves the cursor like Home, Up, PageUp) in the lock screen password field, the polkit authentication dialog, the launcher and other Qt/Quickshell fields. NumLock is on and its LED is lit. A password with digits typed on the keypad is rejected. Refocusing the window or toggling NumLock sometimes fixes it for a while.
Cause
Qt surfaces (the Quickshell lock screen, polkit dialog and launcher) end up with NumLock off in their own XKB state although the seat has it on, so keypad keys arrive as Home, Up, PageUp and so on with no text. One confirmed trigger: Hyprland sends a wl_keyboard.keymap after the modifiers event that carries NumLock, and Qt builds a fresh XKB state from the new keymap with NumLock off. The second keymap exists because fcitx5's virtual keyboard (hl-virtual-keyboard-fcitx5, started by omarchy-fcitx5.service) re-parses the keymap, and libxkbcommon 1.13.2 cannot round-trip a key that belongs to two modifiers: Omarchy's default shift:both_capslock_cancel maps <LFSH> to both Shift and Lock ([XKB-800] Key "<LFSH>" added to modifier map for multiple modifiers). The two keymaps differ, so Hyprland resends a keymap every time the active device flips between the real keyboard and fcitx5. The libxkbcommon side is fixed by xkbcommon/libxkbcommon#1054, in the 1.14.0 betas and not in 1.13.2. The Hyprland side is reported as hyprwm/Hyprland discussion 16371. This is not the only trigger: the original reporter of omacom/omarchy#8552 saw the same failure with fcitx5 stopped, and Qt's long-standing failure to read NumLock at startup (QTBUG-32687) is the other explanation in that thread. Omarchy's per-surface fixes (PR #10530 for polkit, #12667 for the lock screen) are open and not in 4.0.4.
Fix
**Immediate workaround at a stuck prompt:** press NumLock twice. That resyncs Qt for that surface.
**Omarchy 4, lasting fix when fcitx5 is the trigger:** make the physical and fcitx5 keymaps identical by dropping the double-Shift Caps Lock option. Compose on Caps Lock keeps working:
-- ~/.config/hypr/input.lua
hl.config({
input = {
kb_options = "compose:caps",
},
})
kb_options replaces Omarchy's value. If your layout is non-Latin (ru, ua, gr, ir and the others in Omarchy's list), Omarchy's default also adds grp:alts_toggle so you can reach your own layout, so keep it: kb_options = "compose:caps,grp:alts_toggle". Check what is in effect first with hyprctl getoption input:kb_options.
hyprctl reload
systemctl --user restart omarchy-fcitx5.service
Check the installed libxkbcommon. Upstream fixed the round-trip in 1.14.0 (beta only as of 2026-10-05, Arch ships 1.13.2), which should make the two keymaps match with the default option string. That is inferred from the fix, not tested, so keep the override until you have confirmed the keypad works without it:
pacman -Q libxkbcommon
**If the keypad still fails with fcitx5 stopped** (systemctl --user stop omarchy-fcitx5.service, start it again afterwards), fcitx5 is not your trigger. NumLock twice is then the only workaround until Omarchy ships the per-surface keypad mapping from PR #10530 and #12667. The same symptom on a laptop or Apple keyboard with letters turning into digits is a different problem (numlock keypad emulation).
Verify
Lock the session, press keypad digits in the password field without touching NumLock and confirm they appear as dots. Open a polkit prompt (for example pkexec true) and type keypad digits.