Skip to content

Move Macs onto Omarchy's official packages with a standalone omarchy-mac-migrate - #690

Open
maralcbr wants to merge 8 commits into
quattrofrom
mac/migrate-standalone
Open

maralcbr wants to merge 8 commits into
quattrofrom
mac/migrate-standalone

Conversation

@maralcbr

@maralcbr maralcbr commented Oct 4, 2026

Copy link
Copy Markdown
Member

Moving existing Macs onto Omarchy's official packages is now a standalone script, not part of any package (agreed with Ryan on 2026-10-04: the Mac packages move to omacom/omarchy-mac-pkgs without the migration engine; omacom#13362 uses them).

What it adds

Supersedes #603. The code of #601, #602 and #607 is the tool's source.

Tests

  • New fixture suites mac-migrate-test.sh (tester cohort, engine), mac-migrate-legacy-test.sh, mac-migrate-mx-test.sh and mac-move-migration-test.sh: 104 checks. They cover kill -9 after, during and in the middle of every step, deferral before the switch and failure after it, the guard, channel detection and qualification, and leftovers.
  • VM runs on the M1 (aarch64 KVM with the Mac image harness, on the encrypted test image apple-test-f22c43fb7903). The target was a test-lane-signed set: the Converge omarchy-mac and omarchy-mx-mac into upstream Omarchy omarchy#13362 (69d80cc) dev pair, plus omarchy-mac and omarchy-mac-boot from proposal/main without the engine. Everything else came from the real edge.
    • Stable: deferred (75) with the Mac byte-identical.
    • Today's edge: deferred. The dev pair has no dispatcher, and the boot package carries the old engine.
    • Test image → edge: passed.
    • mx-mac → edge: the guest ran the fork's omarchy-dev 4.0.4.r7099 bundle. The rc lane deferred unchanged. On the edge lane, the fork's pair went down to the official build in one transaction, with kills during prefetch and the boot-chain rebuild resumed.
    • Encrypted legacy → edge: the guest was set up as quattro's layout (GRUB, busybox encrypt from cryptdevice=, ESP at /boot, [omarchy-aarch64] lane). The ESP moved to /boot/efi, the unlock to sd-encrypt and the loader to Limine.
    • After each move, pacman.conf was the core configuration and the migrated disk unlocked with the same password through its own boot files and reached a login. The boot unit's verification correctly refused the guest's generic kernel; completion needs a Mac.
  • Two design debate rounds and two code review rounds; every verified finding is fixed (boundary event, lock contention, retired keys in preflight, NoExtract/NoUpgrade, guard scope, update-hook exit, payload tool trust, resume unit kept through user-setup retries, HOOKS preview). The VM runs used an earlier build (879faa65); the fixes after it are covered by the fixture suites.
  • Not done: real Mac hardware (the post-reboot verification needs the Aurora kernel). A power loss inside the pacman transaction itself is not recoverable by the tool, as for any pacman upgrade.

Release (owner): merge, publish bin/omarchy-mac-migrate as release mac-migrate-v1, and publish a quattro omarchy build to the [omarchy-aarch64] lanes so packaged quattro installs get it (checkout installs get it from git pull). Nothing moves until omarchy-mac-pkgs' packages and a dev pair with omacom#13362 are on edge.

…c packages

The migration engine leaves omarchy-mac-boot (omacom/omarchy-mac-pkgs drops it):
one self-contained script built from migrate/src, journaled and resumable, that
moves a quattro, mx-mac or test-image Mac onto its channel's official packages
(the omarchy-dev pair on edge) and the core Apple Silicon pacman configuration,
and defers while that channel has no Mac release.
A migration marks the Mac; the next omarchy update runs omarchy-mac-migrate
before any fork step and stops there for the reboot, or updates the Mac as
before while its channel has no Mac release. The README gives testers the
one-line command.
The boundary is recorded right before the switch's first write, so a reset or
a refusal before it still defers. A second run never touches the state of the
run holding the lock. Preflight trusts no retired key, refuses NoExtract or
NoUpgrade patterns and IgnorePkg globs that hold back what the move installs,
and checks a busybox-encrypted Mac keeps its unlock under the new drop-ins.
Fork leftovers the Mac packages no longer retire are retired here. The update
hook stops the fork update only after a move.
…tion

Preflight runs only regular files from the verified omarchy-mac-boot, never a
link, and refuses a boot package without them. Its HOOKS preview drops the
drop-ins of the packages the transaction replaces. A user setup that succeeds
on a retry no longer releases the unit while the migration is still under way,
and a migration past its switch resumes whatever a detector says.
pacman -Qlq of a package that is not installed fails, which under pipefail
made the preview fail and an encrypted legacy Mac defer for good. The fixture
pacman now fails the same way. The boot unit also resumes a migration past its
switch whatever a detector says.
It runs as root in an empty environment and across the move to another
runtime, so it cannot depend on the runtime's helpers.
@Knutsi

Knutsi commented Oct 21, 2026

Copy link
Copy Markdown

Preflight-only report from a second legacy Mac: MacBook Pro 16" M1 Max (j316c), quattro guided install, encrypted, GRUB, ESP at /boot, edge. sudo bash omarchy-mac-migrate check at 2976e70d exited 0 (migration not run):

Ready: run moves this Mac (legacy, grub boot, encrypted) onto repository https://pkgs.omarchy.org/edge/$arch (edge).

This Mac also has GRUB_FONT=/boot/grub/fonts/unicode32.pf2 from omarchy-mac-setup, so #727's fix applies beyond the M1 Pro.

low_battery() counts peripheral batteries (migrate/src/engine.sh, lines 486–502):

  • Observed here: a Logitech receiver registers hidpp_battery_3 (scope=Device) with no capacity file, and check printed omarchy-mac-migrate: line 550: /sys/class/power_supply/hidpp_battery_3/capacity: No such file or directory before the verdict. The || capacity=100 fallback still works, but the trailing 2>/dev/null doesn't silence that message, because on an assignment-only command it doesn't cover errors raised inside the command substitution (bash 5.3; wrapping it in { …; } 2>/dev/null does).
  • Reproduced with the extracted function on a fake sysfs (not observed on this Mac): with the Mac unplugged at 90% and a scope=Device mouse at a readable 20%, low_battery returns true. Preflight then refuses with "the battery is below 30% and no charger is connected" and exits 75, so the move is deferred while the fork update carries on. The same happens for a battery with present=0.

