Nothing stood between "memory is tight" and "processes die at random":
the kernel OOM killer only fires after an allocation has already failed,
so a machine under pressure thrashes in reclaim until something breaks
somewhere fatal — seen in practice as Hyprland taking SIGBUS mid-memcpy
from a client shm pool that memory pressure was tearing down.
Enable systemd-oomd, which keys on PSI stall time rather than free
pages, and mark app.slice — and only app.slice — as its kill candidate.
Hyprland runs in session.slice, so the compositor is structurally
ineligible as a victim: oomd takes the app that caused the pressure and
the session survives. Thresholds (50% stall over 20s, Fedora's desktop
defaults) live in an oomd.conf.d drop-in; candidacy ships as a vendor
drop-in under /usr/lib/systemd/user so existing users get it on package
upgrade with no per-user seeding.
New installs enable the daemon from enable-services.sh; a migration
covers existing ones, restarting an already-running oomd so it doesn't
keep stale thresholds until reboot.
Deliberately not earlyoom: it triggers only when free RAM and free swap
are both under threshold, so Omarchy's large, mostly-idle swapfiles
keep it asleep through exactly this failure.
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Kvantum is a QStyle plugin, so it only ever painted the three QWidget
apps we ship, and it painted them with its own stock look: Omarchy has
never shipped a .kvconfig, so nothing about it tracked the active theme.
Without it Qt falls back to built-in Fusion while the palette still comes
from QT_QPA_PLATFORMTHEME=gtk3, so everything stays dark and now follows
the theme instead of ignoring it. Measured across a full 1920x1080 frame,
dropping it moves obs-studio by 132 pixels -- it overrides the platform
style with its own stylesheet anyway -- kdenlive by 0.8%, and the
Hyprland share picker by 4.7%. Flatter buttons, classic tabs, a
conventional checkbox tick.
Existing machines get a migration, which the Qt5 stack never did. The
difference is that this removal cannot reach the greeter. kvantum-qt5
goes in the same transaction, since it is the only thing that requires
kvantum, and it takes qt5-svg and qt5-x11extras along with it -- but
qt5-wayland is explicitly installed, so the cascade stops there and
leaves qt5-declarative and qt5-base standing for any sddm theme still
starting the Qt5 greeter.
kvantummanager stays in launcher.hides. The entry costs nothing once the
package is gone, and it keeps Kvantum Manager out of the launcher on
machines that have not run the migration yet.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
kvantum-qt5 only ever existed to dark-mode VLC, and VLC is retired.
qt5-wayland was the Qt5 Wayland platform plugin, but it also quietly
dragged in qt5-declarative, which is what kept sddm's Qt5 greeter alive.
No application we ship runs on Qt5 any more, so the whole stack goes
from fresh installs: about 110 MiB across kvantum-qt5, qt5-wayland,
qt5-declarative, qt5-base, qt5-svg, qt5-x11extras and qt5-translations.
fcitx5-qt keeps a few Qt5 objects it builds regardless, and they simply
go unused.
sddm picks its greeter from the theme's SddmGreeterTheme/QtVersion and
defaults that key to 5, so a theme without it execs the Qt5
/usr/bin/sddm-greeter. That tie is only an optdepend, so pacman would
happily take qt5-declarative and leave a login screen that cannot start.
The theme declares QtVersion=6 now, and the Qt6 greeter renders it
unchanged.
kvantum, the Qt6 style engine behind QT_STYLE_OVERRIDE, was only ever
present as a dependency of kvantum-qt5, so the package list swaps to it
rather than dropping the line. obs, kdenlive and the share picker keep
the styling they have today.
