Skip to content

Omarchy Mac 4.0.5rc1 - #765

Open
malik-na wants to merge 49 commits into
quattrofrom
release/4.0.5rc1
Open

malik-na wants to merge 49 commits into
quattrofrom
release/4.0.5rc1

Conversation

@malik-na

@malik-na malik-na commented Oct 24, 2026 •

Copy link
Copy Markdown
Member

Release 4.0.5rc1: the first quattro release since 4.0.3rc4. Bug fixes only, no boot-chain changes.

This branch is exactly the tree that was tested (994f919d). Merge it with a merge commit, not squash or rebase, so the tested bytes and every contributor's commits land unchanged.

Pull requests, in merge order: #747, #744, #737, #356, #458, #460, #461, #462, #420, #421, #502, #663, #718, #410, #414, #759, #752, #711

Maintainer commits:

Testing (VMs on Arch Linux ARM; no Apple hardware):

  • Both goldens hold the exact published lane bytes. A full omarchy update passes from 4.0.3rc4-2 (rc lane) and from 4.0.2-2 (edge lane), and so does a second update. The desktop session starts.
  • All PRs were also tested one at a time, with a full upgrade or a quick overlay check.
  • CI's test suites match quattro's results (no new failures).

Evidence (in the release workspace, automation/evidence/):

  • Combined upgrade runs of the tested tree: 2026-10-24-release-release-next-rc4-051524 and 2026-10-24-release-release-next-edge-051554.
  • With the version commit, built as 4.0.5rc1-1: 2026-10-24-release-release-next-4.0.5rc1-rc4-053928 and …-edge-053958.
  • The exact recovery command from the release notes: 2026-10-24-release-documented-recovery-{rc4,edge}-*.
  • Fresh install of this tree: 2026-10-24-release-fresh-tip-fresh-local-*.
  • CI test suites at quattro and at this tree: 2026-10-24-release-container-tests-full-{ba546a61f,da07df3bf}.
  • One PR at a time: 2026-10-24-release-pr-<N>-*.

Hardware checks owed by rc testers: #356, #502, #752, #711.

Known: existing Macs need a one-time sudo env OMARCHY_UPDATE_PACMAN=1 pacman -Syu --needed --overwrite '/usr/share/omarchy/*' omarchy/hyprland before omarchy update (#741).

jastincheis and others added 30 commits September 6, 2026 18:25
omarchy-battery-status preferred UPower's charge-*-threshold over the
device's own sysfs value. The macsmc-battery driver never calls
power_supply_changed() for a threshold-only write, so UPower's copy
goes stale the moment battery-charge-limit changes it and never
catches up — the panel kept showing the old limit (e.g. 80%) even
after the slider set a new one (e.g. 87%), even though the real
hardware threshold was applied correctly.

Panel.qml already worked around this same staleness for the slider's
own display by reading the CLI instead of UPower; apply the same fix
here by reading sysfs first and falling back to UPower only when
sysfs isn't readable.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Nyix3VJYzz9JEXHPwHPYTK
omarchy-launch-editor already launches nano, micro, hx, and fresh as TUI
editors, but omarchy-default-editor rejected three of them, so they were
supported yet unreachable through the documented interface. Accept them in
the setter (plus the hx spelling of helix) with omarchy-pkg-add install
paths: nano and micro from the repos, fresh-editor from the AUR.

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
…unit

migrations/1785608166.sh treated a reset-failed error as fatal, but on a
unit that never loaded reset-failed is a normal no-op — the update stopped
with a misleading "Could not reset" over nothing. Tolerate it the way
omarchy-restart-audio already does, and skip the live repair entirely when
the manager cannot load the unit (not packaged here, or user-masked): there
is no failed state to clear and nothing to restart. Real inspection,
restart, and stop failures still stay retryable.

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
When ttfx is absent the launch inside the unconditional while loop fails
instantly, the inner pgrep wait sees no process, and the outer loop retries
forever — filling the terminal with errors and never reaching the input or
focus checks. Guard before any state is changed (cursor hide, background
color), report the missing engine once, and exit.

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
The update check counted any commit HEAD was behind upstream, so the bar
icon lit on every push between releases. Resolve the newest tag reachable
from upstream and report only when HEAD does not contain it; keep the
commit count as a fallback for upstreams with no tags.

