Skip to content

Sync upstream quattro through b8706a3f - #98

Merged
mark-groves merged 11 commits into
quattrofrom
cursor/upstream-changes-pr-b870
Oct 27, 2026
Merged

mark-groves merged 11 commits into
quattrofrom
cursor/upstream-changes-pr-b870

Conversation

@cursor

@cursor cursor Bot commented Oct 26, 2026

Copy link
Copy Markdown

Upstream quattro moved from 2fca7fe0 (25 Oct, already on this fork via #95) to b8706a3f (26 Oct, Snapdragon disk-unlock keyboard omacom#16688). This replays all 11 first-parent commits. None of them touched the fork overlay, so the cherry-picks applied cleanly.

What changed

Hibernation

  • An unencrypted install names the resume device by filesystem UUID instead of a kernel name such as /dev/nvme1n1p4. On a machine with two NVMe disks those names can swap between boots, so resume was looking on the other disk and the session was lost (Name the hibernation resume device by its filesystem UUID omacom/omarchy#16658). Device-mapper names on encrypted installs stay as they are. A rerun rewrites a kernel-name drop-in Omarchy wrote, including one whose offset was empty, and rebuilds the boot image. A failed or interrupted rebuild leaves a marker and the next run finishes it. Migration 1792965062 runs that setup on existing installs whose drop-in names a /dev/ path outside /dev/mapper.

Notifications

Panels

Boot and login

Translations

Kept on the fork

Face, Grok Bot, and Cursor stay in the menu. Cursor stays the official tarball, not a mise tool. FaceChromeCanvas stays immediately before FilterInput. The menu power-profiles provider stays out, and the profile write stays busctl. The fork delta against b8706a3f is the same 223 paths as quattro against 2fca7fe0.

Tests

Passed: hibernation-setup, mkinitcpio-hooks (including the Snapdragon i2c-hid-of cases), network-speedtest, shell-shortcuts, notifications, and i18n (English keybindings match the Hyprland bindings). background-intro-lifecycle skipped: this machine has no Wayland compositor, so the still-intro poll stop was not exercised here.

Open in Web View Automation 

emirb and others added 11 commits October 26, 2026 23:04
* Name the hibernation resume device by its filesystem UUID

Hibernation setup wrote resume= from findmnt's SOURCE column. On an
unencrypted install that is the partition's kernel name, such as
/dev/nvme1n1p4. NVMe names follow the order the controllers finish
probing, so on a machine with two NVMe disks they can swap between
boots: resume= then names a partition on the other disk, the initramfs
finds no hibernation image there, and the machine cold-boots with the
session lost. On an MSI desktop with two NVMe disks the install's root
came up as nvme0n1p4 while resume= named nvme1n1p4, an ext4 partition
on the other disk.

Write resume=UUID=<filesystem UUID> instead. Both resume paths accept
it: mkinitcpio's resume hook hands off to systemd-hibernate-resume on
systemd 255 and later, which resolves resume= with
fstab_node_to_udev_node() to /dev/disk/by-uuid/, and the hook's own
resolve_device() handles UUID= on older systemd. The udev hook ships
60-persistent-storage.rules, so the link exists in the initramfs. The
kernel's early lookup rejects UUID= and leaves resume to the initramfs,
which is where it runs on these installs anyway, since nvme is a
module there.

A device-mapper name such as /dev/mapper/root on an encrypted install
is chosen by whoever opened the device and does not move, so it stays
as it is. If findmnt reports no UUID, setup falls back to the kernel
name it wrote before.

* Rename a kernel-name resume device when hibernation setup reruns

A machine set up before keeps resume=/dev/nvme0n1p2 in its drop-in,
and setup stopped at "Hibernation is already set up" without looking
at it. Now a rerun on a configured machine rewrites the line setup
itself writes, in the current and the Omarchy 3.4 form, with the
filesystem UUID and rebuilds the boot image. Device-mapper names, a
drop-in already naming a UUID and a drop-in edited by hand are left
alone, so a second run changes nothing.

limine-mkinitcpio 1.40 reports "mkinitcpio failed for kernel ...,
skipping." and still exits 0, so setup treats that line as a failed
rebuild, as migration 1791621765 does. On a failed rebuild the old line
goes back, so the drop-in matches the boot image and the next run tries
again. Both lines are written through install_root_file, the staged,
atomic replace the script already uses for its sleep hook, and are
staged before the drop-in changes, so a full disk cannot leave it
half-written. --no-rebuild leaves the rebuild to the caller, as the
installer and the factory reset expect.