Machines already installed keep their Qt5 stack. Shedding it there means
deciding which greeter sddm will actually launch, and that means
reimplementing its config loader: sddm reads every file in sddm.conf.d
whatever its extension, QSettings trims keys and lets the last duplicate
win, and ThemeDir can move the themes entirely. Not worth risking a
login screen for 110 MiB that a reinstall reclaims anyway.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
fcitx5 is what turns the CapsLock compose sequences in ~/.XCompose into
text for Wayland clients -- CapsLock m s for an emoji, CapsLock space n
for your name. It was launched fire-and-forget from Hyprland's autostart
via uwsm-app, so nothing supervised it, nothing restarted it, and it
logged nowhere. When it went away, every compose sequence stopped
working for the rest of the session with no visible symptom beyond
"emoji input is busted", and no record of why.
That is not hypothetical: it was found dead on a running machine with no
coredump, no OOM kill, and nothing in the journal to explain it.
Move it to a systemd user service:
- Restart=always, not on-failure. fcitx5 exits 0 when it finds another
instance already owning org.fcitx.Fcitx5, and a clean exit still
leaves the user with no input method.
- After=/PartOf=graphical-session.target. It needs WAYLAND_DISPLAY and
DISPLAY, which uwsm imports into the user manager before reaching the
target, and its wayland connection dies with the compositor.
- ConditionEnvironment=WAYLAND_DISPLAY. After= is ordering only and does
not stop the unit from being started while the target is inactive. An
update over SSH has a live user manager (pam_systemd) and no graphical
session, and a fcitx5 started there comes up blind -- then stays
active, so the later target activation won't pull in a working one,
because Wants= does not restart what is already running. Skipping the
start leaves the unit enabled and healthy for the next graphical login.
The migration hands over inside a live session only: it enables without
--now, and only when graphical-session.target is active does it drop the
autostart-launched process and start the unit. Because that kills a
fcitx5 that was working a moment ago, a failed start is reported instead
of leaving the session mute with the migration marked complete.
omarchy-restart-xcompose now drives the unit. It still clears any fcitx5
running outside it first: that process owns the bus name, so the unit's
instance would exit on arrival and the stale one would keep serving the
old table -- a restart that reports success and changes nothing.
Side benefit: fcitx5 now logs to the journal under its own unit, so the
next disappearance leaves a record.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Chromium ships in the base packages, so it never goes through
omarchy-install-browser, and a first install stamps every migration as
already applied without running it. That left both bundled extensions
loaded from chromium-flags.conf with no host to talk to: Copy URL did
nothing at all, since the clipboard write and the toast both live in the
host.
The user install now registers both hosts itself, and a migration
repairs the installs that already missed out — Copy URL since 1784763917
landed, Download Video since 1780517689, unless installing a browser or
refreshing Chromium happened to fix them along the way.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VeAd2iERgi7bRSxq2gA1hp
updatedb refuses to run at all on a config that defines a variable twice
("variable `PRUNEPATHS' was already defined"), so any rewrite that misses
an existing line and appends a second one takes the locate index down
rather than fixing it. Two shapes updatedb accepts got missed: a trailing
comment after the value, and a setting indented by whitespace.
Read the existing paths out of the quoted value and write the whole
setting back canonically instead of splicing into a line of unknown
shape. Quotes are not optional to updatedb ("value in quotes expected
after `='"), so a bare value is already a broken config: rewriting it
quoted repairs the file as a side effect.
The tests now hand every rewritten file to the real parser through
`updatedb --config-file`, which is what caught this.
Read the Snapper config as root when the running user cannot read it.
snapper create-config leaves the config root-only, and a config the user
cannot read was passing for one that wants its timeline snapshots kept.
Report the snapshots the drain could not delete. omarchy-migrate writes
the completion marker whether or not the batches succeeded, so there is
no later run to pick up the remainder, whatever the comment claimed.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The machines this targets are the ones with an updatedb already grinding
through every snapshot, and `systemctl start` on an active unit is a
no-op. updatedb reads /etc/updatedb.conf once at startup, so a run that
began before the rewrite keeps burning CPU on the old config until it
finishes. Restart the service instead: it's Type=oneshot and plocate
builds into a temp db, so nothing is lost by replacing the run.