Skipping those the way bin/omarchy-battery-present in core does fixes both. A Mac really below 30% and unplugged still refuses, scoped System or unscoped:

      Battery)
        [[ -r $supply/scope && $(<"$supply/scope") == "Device" ]] && continue
        [[ -r $supply/present && $(<"$supply/present") == 0 ]] && continue
        { capacity=$(<"$supply/capacity"); } 2>/dev/null || capacity=100

@Knutsi

Knutsi commented Oct 26, 2026

Copy link
Copy Markdown

Friendly bump: is anything blocking a review here? Preflight still passes on my M1 Max (j316c, legacy/grub/encrypted, edge). If it lands this week, I'll do a full run on Friday evening and report back. #726 and #727 look like they belong with it too. @maralcbr

@harsha-satyuma

Copy link
Copy Markdown

I used migrated my M1 pro to latest 4.5-RC2. Used claude to fix issue with grub migratio. Below are all the details, generated using claude

Real-hardware report: legacy encrypted GRUB Mac → rc, completed

Machine: MacBook Pro 16" M1 Pro (2021), 16 GB.
Before: omarchy-mac quattro checkout install, 4.0.3rc4-1, [omarchy-aarch64] edge lane, GRUB with the ESP at /boot, busybox encrypt, linux-asahi 7.1.13.
Tool: bin/omarchy-mac-migrate from mac/migrate-standalone @ 2976e70d8 (sha256 07c01b67…c25c, matches the README one-liner).

Result

The migration completed on real hardware. After the reboot, omarchy-mac-migrate-verify.service finished reboot and retire. The Mac now runs omarchy 4.5.0rc2-1 from pkgs.omarchy.org/rc, linux-aurora 7.2.11 and Limine, with the ESP at /boot/efi, unlocking through sd-encrypt (rd.luks.name=). LUKS, snapper and data are intact, and all pending upstream user migrations ran on the next omarchy update.

What I ran:

  1. check: Ready, cohort legacy, grub boot, encrypted, target edge (from the lane).
  2. check --target with an admin target format=1 / type=repository / channel=rc: Ready, target rc.
  3. run --target …: preflight through boot-chain passed. loader failed and rolled back cleanly (see 1).
  4. After the workaround, run resumed at loader and completed. The reboot and verify both passed.

The rollback worked well. After the failed loader, GRUB, the Aurora kernel and the busybox-encrypt initramfs on the ESP were consistent, and the Mac would have booted.

Issues found

1. GRUB_FONT in /etc/default/grub breaks Limine activation on legacy GRUB Macs (blocker).
The quattro setup wrote GRUB_FONT=/boot/grub/fonts/unicode32.pf2. After legacy_stage moves the ESP off /boot, that path no longer exists. setup-boot then runs update-grub (/usr/lib/omarchy-mac/boot/setup/limine-boot.sh:100), and it fails:

/usr/bin/grub-probe: error: failed to get canonical path of `/boot/grub/fonts/unicode32.pf2'.
The boot switch was undone; GRUB boots this Mac as before.
omarchy-mac-migrate: Limine could not be activated; GRUB is still the loader

Workaround: comment out GRUB_FONT, then run again, which resumed at loader and succeeded. My root cause comes from reading the error and the code, but the fix confirms it: the same step passed with only that line changed. Suggested fix: have the legacy adapter drop or comment out a GRUB_FONT under /boot/grub/ in its stage, or have limine-boot.sh not fail the activation when update-grub fails for a GRUB that no longer boots anything. Any quattro Mac set up by the guided installer will probably hit this.

2. Retiring wifi.backend=iwd loses saved Wi-Fi passwords.
retire moves the fork's NetworkManager/conf.d/wifi_backend.conf (wifi.backend=iwd) to leftovers/. NetworkManager then uses wpa_supplicant, but the PSKs had been stored by iwd. Every saved network fails with reason 'no-secrets' on first boot until the password is entered again. The result is no network on first login, and with no network the new shell setup sits through mise @latest lookups (about 30 tools × 20 s timeouts). Suggested fix: copy the PSKs from /var/lib/iwd/*.psk into the matching NetworkManager profiles before retiring, or at least tell the user to expect it.

3. A --target path that doesn't exist is silently ignored.
find_target skips paths that don't exist, so a typo or a failed copy falls back to channel detection and quietly plans edge. Mine failed to copy and check just reported edge. Suggested fix: refuse (exit 75) when --target is given and isn't a readable, root-owned file.

4. prefetch runs about 10 minutes with no output.
legacy_conflicts loops over 25,806 paths and runs a separate awk over the 2 MB conflicts.owned for each one. That's quadratic, and it looked hung. Loading the owner map once (an associative array, or a single awk/join) would make it seconds. A progress line would also help.

5. Harmless but alarming: warning: database file for 'omarchy-aarch64' does not exist.
It's printed during prefetch, from a pacman call that pairs the live pacman.conf (still listing the fork lane) with the tool's database copy. Using the frozen/target config for that call would make it go away.

Note on rc for Macs

The 4.5.0rc2 runtime has Apple Silicon templates for edge only (default/pacman/aarch64-apple/{pacman,mirrorlist}-edge, and omarchy_pacman_default_channel → edge). A Mac migrated to rc therefore:

  • updates fine from pkgs.omarchy.org/rc;
  • reports omarchy-version-channel as unknown / rc, because the mirror list still names mirror.archlinuxarm.org;
  • gets "no rc/stable channel for aarch64-apple" from omarchy-channel-set until those templates ship.

If rc isn't meant to be a Mac target yet, the rc [omarchy] repo carrying omarchy-mac/omarchy-mac-boot makes the tool's "channel carries the Mac" check pass anyway. The tool should probably also require the runtime's aarch64-apple templates for that channel.

Happy to rerun anything or collect logs from this machine.

@patatman

Copy link
Copy Markdown

Full run on real hardware: MacBook Pro 14" M2 Pro (j414s), quattro guided install, LUKS + btrfs + snapper, GRUB, ESP at /boot, GRUB_FONT on the ESP, edge, with Touch ID through aurora-sep. Built from #727's branch (so #726 and #727 included).