Generated with [Devin](https://devin.ai)

Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
theme current and theme list print Title Case names such as "Osaka Jade",
but theme dir appended its argument verbatim and printed a nonexistent
path with exit 0. Slugify the same way theme set does, prefer a
user-installed copy, and fail when the theme is missing.

Fixes #427
pgrep -x foot/ghostty printed matching PIDs, and -g with no glyph made
omarchy-notification-send consume the headline. Quiet pgrep and pass a
real headline.

Fixes #430
omarchy-lazyvim-setup still symlinked ~/.config/omarchy/current, which
the theme-state migration moved to ~/.local/state/omarchy/current.
Fresh nvim configs then failed to load plugins.omarchy-theme.

Fixes #433
mic.sh treated /run/user/$UID/bus as enough to run systemctl --user.
A logged-in --resume has that socket, but sudo -i clears XDG_RUNTIME_DIR,
so the user-bus call aborted install.sh. Only contact the user manager
when XDG_RUNTIME_DIR actually points at it; enablement already persists
via the graphical-session.target.wants link.

Fixes #443
loudnorm declares a 192 kHz output rate, and the AAC encoder then writes
96 kHz. Playing that file switches the Asahi speaker PCM to 96 kHz and
speakersafetyd panics, locking the amps. wf-recorder is already passed
-R 48000; this finalize pass was rewriting it.

Fixes #501
The shell answers a ping before it owns org.freedesktop.Notifications, so the next notification fails and installers exit. Relock a stranded session first, then wait for the name when it was present before the restart.
omarchy-audio-input-set-default moved every source-output to the chosen
input, including asahi-audio's own DSP capture stream (audio_effect.jNNN-mic,
the input half of the mic filter chain). That stream carries
node.dont-fallback, so a target it cannot link to makes WirePlumber destroy
it, and the filter-chain module unloads itself with it: picking the built-in
microphone in the audio panel removed effect_output.jNNN-mic until
WirePlumber was restarted (#700).

Port upstream's rule: skip source-outputs that belong to a filter or loopback
(node.link-group) or that set node.dont-move, and only move application
streams. Refuse a name that is not a source at all: the panel listed the mic
mapping's null sink as an input because Pipewire.defaultAudioSource was that
sink node, and wpctl set-default on a sink id silently changed the default
output, feeding the speakers into the mapping. The panel now also leaves a
sink node out of the input list.

Verified on a MacBook Pro 14" M1 Max (j314): selecting effect_output.j314-mic
through the script keeps the DSP source and its RawMics target; passing the
mapping's sink id exits 1 and leaves the default sink unchanged.

Fixes #700

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
omarchy-menu-keybindings recovers Lua binds by running hyprland.lua under a
stub hl whose catch-all __index returns a noop proxy table. The stock tree
that config loads first calls hl.get_active_monitor() and compares a field
of the answer with a number (qconsole.lua:64), and a table there raises
"attempt to compare table with number". pcall swallows it, but the dofile
has already died before hyprland.lua reaches require("hypr.bindings"), so
every binding a user defines shows in Super+K with its description and no
command, and cannot be run from the menu (#685).

Make get_active_monitor and get_active_window answer nil, which every
caller in the stock config already guards for (qconsole's fit(), clipboard's
active_window_is_terminal()), so the load completes and user binds resolve.
get_config already answered nil for the same reason.

The new test loads the real stock tree under the scanner from a temporary
home with one user bind and checks that bind comes back with its exec
command; it fails on the previous script and passes now. Verified on a
MacBook Pro 14" M1 Max: DEBUG=1 no longer reports the scan failure.

Fixes #685

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
omarchy-update-confirm still linked basecamp/omarchy releases, which
describe upstream versions rather than the omarchy-mac packages this
update installs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
fwupdmgr exits 2 when no updates are available, which set -e turned into
a failure. On Apple Silicon this happens on every run, since fwupd cannot
update Mac system firmware, so also point Mac users at macOS updates.
Edge stopped publishing hyprtoolkit and hyprland-guiutils for aarch64, so
the hardcoded omarchy/ targets fail every install and update with "target
not found". Select only what the synced edge database carries and let the
regular repositories supply the rest; with no readable edge database, keep
selecting all three.

