Skip to content

Merge upstream quattro (China mirrors, captive portal, notification trim) - #91

Merged
mark-groves merged 19 commits into
quattrofrom
cursor/upstream-changes-pr-e737
Oct 23, 2026
Merged

mark-groves merged 19 commits into
quattrofrom
cursor/upstream-changes-pr-e737

Conversation

@cursor

@cursor cursor Bot commented Oct 22, 2026

Copy link
Copy Markdown

Upstream quattro moved from 7075b61c (already on the fork via #90) to 967bbdd7 (2026-10-22 20:58 UTC). 19 first-parent commits, all non-merges, cherry-picked onto fork quattro. Cursor stays on the official tarball. Face, Grok Bot, and Cursor menu entries stay. The power-profile menu provider stays out. The kernels provider stays. Fingerprint setup still goes through omarchy-pam-pair-add and omarchy-apply-polkit-pam. Factory reset still scrubs inherited USBGuard and Thunderbolt trust. The icon font charset is now e900-e90f for the Tailscale mark at U+E90F. Cursor stays at U+E90D. The fork overlay is still 231 paths, the same set and the same hunks against this tip.

Installs in China use Omarchy's China mirrors (omacom#16256)

A timezone in China (Shanghai, Urumqi, and tzdata's older names for them) selects the cn region. Pacman then lists omarchy.cn and pkgs.omarchy.cn/<channel> first, with the global .org servers still behind them as fallback. The packages are the same signed set. Global installs are unchanged, and an existing install stays global until someone opts in. Hong Kong, Macau, and Taiwan map to no region. ArchLinuxCN is not added. omarchy-apply-pacman renders the channel and regional files as the caller and installs them with the same root copies refresh already used. A dev checkout that lacks this machine's region profile is refused before the link, so a switch cannot drop the regional servers. Migration is not required; the marker is /etc/omarchy/region.

The clock and the weather ask Omarchy (omacom#16247, omacom#16240)

timesyncd's fallback is ntp.omarchy.org, then time.cloudflare.com. The drop-in is /usr/lib/systemd/timesyncd.conf.d/10-omarchy.conf, so an admin file in /etc still wins, and it takes effect the next time timesyncd starts. With no stored place, the weather panel and the installer's timezone guess ask geo.omarchy.org (a JSON place with coordinates and a timezone) instead of wttr.in's location lookup and tzupdate's seven geo-IP services. A stored location, and a timezone the install media already set, still win. If the lookup cannot be reached, weather falls back to wttr.in and the installer falls back to the full timezone list.

Captive-portal sign-in, and connectivity on ping.omacom.io (omacom#16186)

NetworkManager's connectivity check is ping.omacom.io. The migration 1792507034 rewrites an existing check and reloads NetworkManager; a failed reload leaves the migration pending. Signing in opens a floating window titled for a network (wired portals use it too) and closes it once the link is online. A Tailscale exit node is paused for that window and restored when the machine is back online, or if the window never opens. The Tailscale menu rows and those notifications use the new Omarchy-font mark.

Apple Silicon pacman drops the unused [aur] repo (omacom#16189)

ALARM's [aur] repository is not in the Apple Silicon template. A Mac's pacman.conf only names repositories Omarchy installs from, matching the other aarch64 templates.

A non-Latin layout keeps a Latin us in front (omacom#14076)

A layout list that already contains us moves that entry to the front instead of adding a second one. us(rus) and us(chr) do not count: they cannot type Latin, so a plain us is still prepended. Keybindings and the greeter password keep a Latin layout in the lead.

The keybindings scan survives a config that reads Hyprland at load (omacom#7564, omacom#16276)

The Lua bind scan stubs hl so a user's hyprland.lua can be read without a compositor. Numeric walks and ipairs now end, comparisons no longer abort the scan, and a runaway config stops on an instruction limit instead of spinning until the 10s timeout. The cache key is v19, so a scan truncated by the old stub is redone.

A menu file read mid-write keeps the rows already on screen (omacom#16268)

A half-written or empty read of the shipped menu, or of the user's extension while it is being saved, keeps the last good rows. A finished file replaces them. An emptied user extension is dropped. A shell that started during an update reloads the shipped menu if it never got a copy.

A large tailnet no longer rebuilds every machine row (omacom#7623)

The Tailscale panel virtualizes the machine list, shares one copy menu, and searches with /. While the panel is closed it polls with --peers=false and fetches the full list on open. An unchanged poll rebuilds nothing, and a poll no longer jumps a scrolled list back to the top.

fcitx5 stays enabled and inert when the binary is absent (omacom#11694)

The user unit now has ConditionPathExists=/usr/bin/fcitx5. A trimmed image that ships the unit without the binary no longer restarts it every two seconds for the whole session.