Quotes are optional in updatedb.conf, so parse the existing paths out of
whatever quoting the file uses and write the setting back in one
canonical form. `PRUNEPATHS=/tmp` previously fell through to the append
branch and got a second PRUNEPATHS line, which drops /tmp from the
pruned set. Comparing whole paths rather than substrings also keeps a
config that already prunes something like /var/lib/machines/.snapshots
from being mistaken for one that prunes /.snapshots.
Prefer $OMARCHY_PATH over the packaged copy when locating the config
script, per docs/migrations.md, so the migration test exercises the
checked-out script rather than whatever release is installed. Skip when
neither exists: omarchy-migrate runs under set -e, so a missing script
would take down every migration queued behind it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
omarchy-update-user-notify.path watched /usr/share/omarchy/migrations, but
pacman writes that directory during every update, including the blessed
omarchy update, which runs omarchy-migrate a step later. The watcher fired a
critical notification for the migrations the update was already applying in
the visible terminal. A watcher cannot tell that apart from a bypassed
pacman -Syu, so the only trigger that never collides with a running update is
a once-per-login check.
The service that already ran at graphical-session.target is now the whole
mechanism, renamed after the command it runs. That is also all the second-user
case needs: markers are per-user, so anyone who did not run the update finds
them missing at their next login.
Login timing means the toast can be sent before the shell has claimed
org.freedesktop.Notifications, so the notifier waits for a live server first.
The wait is omarchy-first-run's, lifted into omarchy-notification-wait rather
than duplicated.
The package keeps omarchy-update-user-notify.service as a symlink onto the new
unit. Existing users hold an absolute wants symlink to the old path, and the
migration that repoints it only runs for users who run an update, which is the
opposite of who the notifier is for.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A fresh install on a Dell XPS 14 spent 73s in "Configuring system", and two
silent gaps in the install log account for 66s of it. Both are the same
redundant work.
hardware/all.sh ran ipu7-camera.sh before ptl-kernel.sh. intel-ipu7-camera
pulls in ipu7-drivers, vision-drivers and v4l2loopback, so DKMS compiled all
three against the stock linux kernel and built a UKI for it. ptl-kernel.sh
then installed linux-ptl, which recompiled the same three modules against the
new kernel, and its pacman -Rdd of linux followed with three dkms removes and
yet another UKI rebuild. The log shows four UKI builds where one would do, and
everything the first script produced was deleted 42 seconds after it was made.
Running the kernel swap first means the DKMS modules are only ever built
against the kernel the machine actually keeps, which returns roughly 25s.
ptl-kernel.sh still installs linux-ptl before removing linux, so one redundant
UKI build for the stock kernel remains. Removing first would save it but would
leave the machine with no kernel at all if the linux-ptl install failed, so the
safe order stays.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017q97Rsk8KwgaPUVq5ArLgM
Clicking the Voxtype or fingerprint invitation did nothing. The launcher
execs setsid, which forks because the transient unit's main process is
already a process group leader, so the unit exits within milliseconds and
systemd's default control-group kill took the still-starting terminal down
with it. Run those units with KillMode=process instead.
The invitation tests had also drifted from the two-unit design and were
failing; teach the systemd-run mock to run the command it is given.
Both notifications are sent from background subshells, so whichever
notify-send registered first won the bottom slot. Give the update toast a
tick to register so Wi-Fi lands newest and stacks on top.
Laptop speakers ship voiced by the vendor's Windows DSP layer, which Linux does
not get. A tuning restores that as a PipeWire filter-chain in front of the
internal speaker sink, matched to the machine by DMI string and expected sink.
Adding a laptop is a directory under default/audio/tunings with two files and no
new code: matching is data. The XPS 14 DA14260 tuning included here was derived by
measuring the xps-audio-linux EasyEffects profile (MIT) and fitting a biquad chain
to it, so no impulse response or other upstream asset is redistributed. It measures
1.24 dB RMS against that reference, and matches its dynamic range within 0.1 LU --
the reference's multiband compressor turned out to contribute nothing, so a linear
chain replaces it. Bass Q is capped deliberately: a closer magnitude fit swung
group delay 31 ms across 63-80 Hz, which smears bass transients.