Result: migrated. After the reboot verify passed ("This Mac now runs repository https://pkgs.omarchy.org/edge/$arch"), retire done, booted linux-aurora 7.2.11-10-1-ARCH through Limine, unlock through rd.luks.name=/sd-encrypt, platform aarch64-apple. All 81 pending Omarchy migrations then ran clean, no failed units from the move. Kept without an official build: obsidian-appimage. #727's GRUB_FONT fix was needed here too.

Two problems, both from aurora-sep:

  1. Preflight refuses on aurora-sep's pin, and the migrator has no aurora-sep handling: omarchy-mac-migrate refuses every Mac with Aurora Touch ID (aurora-sep pin), and leaves aurora-sep behind #778.
  2. loader failed with the staged boot chain does not check: Apple Silicon boot check: /boot/vmlinuz-linux-aurora is not the 7.2.11-10-1-ARCH kernel linux-aurora installed. kernel_release() takes the first module tree whose pkgbase names the kernel; aurora-sep's tree (7.1.12-2-20261013.18-sep-ARCH) is owned by no package, stays after the transaction, says linux-aurora and sorts first, so legacy_build_initramfs copied the old kernel to the root's /boot and the UKI was built from it. The boot check caught it before the reboot. Moving that tree's pkgbase aside and rerunning run resumed at loader and finished. Fix with a legacy fixture test: Take the Aurora release from the tree the package owns #777 (stacked on Drop the fork's GRUB_FONT once the ESP leaves /boot #727).

Notes for the tests: mac-migrate-legacy-test.sh has the Plymouth environment dependency #726 mentions; with it hidden (bwrap --dev-bind / / --tmpfs /usr/lib/initcpio/install) legacy 26/26, mac-migrate-test 49/49, mac-migrate-mx-test 28/28, mac-move-migration-test 4/4, migrate/build --check clean.

@AmmonMatthews

Copy link
Copy Markdown

Report: M2 Air 15" (j415), fork 4.0.3 → rc, completed; loader failed once because the fork's omarchy_hooks.conf survived the transaction as the live file (the official one landed as .pacnew) and legacy_switch_unlock keeps it when a package owns the path

Setup: MacBook Air 15" M2 2023 (apple,j415), omarchy-mac 4.0.3 (quattro guided install, 2026-09-09), GRUB, ESP at /boot, LUKS + btrfs (@, @home, @log), no snapper. Tool at 2976e70d8, /etc/omarchy-mac/migration-target → channel=rc. I had removed a self-built linux-asahi variant first (the preflight otherwise refused: device tree …-fairydust-ARCH/dtbs/… is not one of linux-asahi 7.1.13-3-2-ARCH's, correct behaviour). GRUB_FONT commented out per #727.

run went preflight → … → boot-chain cleanly, then:

loader fail the legacy adapter could not stage the boot switch; GRUB is still the loader

Cause. The fork's /etc/mkinitcpio.conf.d/omarchy_hooks.conf (from the fork's omarchy-settings 4.0.3-1) begins with

HOOKS=(base udev plymouth keyboard autodetect microcode modconf kms keymap consolefont block encrypt filesystems fsck btrfs-overlayfs)

The official omarchy-settings 4.5.0rc2-1 (aarch64, rc) lists the same path in backup=, so the upgrade kept the fork's bytes and put the new file beside them:

[ALPM] warning: /etc/mkinitcpio.conf.d/omarchy_hooks.conf installed as /etc/mkinitcpio.conf.d/omarchy_hooks.conf.pacnew
[ALPM] warning: /etc/mkinitcpio.conf.d/thunderbolt_module.conf installed as /etc/mkinitcpio.conf.d/thunderbolt_module.conf.pacnew
[ALPM] upgraded omarchy-settings (4.0.3-1 -> 4.5.0rc2-1)

legacy_switch_unlock knows this file ("The fork's omarchy_hooks.conf sets the busybox line outright and sorts after the Apple drop-ins. Where no package took it over, it goes too."), but it decides by ownership: after the transaction pacman -Qlq lists the path under the new omarchy-settings, so the file stayed, with the fork's content. mkinitcpio reads it after 00-omarchy-hooks.conf and the 9x-omarchy-mac-*.conf drop-ins, its HOOKS=(… encrypt …) replaced the sd-encrypt line they had composed, and legacy_check_unlock refused the switch. Preflight did not see it coming: future_hooks overlays the new packages' drop-ins on a copy of the live directory, which assumes pacman installs them, and for a backup= file whose contents differ it does not. The legacy test covers the .pacnew case for 10-theme.conf, but its new settings package does not bring omarchy_hooks.conf, so there the fork's file goes with the fork's package and the encrypted scenario passes; the rc aarch64 omarchy-settings does bring it.

Fix on my side:

sudo mv /etc/mkinitcpio.conf.d/omarchy_hooks.conf ~/omarchy_hooks.conf.fork
sudo mv /etc/mkinitcpio.conf.d/omarchy_hooks.conf.pacnew /etc/mkinitcpio.conf.d/omarchy_hooks.conf
sudo mv /etc/mkinitcpio.conf.d/thunderbolt_module.conf.pacnew /etc/mkinitcpio.conf.d/thunderbolt_module.conf

then run again: resumed at loader, finished through reboot. First boot: Limine, sd-encrypt passphrase prompt, omarchy-mac-migrate-verify.service ran reboot + retire and removed itself. On Aurora 7.2.11 the external display that needed a patched kernel before works on the stock kernel on both USB-C ports (DP-2 on one, DP-1 on the other); speakers, mic, webcam, Wi-Fi, Bluetooth and zram work. omarchy update afterwards ran 109 migrations and passed its boot verification. One more reboot since, clean.

Suggestions:

  • In legacy_switch_unlock, decide by content rather than ownership: an omarchy_hooks.conf whose HOOKS=( line names encrypt outright is the fork's, whoever owns the path now. Keep it in the boot-switch backup as today and, when omarchy_hooks.conf.pacnew sits beside it, move the .pacnew into place instead of leaving the path empty. thunderbolt_module.conf.pacnew is harmless but stays behind the same way.
  • Have future_hooks treat a live drop-in the new package lists in backup= the way pacman will (kept, .pacnew beside it), or at least let the loader failure name the drop-in that put encrypt back.
  • Two more fork-era leftovers the retire list could cover: an unowned /etc/mkinitcpio.conf.d/apple_hid_modules.conf (superseded by 92-omarchy-mac-hid.conf) and the grub package plus /boot/grub, now unreachable.