Battery copy, and the idle toggle (omacom#16272, omacom#16271)

A full battery says Fully charged, and a notification leaves out a reading it does not have. A negative power figure is treated as a live draw. The menu rows that clipped are now Show/Hide Battery %, and the idle toggle reads Stay Awake / Allow Sleep.

Notifications are shorter, and they appear at once (omacom#16277, omacom#16296, omacom#16301, omacom#16306)

Phone-number and meeting buttons are gone. Links in the body are the link. A code is copied directly. Snooze is one line, with a button beside the close mark, and Do Not Disturb is no longer in that menu. Cards no longer route away from a fullscreen app. Age shows up once a card is five minutes old. Arrival, height changes, and dismissal jump instead of sliding and fading. Critical titles stay in the normal text colour. Taildrop starts after the session environment is in, so the title is translated; the save path reads ~/Downloads again; image files preview in the card, with the send glyph still underneath if the image fails. Migration 1792648436 clears older notification layout leftovers once, instead of at every startup.

The indicator peek stays shut beside the clock (omacom#16293)

Hovering in the gap to the right of the clock no longer opens the indicator peek. While the peek is open, indicators revealing beside a notch no longer pull the hover edge across a still pointer and flap the peek.

Shell tests stay out of the real home (omacom#16229)

base-test.sh unsets the XDG base directories so a test that points HOME at a scratch directory cannot edit the real config through them. Background pixel tests skip when a window covers the first output. The i18n check is structural unless I18N_STRICT=1. AGENTS.md now says when a test is worth adding, and to read a failure from the suite log instead of rerunning the file to see it.

Open in Web View Automation 

manuaudio and others added 19 commits October 22, 2026 23:04
…g it (omacom#14076)

* Move us to the front of a non-Latin layout list instead of duplicating it

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Keep a plain us in front when the list's us can't type Latin

us(rus) is Russian phonetic and us(chr) is Cherokee. Moving one of those to the front, as XKBLAYOUT=ru,us with XKBVARIANT=,rus would, leaves no Latin layout leading, so the session's keybindings and the greeter's password both stop working. Only reuse a us entry whose variant types Latin; otherwise prepend plain us as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Codex Medium <noreply@openai.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com>
Co-authored-by: Codex Medium <noreply@openai.com>
…les for agents (omacom#16229)

* Keep shell tests out of the real home's XDG directories

* Keep the session's cursor out of the startup cursor test

* Capture the remote session test frame from the first output

* Run the i18n test structurally by default, strict on request

* Skip background pixel tests while windows cover the background

* Keep shell test logs and tell agents how to avoid rerunning tests

* Tell agents when a test is worth adding
ALARM's [aur] repository serves 12 packages, and none of them is
among the 171 that Omarchy installs on a Mac (the base, aarch64 and
aarch64-apple package lists). The generic aarch64 templates dropped
it in omacom#16065; this does the same for Apple Silicon, so a Mac's
pacman.conf only names repositories Omarchy uses.

The Mac image builder (omarchy-mac-installer) refused a template
without [aur] until omacom/omarchy-mac-installer#113, so this must
land after that.
timesyncd's compiled-in fallback is 0-3.arch.pool.ntp.org, so every
Omarchy machine asks Arch's NTP Pool vendor zone for the time. A
vendor drop-in in /usr/lib/systemd/timesyncd.conf.d sets FallbackNTP to
ntp.omarchy.org, then time.cloudflare.com if that does not answer.

ntp.omarchy.org is a name we control, so where it points can change
without an Omarchy update: today a CNAME to 2.arch.pool.ntp.org (the
pool's IPv6 zone, the same servers as now), later Omarchy's own vendor
zone. FallbackNTP stays below any NTP= an admin sets and any server a
network hands out through systemd-networkd.

The drop-in lives in /usr/lib, not /etc, so a user's own drop-in in
/etc overrides it and pacman never leaves a .pacnew. omarchy-settings
installs it on both architectures (omacom/omarchy-pkgs, same change).
It takes effect when timesyncd next starts, at the next boot.
…hy.org (omacom#16240)

* Find the weather location with geo.omacom.io instead of wttr.in

With no location stored, the weather panel asked wttr.in where the
machine is twice: once for the forecast (wttr.in/?format=j1) and once
for the city's name (wttr.in/?format=%l). omarchy-weather-location
asked it again for the name.

geo.omacom.io is Omarchy's own lookup, a Cloudflare Worker that answers
from Cloudflare's view of the network in the shape of weather.json:

  {"name":"Amsterdam","latitude":52.37403,"longitude":4.88969,
   "country":"NL","timezone":"Europe/Amsterdam"}

so parseLocationFile reads it as is. The panel now asks it first and
uses its coordinates, through Model.effectiveLocation, for Open-Meteo and
wttr.in alike. The first fetch waits for the answer rather than let
wttr.in look the machine up meanwhile. If geo.omacom.io cannot be
reached, the panel falls back to wttr.in's own lookup as before. A
stored location still wins over the detected one.