The graph runs as its own PipeWire client under its own config name rather than
loading into the audio daemon. The daemon only reads its config at startup, so a
daemon-loaded tuning could only be switched by restarting PipeWire -- which drops
every PulseAudio client's connection, and applications that do not reconnect
(Spotify) then have to be restarted by hand. Hosting it separately also contains
failure, since a malformed tuning breaks only that service.
Three things about the surrounding audio graph needed fixing for this to behave:
- Volume must live downstream of the tuning. omarchy-audio-output-sink is now the
single definition of which sink an output's volume really uses, shared by the
volume keys, the output switcher's OSD and the audio panel, so they cannot
disagree. It resolves the current default output, which keeps it correct when
headphones are selected while a tuning exists.
- The tuning's own output is a movable sink input, so rerouting "all streams" to a
newly selected output would drag the processing onto headphones, or into the
tuning's own sink, which is a cycle. It is pinned, and stream moves are limited
to streams carrying an application.name.
- The physical sink a tuning fronts is not independently selectable, since picking
it would only bypass the tuning, so it is kept out of the output list.
Applying happens at first-run, not finalize-user, because finalize-user also runs
in the ISO chroot where there is no audio server and nothing would retry.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Voxtype/fingerprint invitations run as a transient
omarchy-*-invitation.service that shows a critical toast and waits for a
click. On click the service execs the install terminal, so the service
stays active for the whole life of that terminal.
The installer ends with omarchy-restart-shell, which re-triggers any
still-running omarchy-*-invitation.service on the assumption that a
still-running unit is an unanswered toast wiped by the shell restart. But
an answered invitation is still running only because it is the parent of
the very install terminal that triggered the restart, so it gets falsely
re-fired and the invitation toast pops a second time right after the
install finishes.
Launch the installer in its own transient unit so the invitation service
exits immediately after the click. Genuinely unanswered invitations still
block in notify-send and are still recovered by restart-shell; answered
ones are gone and no longer match the re-show glob.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WHPDqx5hwLTwq27f7KpM9d
The setup installs packages (libfprint-git, fprintd, usbutils), which
isn't reliable on a fresh install before the databases have been
synced. Defer the invitation with a post-update hook, like Voxtype.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Add a first-run notification, alongside the keybindings/Wi-Fi toasts, that
invites anyone with a fingerprint sensor to enable it. Clicking launches
omarchy-setup-security-fingerprint in a floating terminal.
Detection lives in a new omarchy-hw-fingerprint helper that reads sysfs
(device product descriptor plus a fingerprint-vendor allowlist), so it
works before fprintd/usbutils are installed and without nagging machines
that have no reader. The setup script reuses the same helper as an early
gate, bailing before installing any packages when no reader is found
(replacing the old post-install fprintd-list probe).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The stock Arch updatedb.conf interacts badly with Omarchy's Btrfs
layout in both directions:
- Snapper snapshots under /.snapshots are nested subvolumes reached by
plain directory traversal, so updatedb indexes the entire system once
per snapshot. On machines that accumulated snapshots this means
multi-hour updatedb runs at full CPU, gigabytes of RAM, and a
multi-gigabyte plocate.db (observed: 18 GB db, 7.5 h runs at 96% CPU
with 592 snapshots; 43 MB and ~1 min after the fix).
- PRUNE_BIND_MOUNTS="yes" treats Btrfs subvolume mounts like /home as
bind mounts, so locate finds nothing in home directories at all.