Also noticed after omarchy update: migration 1791581004 warns "The Limine kernel command line leaves out systemd.tty.term.console=dumb; use += rather than = in /etc/default/limine", because omarchy-mac-limine-cmdline writes KERNEL_CMDLINE[default]= from GRUB's defaults, which overrides upstream's omarchy-defaults.conf += additions. Cosmetic (boot verbosity), but worth knowing when upstream migrations assume +=.

Unrelated to the tool, but it shows in its output: this Mac's SMC RTC came up as 1970-01-12 at every boot (macsmc-rtc: setting system clock to 1970-01-12…, timesyncd fixed it a minute later), so the complete file records completed=1970-01-12T01:55:03 and the step journal's post-reboot lines carry epoch 978897. One hwclock --systohc after timesyncd had synced fixed it for good: the next boot logged macsmc-rtc: setting system clock to 2026-10-30T21:31:48 UTC. The tool could do that once the clock is synced, or check date before writing timestamps.

@simon-nicholls

Copy link
Copy Markdown

Report: MacBook Pro 16" M1 Max (j316c), omarchy-mx-mac fork 4.0.4-mac.1 → rc, completed

Tool at 2976e70d, run with --target set to channel=rc. Setup: Limine, not encrypted, btrfs root, separate ext4 /boot, ESP at /boot/efi.

check reported mx-mac, limine boot. run went through without intervention and exited 0. After the reboot, verify, reboot and retire completed. It now runs omarchy 4.5.0rc2-1, linux-aurora 7.2.11 and m1n1-aurora 1.6.1.

Working afterwards: Wi-Fi, speakers, USB audio, Bluetooth, and suspend and resume (by command and by lid). External 4K works over HDMI and over USB-C (120 Hz). The charge limit held.

Things I noticed:

  1. quickshell-git downgrades. The plan moved 0.3.2.r7.g2b4501a-1 to 0.3.2.r0.g4f508be-5, though the same repo has quickshell 0.3.2.r7.g2b4501a-1. In mx_mac_plan, a package name the official repo carries is taken before its mx_mac_counterpart. The bar worked anyway.
  2. .pacnew on a boot hook. A locally differing /etc/initcpio/install/omarchy-vendorfw stayed in place, with the new version left as .pacnew (same for the kmod-static-nodes drop-in). It booted fine unencrypted, and I merged and rebuilt afterwards. It might matter on a Mac that adds encryption later.
  3. Wi-Fi stayed on iwd. wifi_backend.conf was retired, but 20-omarchy-mac-wifi.conf from omarchy-mac 0.2.3 keeps the backend on iwd, and saved networks carried over.
  4. no-secrets can be misleading. Before migrating, I tried wpa_supplicant on this Mac. The password was stored, but association was rejected with status_code=16, and NetworkManager reported that as no-secrets. It may be worth checking for that in the lost-password report above.
  5. fnmode changes. hid_apple.conf with fnmode=2 was retired, so the top row went to the kernel default. Fine for me, but it's a visible change.
  6. Symlinked config. Pending migrations replaced a symlinked ~/.config/omarchy/shell.json with a plain file, and edited a symlinked mise/config.toml, which matters to anyone keeping these in dotfiles.
  7. rc channel display. omarchy-version-channel shows unknown / rc, as reported above.

Thanks ever so much for the tool; loving the Omarchy Mac work!

I ran this with help from Claude (Claude Code), which also reviewed the PR's code beforehand and helped write this report.

@Knutsi

Knutsi commented Nov 2, 2026

Copy link
Copy Markdown

Another data point for the omarchy_hooks.conf failure in @AmmonMatthews' report. My M1 Max (j316c, quattro guided install, legacy/GRUB/encrypted, edge, still on 4.0.3rc4-1, not migrated yet) is set up the same way:

$ head -1 /etc/mkinitcpio.conf.d/omarchy_hooks.conf
HOOKS=(base udev plymouth keyboard autodetect microcode modconf kms keymap consolefont block encrypt filesystems fsck btrfs-overlayfs)
$ pacman -Qo /etc/mkinitcpio.conf.d/omarchy_hooks.conf
/etc/mkinitcpio.conf.d/omarchy_hooks.conf is owned by omarchy-settings 4.0.3rc4-1

It also has GRUB_FONT=/boot/grub/fonts/unicode32.pf2 and the unowned apple_hid_modules.conf. So every encrypted quattro Mac set up by the guided installer probably meets both problems, on edge as well as rc. check at 2976e70d still passes here, because the preflight doesn't see either problem.

Would it make sense to fold #727 and a content-based omarchy_hooks.conf check (with the .pacnew moved into place) into this PR before merging? Otherwise the first run stops at loader on these Macs. I'm holding my full run until then and will report back after a Friday-evening run.

@Knutsi

Knutsi commented Nov 2, 2026

Copy link
Copy Markdown

Follow-up to my comment above, from reading the code and the edge/rc packages on the same M1 Max (legacy/GRUB/encrypted, edge). I haven't run run.