In a 4.0 edge VM (a17a001, emulated), the shell's fetches after a
restart with no location stored:
  before: wttr.in/?format=j1, wttr.in/?format=%l, then Open-Meteo
  after:  geo.omacom.io, then Open-Meteo and wttr.in/52.37403,4.88969
With geo.omacom.io unreachable it fell back to wttr.in/?format=j1, and
storing and clearing a location fetched the same as before the change.

* Guess the installer's timezone with geo.omacom.io instead of tzupdate

The setup form's timezone guess ran tzupdate, which asks seven geo-IP
services at once (geoip.chrisdown.name, ipbase, ipapi.co, worldtimeapi,
reallyfreegeoip, ipwho.is, ipinfo). It now asks geo.omacom.io, Omarchy's
own lookup, for the timezone. When it cannot be reached the prompt
falls back to filtering the full list, as it already did offline, and a
timezone preset by the install media still wins.

In a 4.0 edge VM with the real curl, jq and timedatectl, the prompt
preselected Europe/Amsterdam from geo.omacom.io.

* Ask geo.omarchy.org for the location instead of geo.omacom.io

The geo Worker now answers on omarchy.org, next to Omarchy's other
services such as ntp.omarchy.org. Same Worker, same JSON. geo.omacom.io
no longer resolves.

* Show wttr.in's weather icon when Open-Meteo fails for a detected location

The panel let wttr.in fill an empty icon only when no coordinates were
known. Once geo.omarchy.org supplied coordinates for an install with no
stored location, a failed Open-Meteo fetch left the icon empty, and the
bar hides the weather widget when it has no icon.

Gate the fallback on a stored location's coordinates again, as before
geo.omarchy.org: wttr.in fills the icon until Open-Meteo answers, and
Open-Meteo's icon replaces it when it does.
…ig reads (omacom#7564)

* Make the keybindings menu's Lua bind scan safe against load-time config reads

omarchy-menu-keybindings learns about Lua-only binds by running the user's
hyprland.lua under a stub `hl` whose every lookup answers a truthy noop
table. That stub breaks as soon as a config reads compositor state at load
time, which real configs do:

- `for _, m in ipairs(hl.get_monitors())` never terminates, because noop[1],
  noop[2], ... are all non-nil. The menu exits, the `lua` child keeps
  spinning at 100% CPU under systemd --user until the next reboot. One
  such orphan ran for 27 hours on a laptop after a single SUPER+K.
- `monitor.scale <= 0` (default/hypr/qconsole.lua:64) raises "attempt to
  compare table with number", which aborts the pcall'd scan, so every bind
  declared after that file is silently missing from the menu.

The stub now answers nil for numeric indexes (so ipairs/`#` terminate),
reports length 0, and degrades comparisons, arithmetic, and concatenation
to harmless values instead of raising. The scan is also wrapped in
`timeout 10` so a config the stub still cannot satisfy can never outlive
the menu.

Adds a shell test that scans a fixture config doing exactly those
load-time reads; without the fix it hangs for the full timeout.

* Bump the keybindings cache version so existing caches are rescanned

The cache key hashes only the version, the active keymap and `hyprctl binds`, so a records file written by the scan that aborted in qconsole.lua keeps being served after this fix lands, and the menu goes on dispatching the stale default in place of the user's override until their binds happen to change.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Codex Medium <noreply@openai.com>

* Stop a runaway config inside the bind scan instead of at the timeout

A config the stub cannot end held the menu for the full 10 s before the timeout killed the scan. An instruction-count hook now raises inside the pcall instead, after 50M instructions (about 0.3 s on a worker; the stock config uses about 69k), and the scan reports the binds it found, as it already does for one that errors part way. That result is cached like theirs: a config that runs away under the stub will almost always do so again, and rescanning would only repeat the wait. The timeout stays as the backstop for what the hook cannot reach: a loop in a coroutine, or one that catches the error and keeps going.

Co-Authored-By: AnPod <andrej@podgorsek.de>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Test that a window walk at config load leaves the menu's Lua binds intact

Runs the whole menu under a timeout with a plugin-style ipairs(hl.get_windows()) loop ahead of the stock config, and checks the Lua-only Terminal bind still resolves, which it can only do if the scan got past the loop to that bind.

