Repository navigation
Sync upstream quattro through b8706a3f - #98
Merged
Merged
Conversation
* 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.
|
Important Review skippedBot user detected. To trigger a single review, invoke the
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Upstream
quattromoved from2fca7fe0(25 Oct, already on this fork via #95) tob8706a3f(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
/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. Migration1792965062runs that setup on existing installs whose drop-in names a/dev/path outside/dev/mapper.Notifications
notifications.dismissOnestays for custom bindings. Super+Shift+comma still dismisses all, and Super+Shift+Alt+comma still opens history.Panels
/usr/lib/omarchy/speedtest(omarchy-speedtest) instead of fast.com (Measure network speed with Omarchy's own service omacom/omarchy#16681). It counts only its own payloads, stops when the panel closes, and a missing client says which package to install. The panel's 20-second timer is a watchdog.Boot and login
owe render-statusonce nothing is covered and no theme transition is left (Stop the background intro's frame poll once nothing is covered omacom/omarchy#16731). On a Snapdragon laptop with a still wallpaper that poll had kept running for the whole session.i2c-hid-ofso the internal keyboard types at the disk unlock prompt (Let Snapdragon laptops' keyboards type at the disk unlock prompt omacom/omarchy#16688). The drop-in asksomarchy-hw-aarch64-qualcommand leaves the module out on x86, on other device-tree machines, and where the kernel built it in.Translations
Kept on the fork
Face, Grok Bot, and Cursor stay in the menu. Cursor stays the official tarball, not a mise tool.
FaceChromeCanvasstays immediately beforeFilterInput. The menupower-profilesprovider stays out, and the profile write staysbusctl. The fork delta againstb8706a3fis the same 223 paths as quattro against2fca7fe0.Tests
Passed:
hibernation-setup,mkinitcpio-hooks(including the Snapdragoni2c-hid-ofcases),network-speedtest,shell-shortcuts,notifications, andi18n(English keybindings match the Hyprland bindings).background-intro-lifecycleskipped: this machine has no Wayland compositor, so the still-intro poll stop was not exercised here.