  1. Correction: on edge the fork's omarchy_hooks.conf isn't a trap when it's unmodified. omarchy-settings-dev 4.0.0.r7555 conflicts with omarchy-settings, so pacman removes the fork package, and its unmodified backup file goes with it. The new file has no busybox line. The legacy test's encrypted-channel case covers this, and here pacman -Qii reports the file as [unmodified]. On rc, a same-name upgrade would also replace an unmodified file, so the .pacnew in @AmmonMatthews' report suggests his copy had been changed. The content-based check is still worth adding.
  2. Free space: preflight's minimum is fixed (engine.sh 724-728: 4 GiB + /etc + /boot, plus 512 MiB with the ESP at /boot), but the transaction is a full -Su plus the targets. Here that is 190 packages today: 1.9 GiB downloaded to cache/pkg, kept until retire, and about 6.3 GiB written. Snapper snapshots keep replaced files allocated. Preflight doesn't look at btrfs unallocated space, and the defaults step's pacman -S isn't counted. Summing the resolved download and installed sizes, and warning when snapshots exist or nothing is unallocated, would cover it.
  3. No kernel on the ESP during the transaction: removing linux-asahi runs mkinitcpio's PreTransaction remove hook. Until the PostTransaction install hook, GRUB has no kernel to boot, and after that grub.cfg still names linux-asahi until boot-chain. Power loss in that window needs rescue media. Keeping the old kernel and image aside until boot-chain would avoid it, or the README could say so.
  4. loader's checks after setup-boot (engine.sh 1273-1276) fail without unstaging, so Limine already owns U-Boot's slot (the Take the Aurora release from the tree the package owns #777 case). The message should say that Limine is now the loader, and that run should be repeated before rebooting.
  5. About removing the grub package (suggested above): /etc/default/grub belongs to grub and is still Limine's command-line source (limine-boot.sh, omarchy-mac-limine-cmdline). Removing grub would turn it into .pacsave.
  6. The silent conflict scan also runs in transaction's prepare (cohort-legacy.sh 263), after the boundary, so the same silence comes a second time.
  7. rc templates: omarchy-settings 4.5.0rc3 now ships aarch64-apple/pacman-rc.conf and mirrorlist-rc. The tool's core_pacman_conf still writes [aur] and leaves the mirror list on mirror.archlinuxarm.org, which is probably why omarchy-version-channel shows unknown.
  8. Accept aarch64-apple as the Mac platform in omarchy-mac-migrate #726, Drop the fork's GRUB_FONT once the ESP leaves /boot #727 and Take the Aurora release from the tree the package owns #777 keep tool_version=1. hand_over takes over at >=, so a resumed migration can be handed back to a build without the fixes. Raising tool_version in each fix avoids that.
  9. README: the one-line install fetches release mac-migrate-v1, which doesn't exist yet.

@rbadhesha

Copy link
Copy Markdown

I use Omarchy MX Mac and want an in-place migration to official Omarchy Mac. I did not attempt migration or execute the migration tool.

Read-only inspection confirmed this configuration:

  • Apple MacBook Pro, 14-inch, M2 Max, 2023 (apple,j414c).
  • Arch Linux ARM, MX version file 4.0.4-mac.1.
  • omarchy-dev and omarchy-settings-dev: 4.0.4.r7099.gb8ed395-1, with the MX bundle updater present.
  • linux-aurora: 7.1.12.aurora2-7. Current kernel: 7.1.12-2-7-ARCH.
  • m1n1-aurora: 1.6.1.aurora1-2. uboot-asahi: 2026.07.asahi2-3.
  • Firmware LoaderInfo: Limine 12.9.6. The MX Limine enable marker is present.
  • LUKS2-encrypted Btrfs root, with systemd LUKS parameters in the boot command line.
  • Separate ext4 /boot (2 GiB) and FAT /boot/efi (500 MiB).
  • The disk also contains APFS partitions. I use macOS dual boot, but this inspection did not verify macOS startup.

I read the convergence specification, MX adapter work in #571, and standalone migration proposal #690. I understand that implementation and test reports do not establish released support for this configuration.

Can the maintainers clarify these points?

  1. Does the current omarchy-mac-migrate proposal support this M2 Max, Aurora, Limine, LUKS2/Btrfs configuration?
  2. What blockers or prerequisites remain before the maintainers recommend migration for this configuration?
  3. Is there an official, recommended migration procedure for existing MX Mac 4.0.4-mac.1 installations?
  4. Which release, milestone, or issue will identify migration readiness?
  5. Can a sanitized system inventory or read-only diagnostic output help establish compatibility?

This is a compatibility and readiness question. I do not report a migration failure. Thank you for your guidance.

@rbadhesha

Copy link
Copy Markdown

Tested the development omarchy-mac-migrate preflight on my MacBook Pro 14-inch (M2 Max, 2023), running Omarchy MX Mac 4.0.4-mac.1, Arch Linux ARM, Aurora, and Limine.

Migration tool commit: 2976e70d81700161ccf9ae9bef03bdab5d42e2cb

Command: omarchy-mac-migrate check (using the downloaded script with root privileges)

The tool automatically selected the RC repository (https://pkgs.omarchy.org/rc/$arch). Repository synchronization succeeded, but dependency resolution failed:

:: unable to satisfy dependency 'libaquamarine.so=15-64' required by hyprtoolkit
error: failed to prepare transaction (could not satisfy dependencies)

Exit code: 75 (deferred). No migration was performed.

Separate package metadata inspection found:

  • RC [omarchy]: aquamarine 0.15.1-1.3 provides libaquamarine.so=14-64.
  • Arch Linux ARM [extra]: hyprtoolkit 0.6.0-3 requires libaquamarine.so=15-64.

These versions suggest an ABI mismatch between the selected repository packages. The preflight's temporary repository databases were not retained, so the exact package selection could not be independently reconstructed.

The preflight stopped before boot-helper execution, so compatibility with my encrypted Btrfs/LUKS2 and Limine configuration remains unverified.

Are you aware of this dependency mismatch, and is a compatible package set planned for the migration preflight?

Thanks!

@simon-nicholls

Copy link
Copy Markdown

@rbadhesha On the dependency mismatch, from my M1 Max, which I migrated with 2976e70d on 1 Nov (report above):

Cause. On 2 Nov, Arch Linux ARM's extra moved to Hyprland 0.57 (aquamarine 0.16, hyprgraphics 0.6, hyprtoolkit 0.6.0-3). Omarchy's rc aarch64 repo still has hyprland 0.56.2-7 and aquamarine 0.15.1-1.3, which come first. The same clash stops omarchy update on already migrated rc Macs. It's tracked in omacom#17368. So the preflight deferring with exit 75 is the right result. Your configuration isn't the problem, and that stopping point is a safe one.

A likely fix. omarchy-settings 4.5.0rc3 ships default/pacman/aarch64-apple/mirrorlist-rc, which points at https://arm-rc-mirror.omarchy.org/$arch/$repo, a frozen Arch Linux ARM snapshot. I ran a dry run against it with today's databases:

  • It has hyprland 0.56.2-4, aquamarine 0.15.1-1, hyprgraphics 0.5.1-4, hyprtoolkit 0.6.0-1 and hyprland-guiutils 0.2.2-4.
  • With [omarchy] rc in front, it resolves with no conflicts.