Co-Authored-By: nekomario28 <206304251+nekomario28@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Run the scan test's lua with -E, as the menu does

Otherwise a LUA_INIT in the tester's environment runs ahead of the fixture, and a LUA_PATH changes what a config's require finds, either of which can fail a test the menu itself would pass.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Check the bind scan finishes without a wall-clock budget

The scan tests failed if a scan took five seconds or more, which a slow or loaded machine can hit with nothing wrong. A scan that never ends is already caught without a clock: run directly, the test's timeout kills it and fails on the exit status; run through the menu, the menu's own timeout stops it before the Terminal bind the test looks for. The outer timeouts are now long enough that only a hang reaches them. The approach is omacom#15704's.

Co-Authored-By: Dylan Kenney <dylan@dkenney.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Keep indexing a stub collection working while ipairs over it ends

Answering nil for every numeric index ended `ipairs(hl.get_monitors())`, but it also made `hl.get_monitors()[1].name` raise, which the scan read before this change. The scan then aborts and every bind declared after that line is lost. Make `ipairs` see the stub as empty instead, so loops still end without running their bodies against stand-in values, and indexing reads a stub again.

Co-Authored-By: Codex Medium <noreply@openai.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: omarchybot <317366263+omarchybot@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Codex Medium <noreply@openai.com>
Co-authored-by: AnPod <andrej@podgorsek.de>
Co-authored-by: nekomario28 <206304251+nekomario28@users.noreply.github.com>
Co-authored-by: Dylan Kenney <dylan@dkenney.com>
* Keep the menu's rows when a source file is read mid-write

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Give each changed menu source a fresh reread budget and treat an empty catalog as torn

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Simplify the torn menu read fix to keeping the last good rows

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…sign-in flow (omacom#16186)

* Drop the help line under the Open Captive Portal button

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Show -- for ping and packet loss until the internet is reachable

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Check connectivity against ping.omarchy.org instead of Arch's endpoint

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Keep reporting lost pings while the internet is reachable

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Leave the connectivity migration pending when the reload fails

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Pause the Tailscale exit node and re-check every 2s while signing in to a portal

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Hold the portal's exit node pause until online and let one click own it

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Move the connectivity check to ping.omacom.io, outside any HSTS scope

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Use the Tailscale glyph on the exit node notifications

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Add the Tailscale mark to the icon font and use it for Tailscale

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Sign in to captive portals in a floating window that closes once online

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Draw Omarchy icon font glyphs in notifications with the icon font

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Call the portal button Sign into Wi-Fi and drop the hero's sign-in line

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Translate the exit node notifications through the CLI catalog

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Read nmcli's connectivity in the C locale and close a portal window that maps late

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Call the portal button Sign into network, since wired portals use it too

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Test the connectivity check migration's NetworkManager reload

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Launch the portal window in the background so the sign-in wait starts at once

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Stop and restore the exit node when the portal window fails to open

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Keep a freshly started browser's failure visible to the sign-in command

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…esn't have (omacom#16272)

* Battery notification says Fully charged and leaves out readings it doesn't have

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Treat a negative power reading as a live draw, not idle

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…e shell (omacom#7623)

* tailscale: survive huge tailnets by virtualizing the machine list

On a corporate tailnet with ~3,900 machines the tailscale widget makes the
shell unusable: `tailscale status --json` is ~4.6MB, and every 30s poll
reassigns `peers`, rebuilding one eager row per peer - each with its own
copy-menu Popup - on every bar. That is ~500k QML object creations per
refresh: ~4.6s of shell CPU per poll, ~15% average CPU, multi-second
main-thread stalls, and delegate trees that ratchet RSS into the gigabytes
over a workday.

Rebuild the hot paths so tailnet size stops mattering:

- Render machines through a clipped, virtualized ListView (reuseItems)
  instead of a Repeater in a Column, so only the visible rows exist.
- Add a search field over the machines (`/` focuses it, Escape clears),
  with the match count in the section header - a 4k-row list is not
  scrollable without one.
- Share one copy-menu Popup across rows instead of instantiating one per
  peer.
- Only reassign peers/exitNodes/mullvadRegions when their content actually
  changed, so an unchanged poll rebuilds nothing.
- Poll with --peers=false (~3KB) while the panel is closed; fetch the full
  peer list on panel open and explicit refresh.