Fixes #741

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The locked binding once read true after an unlock, though the request,
the session lock, and its secure state were all false. A lock before
suspend then took the session for locked and did nothing, and the
machine slept unlocked. Lock requests, the status reply, and stranded
lock recovery now read those three values directly.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
(cherry picked from commit d1a614c)
Qt 6.12 adds a Color type to QtQuick that shadows the qs.Commons Color
singleton, so every unqualified Color.* lookup resolves to undefined and
the bar, launcher, menus, notifications, lock screen and polkit dialog
render without colors (the launcher shows as a black box).

Port of upstream omacom#14553: import qs.Commons as Commons and
reference the palette as Commons.Color.*, keeping the singleton's name so
plugins written against upstream keep working.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
omarchy update runs under script(1), so the orphan check always sees a
terminal and stops at "Remove N orphaned package(s)?" even after -y
promised not to ask. An unattended run, including the one channel
switching starts, waits there forever. With OMARCHY_UPDATE_UNATTENDED
set, list the orphans and keep them, as the non-terminal path already
does. An interactive update still asks.
A fresh install computes the Hyprland targets before its first sync, so
the [omarchy] database is not there yet and the selection falls back to
all three. Edge no longer publishes hyprtoolkit or hyprland-guiutils for
aarch64, and the install stops at "target not found". Sync first so the
selection sees what edge carries; the -Syu follows immediately.
fix-arm-packages.sh chose its targets before it had added or synced the
[omarchy] repository, so on the machines it exists for the selection
fell back to every target and stopped at the packages edge no longer
publishes. Sync after preparing the sources and choose then, with stdin
kept away from pacman, since the documented pipe puts the script there.
The dry run still reports what the machine's databases select now.

The recovery test's pacman stub prompted on every call without
--noconfirm, which a repository listing never does; it now answers the
listing as edge does today and checks the sync comes before the choice.
The first quattro release since 4.0.3rc4: bug fixes only, on the same
unsigned RC4 trust bootstrap. The boundary test pinned the version to
4.0.3rc4; it now accepts the 4.0.5 line and keeps its checks that no
strict-signing migration ships.
@malik-na
malik-na requested a review from maralcbr October 24, 2026 01:07
@greptile-apps

greptile-apps Bot commented Oct 24, 2026 •

Copy link
Copy Markdown

RetriggerView in GreptileConfidence Score: 4/5

[Medium impact] The changes reach package upgrades and several everyday desktop workflows.

The PR still needs the channel-switch package selection fix before merging.

Findings

  1. P1 Channel switches use stale packages ▶

Summary

This release collects fixes for updates, audio, shell restarts, themes, and hardware setup.

Changes since the previous review:

  • ARM installs and recovery choose Hyprland packages from the synced repository.
  • Audio input choices leave speaker output and the built-in Mac microphone alone.
  • Normalized screen recordings keep their audio at 48 kHz.
  • Battery status reads charge limits from the device’s current settings.
  • Lock decisions use the session’s current lock state.
  • Shell restarts restore a needed lock before waiting for notifications.
  • Unattended updates skip orphan prompts and release checks use the right channel.
  • Firmware updates report their results through the update workflow.
  • Font changes complete with the right settings and shell refresh.
  • Editor choices accept more names and LazyVim setup handles them.
  • The keybindings menu includes the user’s own key bindings.
  • Theme commands resolve the active theme directory reliably.
  • The screensaver continues when optional font files are missing.
  • Shell controls and panels use explicit references to the shared color palette.
  • The version file marks this release as 4.0.5rc1.

No new actionable issues were found in these changes.