The tool's core_pacman_conf keeps the live mirror.archlinuxarm.org list (point 7 in @Knutsi's comment). Using the rc template's mirror list when the target is rc would probably avoid this break and later ones like it. It should also fix omarchy-version-channel reporting unknown, which checks for that mirror. I haven't run the tool with this change.

For readiness, my run is one data point: M1 Max, 4.0.4-mac.1, Limine, unencrypted, completed before the Arch Linux ARM bump. It doesn't cover LUKS2 or the M2 Max, so the rest is for the maintainers.

@martythedev-tech

Copy link
Copy Markdown

Follow-up from the M2 Air 13" (J413, legacy cohort, GRUB, unencrypted ext4) that ran omarchy-mac-migrate @2976e70d on 2026-11-02.

After comparing it package by package with a fresh 4.5.0rc3 install (MBP 16" M1 Pro, installer 2.3.13), I found that the migration switches the kernel/boot core (linux-aurora, m1n1-aurora, omarchy-mac, omarchy-mac-boot), but channel_packages() ("every package a migrated Mac takes from its channel's [omarchy]") stops at linux-aurora, m1n1-aurora and uboot-asahi, which leaves the rest of the platform on the old Asahi packages:

Migrated machine Fresh install
asahi-audio, alsa-ucm-conf-asahi aurora-audio (+ speakerguardd), alsa-ucm-conf-aurora
mesa 26.2.6, vulkan-asahi mesa-aurora, vulkan-aurora 26.3-devel
asahi-bless aurora-bless
uboot-asahi uboot-aurora
— omarchy-aurora-firmware (+ aurora-vendor-order initramfs hook)
asahi-meta (+ wifisync, btsync, fix27, configs, tiny-dfr) —

Everything kept working, so nothing is broken, but a migrated Mac quietly ends up on a different audio/graphics stack from a fresh one. That makes bug reports from migrated and fresh machines hard to compare.

The userspace swap went cleanly by hand (the aurora packages provide the asahi names, so omarchy-mac and other dependents are satisfied):

pacman -S --needed --noconfirm --ask 4 aurora-audio alsa-ucm-conf-aurora aurora-bless mesa-aurora vulkan-aurora

Reboot afterwards: Mesa 26.3-devel in use, the aurora-audio DSP graph loaded, no regressions so far. I haven't swapped the boot pieces (uboot-aurora, omarchy-aurora-firmware) yet.

Related, also only on migrated machines (different cause): A mic gain lowered before the mic mapper existed stays stranded on the hidden DSP source; filed as #789.

Would it make sense for the migration's target set to include these swaps, so a migrated Mac ends up matching a fresh install?

@greptile-apps

greptile-apps Bot commented Nov 5, 2026 •

Copy link
Copy Markdown

RetriggerView in GreptileConfidence Score: 4/5

[High impact] The PR can change the package and boot setup of eligible Apple Silicon Macs.

Reject missing explicit targets before merging so the tool cannot silently install a different package set.

Findings

  1. P1 Missing targets silently fall back ▶
  2. P2 Checking changes user settings ▶

Summary

Adds a standalone, resumable move from the Mac forks to official Omarchy packages.

  • Eligible Macs can move to official packages with one resumable command.
  • Mac-specific installs carry their boot setup into official packages.
  • Marked Macs switch before the fork’s normal update runs.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Run migration] --> B[Choose target and check signed packages]
  B -->|Not ready| C[Defer]
  B -->|Ready| D[Freeze plan and back up]
  D --> E[Download and rehearse transaction]
  E --> F[Switch repositories and arm boot resume]
  F --> G[Install packages]
  G --> H[Set up boot chain and users]
  H --> I[Verify and remove guard]
  I --> J[Wait for reboot]
  J --> K[Verify running kernel and boot files]
  K --> L[Retire fork state]
Loading

Reviews (1) · Last reviewed commit: "Merge branch 'quattro' into mac/migrate-..." · Reviewed by Greptile

Comment thread migrate/src/target.sh
Comment on lines +132 to +137
for candidate in "$@" "$admin_target"; do
if [[ -n $candidate && -e $candidate ]]; then
printf '%s\n' "$candidate"
return
fi
done

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 Missing targets silently fall back

find_target skips an explicit --target path when the file does not exist. For example, run --target /etc/omarchy-mac/candidate.typo instead uses the administrator's default target or the detected channel. The tool can then replace packages and the boot chain with a different set than the operator requested. Reject a missing explicit target before trying any fallback.

Comment thread migrate/src/engine.sh
Comment on lines +1509 to +1510
if [[ -f $complete && -f $plan/target-id ]]; then
tidy_completed

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

P2 Checking changes user settings

check calls tidy_completed before checking check_only. After a completed migration with pending user setup, this retries commands that change the user's files and settings. It can also remove the boot unit and saved tool. Skip completed-state cleanup during check so the advertised read-only command stays read-only.

@alexpomsft

Copy link
Copy Markdown

Field report: encrypted M2 Max / apple,j414c, legacy Quattro/custom Aurora SEP installation -> official Omarchy 4.5 stable, preserving the existing installation.

Important qualification: this used a locally modified build based on #690 head 933d74e459af36531f4b3d5771e165b627ece1a2, integrating the #726/#727/#777 fix stack plus local SEP-transition handling, stable archived repositories, an exact-artifact approval pause and manual post-reboot retirement hold. This is not a success report for the unmodified PR or a recommendation to run the stock script on custom SEP systems.

The coordinated target included Aurora audio/UCM, Mesa/Vulkan, bless, U-Boot, firmware and the official libfprint provider, addressing the package-parity concern in #690 (comment). The custom aurora-touchid package was removed through pacman while preserving SEP state and the official Touch ID opt-in; #778 remains relevant to the stock migration.