Measured on a 3,933-peer tailnet with two bars: a full refresh drops from
~4.6s of shell CPU (plus ~7MB retained) to ~0.17s, steady state from ~15%
average CPU to under 1%, and RSS holds at ~520MB instead of climbing past
7GB. Keyboard navigation, the copy menu, taildrop send, and small-tailnet
behavior are unchanged; the cursor and copy actions operate on the filtered
list.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* Queue full-peer fetches behind in-flight light polls; reset shared copy cursor

Review follow-ups: a full-peer request that arrived while a --peers=false
poll was still running was silently dropped, leaving a freshly opened panel
stale for up to a whole refresh interval - remember it and launch the full
fetch when the light process exits. And the shared copy menu kept the cursor
position from the previous machine's menu, so Enter could copy an unexpected
field - reset it whenever the menu opens for a different peer.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* Keep the machine list still under background polls; fix search-field edges

Review follow-ups. Qt resets contentY whenever a JS-array model is
reassigned, and on a large tailnet any poll where a peer changed publishes
a new array - so the machine list scrolled itself to the top under the
user's finger, roughly every refresh while the panel was open. Remember
the position the user (or keyboard cursor) chose and restore it on model
change; filter changes and panel opens reset it explicitly.

Also: clear the search query when its field disappears (small-tailnet
account switch) or the panel closes, so an invisible filter cannot hide
machines; make showPeers filter-aware so the keyboard cursor cannot enter
a section that renders nothing; consume Up in the search field instead of
letting it move the section cursor; show "Loading machines" rather than
"No machines found on this tailnet." before the first full peer fetch has
answered; and publish mullvadExitNodes unconditionally so it cannot go
stale behind the regions signature guard.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* Clear peersLoaded when the tailnet goes down

peersLoaded was set on every full status parse and only ever cleared by resetUnavailable, which a normal stop/start never reaches: `tailscale down` parses fine and exits zero, so it lands in the light-poll branch that clears peers and leaves the flag set. Stop Tailscale, start it again while the panel is closed so only light polls observe the transition, and the next open has peers empty with peersLoaded still true — the panel then states "No machines found on this tailnet." for the whole of the full fetch, which is the claim peersLoaded was added to avoid.

Tie the flag to what it names instead: a full poll sets it only when the tailnet is actually running, and the light-poll branch that clears peers clears it too.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Codex XHigh <noreply@openai.com>

* Scope the remembered scroll position to its account

rememberedY was reset on panel open, filter changes and explicit
positioning, but survived the peer set becoming a different tailnet:
scroll deep into a 4000-machine account, switch away and back, and the
list restores a position in a machine list the user has never seen.
Reset it (and the peer cursor) when selectedAccountId changes, so an
account switch starts at the top the way it did before the scroll fix.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* Keep a failed account poll from resetting the machine list

`tailscale switch --list --json` publishes an empty selected id on any non-zero exit — including when `pollWatchdog` reaps it — while the user is still on the same tailnet, and the recovery poll then publishes the id back. Reading either edge as an account switch scrolled a hand-scrolled list to the top, which is the scroll loss this branch exists to prevent arriving through a different signal. Compare against the last account actually seen instead, so only a move between two real tailnets resets.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Omabot <omabot@omarchy.org>
Co-authored-by: Codex XHigh <noreply@openai.com>
…com#16276)

omacom#7564 made the stub answer every index so that `hl.get_monitors()[1].name` keeps reading, but that also means `while monitors[i] do i = i + 1 end` never finds a nil. The instruction limit stops it, and every bind declared after the loop is lost. Answer indexes 1 and 0 with the stub and every other number with nil: the first element and `list[#list]` (0, since the stub's length is 0) still read, a walk by index ends after a step, and `ipairs` still sees the collection as empty. The cache version moves to v19 so scans truncated by such a loop are redone.

Co-authored-by: omarchybot <omarchybot@users.noreply.github.com>
Co-authored-by: Codex Medium <noreply@openai.com>
Co-authored-by: muse-spark-1.3-contributor <noreply@openrouter.ai>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Keep the fcitx5 unit inert when fcitx5 isn't installed

fcitx5 is in omarchy-base.packages, but a trimmed or derived image can
ship the unit without the binary, and first-run/enable-user-units.sh
enables this unit unconditionally either way. With Restart=always and
RestartSec=2, that install spends its whole session failing at
203/EXEC twice a second.

On the aarch64 image I found this on, the unit had restarted 20,943
times in one session and accounted for 105,828 journal lines, which is
most of what the log contains.

enable-user-units.sh already says ConditionPath* in the unit files is
what keeps an enabled unit inert where it does not apply, and
omarchy-tailscale-receive.service guards its binary exactly this way.
This unit was missing its half of that.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ALLhsWEVznJX5Js4CM59ZV

* Assert the fcitx5 binary guard and correct its stated retry rate

Removing ConditionPathExists=/usr/bin/fcitx5 left every existing fcitx5 assertion in systemd-test.sh passing, so the restart loop could come back unnoticed. The comment said the unit fails twice a second; RestartSec=2 makes it once every two seconds, which is what a worker without the binary showed (13 restarts in 30s).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Codex Medium <noreply@openai.com>

---------

Co-authored-by: Ruslan Kharitonov <ruslan.kharitonov@gmail.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com>
Co-authored-by: Codex Medium <noreply@openai.com>
* Span the notification menu across the card with even padding

The snooze menu and More list sat in the text column, indented under the
title and flush against the bottom edge. Draw the action area below the whole
content block, with the card's side padding underneath.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Refresh the notification clock when cards come back on screen

The clock only ticks while toasts are showing, so replayed history read every
card against the moment the last toast left and called them all "now".