Reviews (2) · Last reviewed commit: "Document the one-time update for Macs st..." · Reviewed by Greptile

Comment on lines +12 to +16
if ! edge=$(pacman --config "${OMARCHY_PACMAN_CONFIG:-/etc/pacman.conf}" -Sl omarchy 2>/dev/null) || [[ -z $edge ]]; then
edge=$(printf 'omarchy %s\n' hyprland hyprtoolkit hyprland-guiutils)
fi
for package in hyprland hyprtoolkit hyprland-guiutils; do
if awk -v name="$package" '$2 == name { found = 1 } END { exit !found }' <<< "$edge"; then

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1 Channel switches use stale packages

omarchy_arm_package_targets now chooses packages from the live database, but channel switches refresh only their separate, staged database before calling this helper. If the live database still lists a package that edge stopped publishing, the helper adds that removed target to the staged transaction. The switch then fails with “target not found,” even though its refreshed database has the correct list. Let channel preparation select targets from its staged configuration and database.

@malik-na
malik-na requested review from scottjones and removed request for dhh, maralcbr and ryanrhughes October 24, 2026 01:12

@maralcbr maralcbr left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks Naeem, this is a careful release branch: the merge order is clear, every PR has its own evidence, and the maintainer commits close real gaps (the database sync before choosing the Hyprland targets in install.sh and fix-arm-packages.sh, and -y no longer stopping on orphans). One thing before merge, plus CI:

  • The sleep-lock migration now retires a repair it hasn't done. migrations/1785608166.sh:55 treats every LoadState other than loaded as nothing to repair, so error and bad-setting exit 0 and omarchy-migrate writes the completion marker. On quattro the failing reset-failed/restart exited 1 and the migration stayed pending, which is what agents/skills/migrations.md asks for. The comment says the intent is units that were never packaged or are masked, so please skip only not-found|masked, keep loaded on the repair path, and exit 1 for anything else. A bad-setting case in sleep-lock-environment-migration-test.sh that checks the exit status and the missing marker would pin it.
  • CI hasn't reported at da07df3b. Both main.yml and install-vm.yml run on pull requests, but only Greptile shows on the head. Please get both to run before merge, ideally on the head that carries the migration fix.

Smaller things, fine as follow-ups before 4.0.5 final:

  • Channel switches still choose targets from the live database. install/helpers/arm-channel.sh:284 calls omarchy_arm_package_upgrade_args against /etc/pacman.conf, while the transaction resolves against $stage/db, synced at :272. If edge drops a target between a Mac's last sync and a channel switch, the stale live database still selects it and the staged -Sup fails with "target not found". Passing the stage database to the selection would fix it. The ordinary update is fine, since omarchy-update-keyring:13 syncs first.
  • The LazyVim path fix only reaches new configs. omarchy-lazyvim-setup exits early when ~/.config/nvim exists (:21-23), so anyone who already ran it keeps omarchy-theme.lua pointing at ~/.config/omarchy/current/theme/neovim.lua. Both earlier relink migrations only touch theme.lua. A narrowly matched migration for omarchy-theme.lua would fix it.

Nits:

  • bin/omarchy-screensaver:19 guards ttfx, which is a default package. On a Mac I think the guard is right, since the installer skips default packages that fail to install (install.sh:284), but a one-line comment saying so would explain the exception to the AGENTS.md rule.
  • bin/omarchy-theme-dir:25: plain $OMARCHY_PATH rather than ${OMARCHY_PATH:-}, per the runtime rule.
  • tests/test-update-firmware.sh would sit better in test/shell.d/ with the shared helpers.

Checked and fine: the Qt 6.12 palette port leaves no unqualified Color. references in shell/, every lock decision in lock/Service.qml now goes through currentlyLocked(), and the audio input command refuses non-sources and leaves filter captures alone. Running ./test/all in a Debian container with CI's packages gives 3414 ok against quattro's 3386, with the same environment failures as quattro; the new user-binds test needs gawk, which Arch ships. tests/all and omarchy commands --check (472 commands) pass.

For release (not the source merge): the rc notes need the one-time sudo env OMARCHY_UPDATE_PACMAN=1 pacman -Syu omarchy/hyprland for Macs still on the old helper (#741), and the hardware checks you listed for #356, #502, #752 and #711 are still owed by rc testers. Since the migration fix changes the tested tree, a quick re-run of the rc and edge upgrade lanes at the new head would keep the evidence matching what merges.

Comment thread migrations/1785608166.sh Outdated
echo "Could not inspect omarchy-sleep-lock.service: $load_state"
echo "The pre-suspend lock repair will be retried by omarchy-migrate."
exit 1
elif [[ $load_state != "loaded" ]]; then

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This also skips error and bad-setting, and the migration is then marked done. Please skip only not-found|masked and exit 1 for anything other than loaded, so a real load error stays pending as it did before.