* Rename kernel-name resume devices on existing installs

Every unencrypted install so far carries a kernel name in
/etc/limine-entry-tool.d/resume.conf, and nothing reruns hibernation
setup on its own, so the previous change would only reach new installs
and factory resets. The migration runs setup on machines whose drop-in
names a /dev/ path outside /dev/mapper and whose resume hook is in
place; setup renames the device and rebuilds the boot image.

Machines without that drop-in, encrypted installs, and drop-ins already
naming a UUID are skipped. If the boot image is not rebuilt, setup puts
the old drop-in back and exits 1, so the migration stays pending and
the next omarchy-migrate retries it.

* Rename the resume device after the empty-offset repair

A drop-in from a setup whose offset lookup failed reads
resume=/dev/nvme0n1p2 resume_offset= and does not match the line the
rename looks for, which needs a number there. The rename ran first and
skipped it, then the offset repair filled the offset in and rebuilt the
boot image with the kernel name still in it, and setup exited 0, so the
migration was marked done.

Run the rename after the offset repair, which leaves the device as it
found it. Such a drop-in now gets its offset, then the UUID, and the
boot image carries both.

* Move the resume device tests into the hibernation setup test

The repository keeps a command's tests in its area's existing file, and
hibernation-setup-test.sh is that file. Its locale case still skips on
a machine without hibernation; the resume device cases redirect every
path they write and run anywhere. Trimmed to the cases that carry
logic: the UUID and device-mapper names on first setup, a rerun
renaming a kernel name once (with and without an empty offset), the
restore when a kernel's image is not rebuilt, and drop-ins setup does
not own.

* Retry a rebuild of the renamed resume device until it succeeds

When the boot image was not rebuilt after the rename, setup put the old
line back. That retry only held if the restore itself succeeded and the
run lived long enough to do it. Closing the update terminal while the
image builds kills setup in between: the drop-in already names the
UUID, the boot image still names the kernel device, and the next
omarchy-migrate saw a UUID, exited 0 and marked the migration done. In
a VM with the 4.5.0rc1 install, a SIGHUP to the migration's process
group a second into the build left exactly that, with
/boot/limine.conf still carrying resume=/dev/nvme0n1p2 for good.

Setup now writes /var/lib/omarchy/hibernation-resume-rebuild before it
touches the drop-in and removes it only after a rebuild that
limine-mkinitcpio did not report as skipped. While the marker exists,
setup rebuilds, and the migration runs setup again. A failed, skipped
or interrupted rebuild therefore leaves a state the next run finishes,
and the restore goes away. --no-rebuild callers rebuild themselves and
get no marker.
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Of the twelve panels, these two are the only ones setting centerOnBar; the
other ten leave it false and open under the widget that was clicked. On a bar
with the clock at one end, the calendar opened in the middle of the screen
instead, which reads as a different control responding.

KeyboardPanel positions from anchorScreenPos and clamps the result into the
screen, so anchoring cannot push a popup off the edge -- the centred setting
was not protecting anything, only ignoring where the widget sits.

Co-authored-by: tayowrld <tayowrld@users.noreply.github.com>
Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
… its buttons (omacom#16699)

* Make Super+comma dismiss, invoke, or focus the newest notification by its buttons

A card with no buttons is dismissed, one with a single button has it
pressed, and one with several takes the keyboard on that card so you can
pick. notifications.dismissOne stays for custom bindings.

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

* Make Super+Alt+comma dismiss the newest notification

Super+comma now acts on a card by its buttons, so the plain dismiss moves
to Super+Alt+comma. Clicking a card from the keyboard stays in keyboard
mode (Super+Ctrl+comma, then Enter).

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

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…acom#16701)

* Count a notification's click as one of its actions on Super+comma

A web notification like WhatsApp's has no buttons, only a click that opens
it, so Super+comma dismissed it. Its click now counts: alone it is clicked,
and beside a button the card takes the keyboard to pick.

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

* Count the sender's own window as a notification's click on Super+comma

A native app like Signal may send no actions at all, but a click still
takes you to its window, so that counts as the card's action too.

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

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Measure network speed with Omarchy's own service

The speed test pulled fast.com's endpoints with a hard-coded token and
reported the interface's whole byte counter in one-second samples, so
any other download on the machine was added to the result: in an edge
VM it read up to 1074 Mbps down beside a 120 Mbit/s unrelated download,
and up to 1045 Mbps up on a 1 Gbit/s line. Closing the panel during the
upload also left four of its upload loops running; they were still
starting new uploads 30 seconds later.