Resulting versions include:

Component Installed version
Omarchy/settings 4.5.0-3
Mac support / boot 0.2.3-2 / 20261104-1
Aurora kernel package 7.2.11.aurora14-1, running 7.2.11-14-1-ARCH
m1n1 / U-Boot 1.6.1.aurora22-1 / 2026.07.aurora6-1
Firmware 20261014-7
Audio / UCM 4.1.aurora3-3 / 10.aurora3-1
Mesa/Vulkan 1:26.3.0_devel.20261031.231483.a1969111475-1
libfprint-git 1:1.94.100.r10.g6f9479c-5

One actionable snag: the package transaction completed, but loader staging stopped with umount: /boot: target is busy. A running Steam/FEX libkrun VM held the ESP open. The adapter undid its loader-stage changes; after gracefully shutting down that VM, the same journaled attempt resumed at the loader step and succeeded. No forced unmount was needed. An early ESP-holder check or explicit warning to close Steam/FEX VMs before relocating /boot would improve the experience.

After applying the normal pending Omarchy configuration migrations and rebooting again, the official clean-environment boot check reported:

Apple Silicon boot check: running linux-aurora 7.2.11-14-1-ARCH; installed boot files match

No pending-reboot or rollback override was used. No failed system/user units were reported, and normal operation looks good so far. On J414, active speakersafetyd with inactive speakerguardd matches the shipped model guard; the latter is selected for J700.

The required core platform providers match the fresh-image builder's policy, but I have not compared the entire published image inventory or exhaustively tested every hardware path. Recovery backups and legacy boot assets are retained; final retirement has not run.

@rbadhesha

Copy link
Copy Markdown

Follow-up to my earlier M2 Max compatibility question and blocked preflight report: the in-place migration succeeded.

Machine: MacBook Pro 14-inch (M2 Max, 2023), previously Omarchy MX Mac 4.0.4-mac.1, with Aurora, Limine, and encrypted Btrfs/LUKS2.

Qualification: I used a locally corrected build based on 2976e70d81700161ccf9ae9bef03bdab5d42e2cb, not the unmodified PR. The correction used the frozen RC Arch Linux ARM mirror for the target package resolution. This resolved the dependency mismatch from my earlier report. The corrected tool SHA256 was 02d52730a76d38ad1c301db188fc21e021878cba7d789b04d3bd2390b4e6b96a.

Results:

  • Official Omarchy/settings 4.5.0-3, RC channel.
  • Mac support 0.2.3-2, boot package 20261104-1.
  • Active Aurora kernel 7.2.11-14-1-ARCH.
  • Post-reboot verification and retirement completed. An earlier verification attempt selected a retained old module tree and failed. A later attempt passed and wrote the completion record.
  • All 112 subsequent user migrations completed. System and user service checks reported no failed units, and Hyprland reported no configuration errors.
  • macOS still boots normally.

The migration left the older audio and graphics packages installed, consistent with the package-selection concern reported above. After migration, I separately replaced asahi-audio, alsa-ucm-conf-asahi, mesa, and vulkan-asahi with:

  • aurora-audio 4.1.aurora3-3
  • alsa-ucm-conf-aurora 10.aurora3-1
  • mesa-aurora and vulkan-aurora, both 1:26.3.0_devel.20261031.231483.a1969111475-1

The transaction also installed speakerguardd. On this M2 Max, speakersafetyd remains active and speakerguardd is inactive. Speakers sound normal. New OpenGL and Vulkan processes identified the M2 Max hardware and the new drivers, with accelerated OpenGL rendering. I did not measure microphone capture quality.

Boot-selection and firmware packages remain as installed by the migration. Recovery backups and migration records remain available.

Thanks for the frozen-mirror guidance and the migration work. This follow-up records the successful outcome of my earlier preflight report. The local corrections and separate audio/graphics replacement remain important qualifications.

@enriquebv

Copy link
Copy Markdown

Field report: unmodified tool, MX Mac 4.0.4-mac.1 → official Omarchy 4.5.0 stable, encrypted, in place. It completed. One post-reboot verify failure, root cause below.

Machine: MacBook Pro 13" M1 (apple,j293). MX Mac 4.0.4-mac.1 (omarchy-dev 4.0.4.r7099.gb8ed395-1), linux-aurora 7.1.12.aurora2-7, Limine, LUKS2 + Btrfs, ext4 /boot + FAT /boot/efi. No aurora-sep.

Tool: bin/omarchy-mac-migrate at 933d74e4 (sha256 07c01b67…c25c, matches the README). No code changes.

Avoiding the aquamarine ABI clash without patches: /etc/omarchy-mac/migration-target with channel=stable, and /etc/pacman.d/mirrorlist pointed at the frozen mirror from 4.5.0's aarch64-apple/mirrorlist-stable:

Server = https://arm-stable-mirror.omarchy.org/$arch/$repo

core_pacman_conf includes that file, so the tool resolved against the frozen snapshot and check was Ready. This tests @simon-nicholls's suggestion. Having the tool use the channel's mirrorlist template itself would remove the need for local patches.

Verify failure after the reboot:

this boot runs 7.2.11-14-1-ARCH, not linux-aurora 7.1.12-2-7-ARCH; the backups are in /var/lib/omarchy-mac/migration/backup

Cause: a race with linux-modules-cleanup.service. On the first boot the old /usr/lib/modules/7.1.12-2-7-ARCH tree was still there when omarchy-mac-migrate-verify.service ran. The cleanup moved it to /usr/lib/modules/.old during the same boot, but by then kernel_release linux-aurora had already picked it. The verify unit has no ordering against the cleanup. After=linux-modules-cleanup.service in the unit, or having kernel_release take only the package's current tree, should fix it. This is probably the failure @rbadhesha saw on the first verify attempt. Re-running systemctl start omarchy-mac-migrate-verify.service completed reboot and retire (State: complete).

Result: omarchy/omarchy-settings 4.5.0-3, omarchy-mac 0.2.3-2, omarchy-mac-boot 20261104-1, linux-aurora 7.2.11.aurora14-1, Limine. The 112 pending Omarchy migrations completed, there are no failed units, and hyprctl configerrors is clean.

+1 on package parity: asahi-audio/alsa-ucm-conf-asahi/mesa/vulkan-asahi stayed. I swapped them by hand for the aurora-* packages, and the J293 speakers work (speakerguardd stays inactive here, its unit is -C AppleJ700).