Comment thread bin/omarchy-lazyvim-setup
# Wire LazyVim to follow the current Omarchy theme
mkdir -p "$HOME/.config/nvim/lua/plugins"
ln -sf "$HOME/.config/omarchy/current/theme/neovim.lua" \
ln -sf "$HOME/.local/state/omarchy/current/theme/neovim.lua" \

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Right for new configs, but existing ones exit at line 23 and keep the old omarchy-theme.lua target; a small migration would repair those.

Comment thread bin/omarchy-screensaver
# Without ttfx there is nothing to draw, and a respawn loop over a missing
# command paints the terminal with launch errors. Say so once and leave
# before any state — hidden cursor, background color — needs restoring.
if omarchy-cmd-missing ttfx; then

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: a comment that Mac installs can miss default packages (install.sh:284) would explain why this one is guarded.

Comment thread bin/omarchy-theme-dir
if [[ -d $HOME/.config/omarchy/themes/$theme ]]; then
echo "$HOME/.config/omarchy/themes/$theme"
else
elif [[ -d ${OMARCHY_PATH:-}/themes/$theme ]]; then

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: $OMARCHY_PATH is always set; no need for :-.

The load-state check skipped the live repair for every LoadState other
than loaded, so a unit systemd found but could not load (error,
bad-setting) exited 0 and omarchy-migrate marked the repair complete
without having fixed anything.

Only not-found and masked mean there is nothing to repair. Any other
unloaded state now reports what it found and exits non-zero, so the
migration stays pending and is retried, as it was before the skip was
added.
Outside a checkout, resolve_helper took the helper installed under
/usr/share/omarchy (or OMARCHY_PATH, which points there) before fetching
the published one. On a Mac stuck on 4.0.2-2 or rc4 that installed
helper still hardcodes omarchy/hyprland, hyprtoolkit and
hyprland-guiutils, so the documented curl | bash recovery asked edge for
packages it no longer publishes and died with "target not found".

A checkout still uses its own helper. Otherwise the downloaded helper
wins, and the installed copy is kept only as a fallback when the fetch
fails: a current installed helper still recovers the machine, a stale
one fails no worse than having none, and the existing test that serves
the published copy from a file keeps exercising the download path. The
fallback says it is being used so a failure there can be read.
A Mac already on 4.0.2-2 or 4.0.3rc4 cannot update itself past #741:
the installed updater still asks edge for hyprtoolkit and
hyprland-guiutils, which it no longer publishes, so the transaction
that would replace that updater fails first. Give those users the one
manual transaction that gets them onto the fixed package, with the same
--needed and --overwrite '/usr/share/omarchy/*' that the regular update
passes, under the error text they will search for.
@malik-na

Copy link
Copy Markdown
Member Author

Thanks Marcelo. Three commits are on top of da07df3:

  • 07c6c5adb — the sleep-lock migration now skips only not-found|masked, repairs loaded as before, and exits 1 for any other LoadState, so it stays pending. A bad-setting case in sleep-lock-environment-migration-test.sh runs the real omarchy-migrate and checks the exit status and that no marker is written.
  • 741c50808 — fix-arm-packages.sh, when piped, now prefers the helper it downloads over the copy installed in /usr/share/omarchy (the installed copy is only a fallback when the download fails, with a warning). A Mac already on the old helper can be unstuck by the README one-liner once this is on quattro. Two tests, one with an old-style installed helper.
  • f8433900d — README: a troubleshooting section with the one-time command for a Mac already stuck on target not found.

Re-ran the rc (4.0.3rc4-2) and edge (4.0.2-2) upgrade lanes and a fresh install at f8433900d: all pass, with the desktop session up and a second update a no-op.

Follow-ups before 4.0.5 final, as you suggested: channel switches choosing targets from the staged database, a migration for existing omarchy-theme.lua links, and the nits.

CI: no workflow has run on this repository since 2026-10-08, not even the daily schedule, so this looks like the org's Actions billing rather than the triggers. I'm checking it.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.