Run Omarchy's own client instead. It is built and packaged on its own
and installed at /usr/lib/omarchy/speedtest; the launcher replaces
itself with it and passes the panel's PID, so the client stops when
the panel goes away. The client counts only its own test payloads over
eight persistent connections, keeps connection warm-up out of the
measurement and prints the running rate as before, so the panel's
protocol does not change. In the same VM it read 919 Mbps down alone,
808 beside the unrelated download, and left no process or connection
two seconds after the panel closed mid-test. Since the client owns
warm-up and the five-second window, the panel's timer becomes a
20-second watchdog that reports a failure instead of stopping silently.

Accept exactly one direction: `down extra` used to run a download.

* Name the package when the speed-test client is missing

Without /usr/lib/omarchy/speedtest the launcher failed with exec's "No
such file or directory" and status 126, which the panel showed as is.
Say which package to install instead, and exit 1 like a failed test.
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
…m#16731)

* Stop the background intro's frame poll once nothing is covered

When OWE takes the background for a login intro, BackgroundIntro starts
a 16 ms timer that runs `owe render-status` until a video frame shows.
If the background is handed back before such a frame is reported,
finishStartup() (or the 10 s startup fallback) drops the cover, and
nothing stops the timer: frameStatus discards every answer once there
is no cover or theme token, so the shell keeps starting a process per
frame for the rest of the session. It was running at 62.5 `owe` execs
a second on a Snapdragon X1 laptop logged in on a still wallpaper.

Measured on that laptop, idle desktop at 25% brightness, 30 s counters
and 5 minutes of battery power, with and without a loop running
`owe render-status` every 16 ms (97 a second; the shell managed 62.5):

                         polling       not polling
  CPU clusters idle      0% / 27%      87% / 87%
  scheduler ticks        2,385/s       415/s
  call_rcu               1,820/s       35/s
  battery power          3.27 W        2.89 W

Stop the timer from its own tick once there is no cover and no theme
transition left, so every path that clears them ends it.

* Test the still intro handoff inside the lifecycle test

The still-wallpaper case shared all of the lifecycle test's setup: the
staged service, the launcher stub and the owe stub. Run it as a second
fixture of that test instead of a test file of its own; the owe stub
reports a still and logs each poll when INTRO_TEST_POLLS is set.
…com#16688)

* Let Snapdragon laptops' keyboards type at the disk unlock prompt

The internal keyboard of a Snapdragon X laptop is HID over I2C, described
by the device tree, so it binds i2c-hid-of. mkinitcpio's keyboard hook
adds drivers/hid/hid-* and the USB and serio drivers, not the I2C HID
transport, and nothing else names it, so an encrypted install showed the
LUKS prompt and took no keys. Only the Yoga Slim 7x's board setup names
the module for its own keyboard.

Ship a drop-in that early-loads i2c-hid-of on machines with a device tree
whose kernel builds it as a module, the way thunderbolt_module.conf
handles thunderbolt: named without ?, since a systemd initramfs copies
MODULES into modules-load.d, and left out on x86 and where it is built
in. omarchy-settings ships etc/ as it is, so installed machines get it
with the package and their next kernel or firmware update rebuilds the
image.

On an ASUS Zenbook 14 UX3480QA (Snapdragon X X1-26-100, keyboard
0B05:4B42 on i2c22) installed encrypted with linux-omarchy 7.3, the LUKS
prompt took no keys; with i2c-hid-of in the image (i2c-hid-of.ko and its
dependency i2c-hid.ko) the passphrase types and the disk unlocks.

* Ask the Snapdragon detector before early-loading i2c-hid-of

The drop-in read the device tree itself to decide. Ask
omarchy-hw-aarch64-qualcomm instead, which also answers for an image
built away from its machine, and leave the module out when the omarchy
runtime that carries the detector isn't installed yet, as the hooks
baseline does. Device-tree machines that aren't Snapdragon (Apple
Silicon, a plain aarch64 board) no longer get it; their keyboards don't
use it.
@coderabbitai

coderabbitai Bot commented Oct 26, 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: a956859c-e590-497e-87a4-ed0be192415a

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.

@mark-groves
mark-groves merged commit 323e230 into quattro Oct 27, 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.

5 participants