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Clear only the shipped engine's notification leftovers at startup

The live/, icons/, reply-demo/, policy.json and icon-cache cleanup was for
files written by drafts of the Omapager PR that never shipped.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drop phone and meeting detection from notifications

Cards offered Copy number for phone-shaped text and relabelled meeting links
as Join. Links still get Open link; codes still get Copy code.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Drop the notification modes that route cards away from fullscreen

all-away and steam-away sent new cards to another display, or held them in
Recent, rather than stepping aside. A saved -away value now reads as the
default, all.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Show a notification's age only once it is five minutes old

A fresh card said "now", which told you nothing the card arriving had not.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Let a notification's link be clicked in its text instead of a button

The Open link button and its headline mark repeated a link the body already
draws, underlined and clickable.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Draw notification bodies once, as StyledText, so every link stays clickable

The body was drawn as RichText, with a plain-text copy for bodies over the
line cap because RichText cannot elide. Links in that copy could not be
clicked. StyledText covers Markup's tags, elides, and colours links itself.
Bare URLs are also linked beside formatting now, unless the sender wrote its
own anchors.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Copy a notification's code directly rather than through offers

With the link and phone buttons gone, a card's only offer is its code. The
offerTaken/takeOffer plumbing and the offer <kind> IPC call become copyCode.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Describe the notification demo's interactive scene without phone offers

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…6271)

Show/Hide Battery Percentage clipped in the menu, so it is now Battery %,
and the idle toggle reads Stay Awake / Allow Sleep instead of Stay Awake
On / Off.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…ooze on one line (omacom#16296)

* Recognise Brave Origin as a browser in notifications

* Put the toast snooze menu on one line and drop Do Not Disturb from it

* Remove notification options nobody needs

* Add a snooze button beside the close mark

* Remove notification IPC methods nothing calls

* Stop offering to copy phone numbers from notifications

* Wrap notification buttons instead of folding extras under More

* Drop the notification icon override folder

* Clear older notification layouts by migration, not at every startup

* Make the notification demo one run that shows everything

* Drop notification tests that restate the code

* Drop the Mark as read button no real sender here sends

* Drop an unused notification property and stale doc lines
…red title (omacom#16301)

* Start the Taildrop receiver after the session environment is imported

It started before UWSM imported OMARCHY_PATH, so omarchy-i18n failed and
every incoming file was announced with an empty title and body.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Shorten the Taildrop save location to ~ again

An unquoted ~ in the replacement re-expands to HOME, so notifications read
"Saved to /home/me/Downloads" instead of "Saved to ~/Downloads".

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Preview received Taildrop images in their notification

Taildrop only accepts files from your own devices, and the shell already
decodes notification images from every other app, so a received screenshot
shows itself again instead of a bare glyph. Other files keep the send glyph.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Keep critical notification titles in the normal text colour

Critical titles turned red, which reads as an error for things that merely
wait to be dismissed, like a received file. The title is already bold.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Keep the send glyph under Taildrop image previews

Omarchy's own cards get no letter fallback, so an image that fails to load
left the card with no mark at all. The shell already shows a loaded image
ahead of the glyph.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…om#16293)

* Leave the indicator peek shut while hovering right of the clock

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Hold the indicator peek's edge still while the peek is open

Beside a display notch, indicators right of the clock pull the center box's
end toward the start as they reveal, which closed the peek under a still
pointer and reopened it in a loop. Move the hover rule into BarModel so the
bar test exercises it at both sides of the edge.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
omacom#16306)

* Show and dismiss notifications instantly instead of sliding and fading

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Keep notification arrivals and departures instant all the way through

Late height changes on a new card, the next card's text taking the front,
and clearing several cards now jump too. Departure drops the fade-era
resting placement, so a leaving card is hidden and takes no input, and its
deferred removal no longer cuts short an unrelated animation.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…macom#16256)

* ✨ feat(region): seed offline chinese input

- ✨ make first-login pinyin available without downloads
- ⌨️ preserve compose and terminal prefix shortcuts

Amp-Thread-ID: https://ampcode.com/threads/T-01a0df38-0ea7-759b-bcff-29da223cc567
Co-authored-by: Amp <amp@ampcode.com>

* ✨ feat(pacman): preserve regional defaults

- ✨ share regional repository generation across install and refresh
- 🛡️ reject invalid profiles before replacing configuration