Configure updatedb.conf at install time and migrate existing installs,
then rebuild the index in the background. Both settings are matched
tolerantly and appended when absent, so a hand-edited updatedb.conf is
fixed rather than silently skipped.
quickshell 0.3.0's qs kill returns before the instance exits, forcing
restart tooling to babysit pids. Upstream fixed kill to block until the
process is gone, but only on master, so the Omarchy repo now carries
quickshell-git pinned past that fix. New installs get it from the base
packages; existing installs migrate with a single --ask 4 transaction so
the conflicting quickshell package is replaced in place, then restart
the shell after the update to pick up the new binary.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Use omarchy-display-text-size 11 instead of a lone GTK text-scaling-factor,
so the XPS 13 nudge moves shell, GTK, and terminal text together.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
graphical.target was gated on network-online.target: cups-browsed orders
itself After=network-online.target and sits in multi-user.target, so the
session waited for NetworkManager-wait-online (DHCP/Wi-Fi association)
before coming up. On the bench VM that was ~140ms, but it's the classic
multi-second boot stall on real Wi-Fi.
Mask NetworkManager-wait-online.service so network-online.target is
reached immediately and nothing in the boot waits for connectivity;
NetworkManager still comes up and cups-browsed browses once the link is
live. Mirrors the existing systemd-networkd-wait-online mask. Enabled at
install (enable-services.sh) and masked on existing installs via
migration.
Verified on a fresh boot: the graphical.target critical chain no longer
passes through network-online.target / NetworkManager-wait-online.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
omarchy-update-user-notify.path used PathExistsGlob= on the packaged
migrations directory. That directive is level-triggered: systemd re-checks
it every time the triggered unit deactivates and fires again while the glob
still matches. Since applied migrations stay on disk forever (state lives in
~/.local/state/omarchy/migrations), the glob always matches, so the oneshot
notifier re-triggered itself in a tight loop (~26-66 starts/sec) — burning
about a core and flooding the journal for the whole session.
The loop existed since the unit was introduced, but the default start-rate
limit killed it after 5 iterations, taking the .path unit down with
'unit-start-limit-hit'. That symptom was reported as #6174 and fixed
yesterday by setting StartLimitIntervalSec=0 — which removed the only brake
and turned the capped hiccup into an unbounded busy-loop.
Fix the actual cause instead:
* Drop PathExistsGlob= from the .path unit, keeping the edge-triggered
PathModified= watch for updates that land mid-session.
* Revert the StartLimitIntervalSec=0 override; with the level trigger gone
there is no self-re-fire to trip the limit, and the default limit is a
useful backstop again.
* Preserve the once-per-login pending check the glob used to provide by
giving the service its own WantedBy=graphical-session.target, enabled at
first-run alongside the other user units.
* Add a migration that daemon-reloads, revives a rate-limit-killed .path,
restarts the watcher, and enables the login-time notifier on existing
installs.
Verified with transient path/service units: the old config runs the service
200 times in 3 seconds; the new config runs it zero times while idle and
exactly once when a new migration file lands.
Thanks to @HANCORE-Linux for finding and diagnosing the problem.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The 2560x1600 panel on the 2026 XPS 13 (DX13260) renders the default
scaling a touch large. Ports f1fa7f69 from master.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The sof-audio-pci-intel-* driver family needs sof-firmware to boot the
DSP, but mainline linux only optdeps it, so affected installs came up
with a Dummy Output sink only. Replace the Panther Lake check with
omarchy-hw-intel-sof, which detects any Intel PCI audio controller.
Ports ffa81021 from master, which adapts basecamp/omarchy#6200.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Bar-widget panels (audio, bluetooth, network, power, monitor) used to be
toggled via their own per-plugin IpcHandler targets. Quickshell resolves
duplicate targets first-handler-wins, so after a plugin or bar reload the
stale handler of the destroyed widget instance kept claiming the target
and the hotkeys went dead.
The shell root's IpcHandler lives outside the reload cycle, so summon,
hide, and toggle now go through `omarchy-shell shell toggle <plugin>`
and the shell routes to the live widget instance via the bar's slot
registry. Panel plugins that are also panel/overlay/menu kinds keep
using the panel loader path. Failures to find a live widget are logged
so a widget missing from the bar layout stays diagnosable.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