Amp-Thread-ID: T-01a0df38-0ea7-759b-bcff-29da223cc567

* ✅ test(region): guard runtime defaults

- ✅ protect global behavior and regional mirror ordering
- ⌨️ guard personal settings and reserved input shortcuts

Amp-Thread-ID: T-01a0df38-0ea7-759b-bcff-29da223cc567

* 📝 docs(region): explain china bootstrap choices

- 📝 clarify optional defaults, trust requirements, and opt-out behavior
- ⚠️ document snapshot compatibility and release validation limits

Amp-Thread-ID: T-01a0df38-0ea7-759b-bcff-29da223cc567

* 🐛 fix(channel): keep regional repos when switching to dev

- 🛡️ refuse a dev checkout that lacks this machine's region profile before linking it, since its own refresh would silently replace the regional pacman files
- ✅ isolate channel tests from the host region marker

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* 🐛 fix(pacman): keep existing config file modes

- 🐛 refresh used to copy over the existing files, so a customized mode must survive the shared regional applicator too

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* 📝 docs(region): drop the unreachable menu input gate

- 📝 the menu bypasses the input method, so requiring Pinyin there made the release procedure impossible until upstream fixes it
- 📝 record the required `skel` profile and the dev-switch region check

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* ♻️ refactor(pacman): render regional config without root

- 🔒 upstream refresh now gives root only command-scoped copies, so regional files are rendered as the caller and installed with the same four `sudo cp` calls
- ✅ exercise refresh only inside the sudo-boundary sandbox; the old host-level test reached real sudo once refresh pinned its PATH

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* 📝 docs(plans): move the china proposal to a discussion

- 📝 upstream plans are implementation plans; the open questions belong in the discussion that links these changes

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K3Jna1ogUUviYLJzhsG6Vr

* 🔥 refactor(region): drop input defaults from the cn profile

- 🔥 keep this change to the regional build architecture; Chinese input gets its own proposal instead of fixing one engine here
- ✅ assert the shipped profile adds only its repository keyring

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K3Jna1ogUUviYLJzhsG6Vr

* 📝 docs(region): scope the cn profile to repositories

- 📝 frame USTC and ArchLinuxCN as a bridge until an official China mirror exists
- 📝 record that language, timezone, and input defaults follow as separate changes

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01K3Jna1ogUUviYLJzhsG6Vr

* 🐛 fix(pacman): remove the render dir on every refresh exit

- 🐛 a failed render, a refused sudo prompt, or a signal left the scratch directory behind; repeated attempts piled them up
- 🔒 the cleanup runs before the sudo revocation, which stays the last step of the exit path

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdXqTwm63rRMwotZ1xqLC8

* 🐛 fix(channel): require a usable region profile for dev

- 🐛 an empty profile directory or a helper that cannot run passed the guard, linked dev, and then broke the refresh
- ✅ the success case now uses a complete profile

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdXqTwm63rRMwotZ1xqLC8

* ✅ test(pacman): isolate the regional refresh boundary test

- ✅ the render read the host's region marker, so a global machine never rendered China and a China machine failed on missing fragments
- ✅ check the finished China files root installs, not only the copy order

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdXqTwm63rRMwotZ1xqLC8

* ✅ test(pacman): sandbox the region for every boundary test

- ✅ The channel and update boundary tests copied the real render and dev-region guard, so on a China install they read the host's `/etc/omarchy/region` and failed before reaching their checks
- ♻️ The shared fixture now points both lookups at a sandbox target that starts global, and the regional test only writes `cn` there

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SdXqTwm63rRMwotZ1xqLC8

* ✨ feat(region): map timezones to regions

- ✨ one iso serves every region, so the installer derives it from the timezone the user already picks
- ✅ reject unknown zones and zones claimed by two regions

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QuATgjU4mnqpqSLAKMQ14g

* 📝 docs(region): describe timezone-derived regions

- 📝 the --region build flag is gone, so the installer contract and validation plan change with it

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QuATgjU4mnqpqSLAKMQ14g

* ✨ feat(region): put the china channel mirrors first

- ✨ omarchy now runs per-channel china replicas, so they should lead instead of trailing as a fallback
- ♻️ prepend per-channel fragments so each channel's own mirror stays the fallback
- 🔥 drop the ustc arch mirror, which was never a stable or rc snapshot replica

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QuATgjU4mnqpqSLAKMQ14g

* 📝 docs(region): explain mirror order and fallback

- 📝 pacman's three-error skip and first-server databases decide what the china replicas must serve

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QuATgjU4mnqpqSLAKMQ14g

* 🐛 fix(channel): require every region fragment for dev

- 🐛 leaving dev refreshes stable or rc with the checkout's tools, so a profile missing those fragments would drop regional repositories
- ✅ refuse a profile missing any single fragment

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QuATgjU4mnqpqSLAKMQ14g

* ✅ test(region): derive china zones from tzdata

- ✅ dropping a backward-compatible zone name from the profile now fails the test

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QuATgjU4mnqpqSLAKMQ14g

* ♻️ refactor(pacman): branch on render with if/else

- ♻️ follow the repo style instead of exiting early to skip the install block

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QuATgjU4mnqpqSLAKMQ14g

* 📝 docs(region): state marker mode and archlinuxcn risk

- 📝 refresh reads the marker unprivileged, and rolling archlinuxcn packages can outpace stable snapshots

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QuATgjU4mnqpqSLAKMQ14g

* 📝 docs(region): scope the refresh render boundary

- 📝 state that same-user code is outside the refresh boundary, so reviewers see the trade-off was deliberate

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QuATgjU4mnqpqSLAKMQ14g

* 🐛 fix(test): skip exact zone match without tzdata.zi

- 🐛 minimal containers lack the link table, so the exact china zone check failed on a correct profile

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QuATgjU4mnqpqSLAKMQ14g

* ✅ test(region): pin the platform in region tests

- ✅ tests no longer depend on the host's hardware, so they pass on ARM dev machines and the incomplete-profile case fails on the missing fragment, not the missing platform helper
- 📝 profiles add x86 repositories only, so ARM machines keep their own templates in a region

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* ✅ test(pacman): follow finalization into `omarchy-apply-pacman`

- ✅ finalization and refresh now restore templates through the apply command, so the template and deferred-hardware fixtures redirect its target etc instead of the post-install script's, and the refresh expects the unprivileged render
- ♻️ the apply command asks `omarchy-hw-platform` from PATH like its callers, so the fixtures' platform stubs reach it

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Expect the rendered refresh in quattro's newer channel tests

Two quattro changes postdate the region series. omacom#15903 keeps a refresh
on the machine's own channel and tests it by the template root copies
into /etc; with the renderer, root copies the rendered file, so those
checks now look for the template in the render step. omacom#16065 gave
aarch64 stable and rc, so the region test no longer expects aarch64 to
reject them: it applies every channel a platform has templates for and
expects the rest to be refused.

* Point China installs at Omarchy's China mirrors only, with no new repository

The China profile added ArchLinuxCN from USTC and its keyring. Nothing installed or updated needs that repository, and pacman trusts a populated keyring for every repository, so its packagers' keys would have been able to sign packages for [core] and [omarchy] on every China machine; it is also rolling while stable and rc are snapshots. It is gone, with the keyring package.

What China installs need is the same packages from closer servers. Besides the channel's Arch mirror on omarchy.cn, the [omarchy] repository now lists pkgs.omarchy.cn/<channel> first, the China replica of pkgs.omarchy.org, which carries the largest downloads. Both replicas serve the same signed packages, .org stays the fallback, and global installs are byte for byte unchanged.

A test pins Hong Kong, Macau and Taiwan, under every tzdata name, to no region. Existing installs stay global; the doc gives the two commands to opt in.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Find the rendered files from root's copy in the regional refresh test

The China render now writes pacman.conf with awk rather than copying the template, so the test reads the rendered directory from root's copy instead.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Check the China profile's mirrors, not its exact file list

The region test listed every file in default/regions/cn, so adding any
unrelated file to the profile failed it. What the test is for is that
the profile adds no repository and no key: every fragment is a Server
line for an omarchy.cn mirror. Keep that check and drop the list.

* Give the stale -dev build test the pacman renderer

The refresh now renders pacman's files with omarchy-apply-pacman. This
test, added to quattro after the region series began (omacom#15903), builds
its own sandbox and did not have the renderer, so its refresh failed
before reaching the switch it tests.

---------

Co-authored-by: Kros Dai <xdanger@gmail.com>
Co-authored-by: Amp <amp@ampcode.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com>
@coderabbitai

coderabbitai Bot commented Oct 22, 2026 •

Copy link
Copy Markdown

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration
  • Source mode: Auto
  • Repository: Unavailable
  • Branch: Unavailable
  • File: No configuration file from this repository applied
  • Commit: Unavailable
  • Layers: Organization UI
  • Review profile: CHILL
  • Plan: Advanced
  • Run ID: 7cd6c597-cc45-4c36-913e-a4c3215e43b7

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

  • 🔍 Trigger review
  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

@cursor
cursor Bot marked this pull request as ready for review October 22, 2026 23:11
@mark-groves
mark-groves merged commit 784d249 into quattro Oct 23, 2026
4 checks passed
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.

9 participants