@seanfreiburg

Copy link
Copy Markdown

Field report: successful migration, legacy GRUB + LUKS, MacBook Pro 16" M1 Max (j316c)

Start: fork omarchy 4.0.2-2 (edge lane), GRUB, encrypt hook, LUKS2 + btrfs (@, @home, @log). Tool at 933d74e (sha256 verified).
End: omarchy 4.5.0-3 + omarchy-mac 0.2.3-2 on stable, Limine, linux-aurora 7.2.11, sd-encrypt. The verify service completed reboot and retire on the first boot. No failed units, and dotfile symlinks were untouched.

Pre-run prep (from earlier reports in this thread):

  • Commented out GRUB_FONT in /etc/default/grub (Drop the fork's GRUB_FONT once the ESP leaves /boot #727).
  • Wrote /etc/omarchy-mac/migration-target with channel=stable (otherwise it picks edge from the fork lane).
  • Pointed the mirrorlist at the frozen arm-stable-mirror.omarchy.org, to avoid the aquamarine/Hyprland ABI mismatch. check then returned Ready.

Run notes: the prefetch took about 4.5 min. 139 locally-newer packages were kept, and a few were downgraded slightly (hyprland-preview-share-picker, localsend, yay). The transaction and boot-chain steps were clean.

Two failures, both fixed by hand and resumed cleanly:

  1. loader: "the initramfs HOOKS after the switch do not unlock the root through systemd… block encrypt…"
    On 4.0.2-2, /etc/mkinitcpio.conf.d/omarchy_hooks.conf (and thunderbolt_module.conf) were not backup= files, but the new package declares them as backup. The on-disk copy still had the old HOOKS=(… encrypt …), so pacman kept it and wrote the new one as .pacnew. Moving both .pacnew files into place fixed it, and the rerun passed loader. Suggestion: in legacy_switch_unlock/legacy_drop_hooks, handle an omarchy_hooks.conf.pacnew next to an unmodified fork copy (or detect the fork's known contents).

  2. defaults: "Creating config failed (creating btrfs subvolume .snapshots failed since it already exists)" → "cannot set up Snapper for the root filesystem"
    An earlier fork Snapper had left @/.snapshots (with snapshots 7–11) but no /etc/snapper/configs/root, and snapper was no longer installed. install/config/snapper.sh runs create-config /, which hits a hard error rather than the exit 3 that repair_snapper treats as "left for manual repair". I moved the subvolume aside (mv @/.snapshots @fork-snapshots-old from subvolid=5) and reran; defaults → verify → unpin passed. Suggestion: treat an existing /.snapshots without a config as the manual-repair case (or adopt/move it), and preflight could flag it.

After the migration (omarchy update): coming from the fork's edge lane, about 138 packages were newer than stable (Hyprland 0.57.1 vs 0.56.2, etc.). omarchy update ran most migrations, then failed at 1791403310 (titlebars). omarchy-hyprland-titlebars pins hyprland=0.56.2, which forces downgrading aquamarine and hyprgraphics, and that breaks hyprtoolkit. The fix was to align with stable (snapper snapshot first):

  • pacman -Rdd clang-libs: newer clang is split into clang + clang-libs. Stable's monolithic clang 23.1.3 conflicts on 318 files.
  • OMARCHY_ALLOW_DIRECT_PACMAN=1 pacman -Syuu, then omarchy update. All remaining migrations passed, and the boot check passed after the reboot.

Suggestion: since edge-lane Macs land ahead of stable, the tool could offer the -Syuu alignment (or at least warn about it) when the target is stable.

Thanks for the tool. The journal/resume design made both of these painless to recover from.

@seanfreiburg

Copy link
Copy Markdown

Follow-up to the report above: Steam was broken after omarchy update, now fixed.

The "Orphan system packages" step of migration 1793667581 removed fex-emu-rootfs-arch, mesa-fex-emu-overlay-i386 and mesa-fex-emu-overlay-x86_64 (~1.4 GB). Neither omarchy-steam-fex (depends: bash coreutils python steam muvm FEX-Emu) nor FEX-Emu requires them, so they look like orphans. However, /usr/share/fex-emu/RootFS was left empty, and omarchy-launch-steam failed (muvm exits 248, "environment variable not found").

Fix:

sudo pacman -S --asexplicit fex-emu-rootfs-arch mesa-fex-emu-overlay-i386 mesa-fex-emu-overlay-x86_64

Steam launches again after that.

Suggestion: omarchy-steam-fex should depend on the rootfs and overlays (or the orphan cleanup should keep them), since any Mac that migrates with Steam installed will probably hit this.

@RaoulBrueker

Copy link
Copy Markdown

Field report: second MacBook Pro 13" M1 (apple,j293), MX Mac 4.0.4-mac.1 → 4.5.0 stable, encrypted, completed

Same starting point as @enriquebv's report (omarchy-dev 4.0.4.r7099.gb8ed395-1, Limine, LUKS2 + Btrfs, ext4 /boot, no aurora-sep), unmodified tool at 933d74e4 (sha256 matches). Prep: GRUB_FONT commented out, mirrorlist on arm-stable-mirror.omarchy.org, and a stable target file. check → Ready, run → clean, then omarchy update ran all pending migrations with none skipped and no failed units. It needed no manual fixes beyond the two notes below.

  • Verify race reproduced: the first boot failed with this boot runs 7.2.11-14-1-ARCH, not linux-aurora 7.1.12-2-7-ARCH. Re-running omarchy-mac-migrate-verify.service completed it. Since this is the second J293 to hit it, it looks deterministic on this path rather than a one-off.
  • Target file: a file with only channel=stable is refused (is not format=1). It needs all three lines:
    format=1
    type=repository
    channel=stable
    
  • Suggestion: when resolution fails, the refusal shows only tail -n 1 of resolve.log (could not satisfy dependencies), and the work dir is gone afterwards. The useful lines (:: unable to satisfy dependency 'libaquamarine.so=15-64' required by hyprtoolkit) are in the --print-format stdout. Including them in the refusal would make this failure self-explanatory.

Thanks for the tool!

@GehrmannSolutions

Copy link
Copy Markdown

Field report: MacBook Air 13" M3 (apple,j613), legacy GRUB + LUKS + self-built boot.bin → official rc, completed

As far as I can see, this is the first J613 in this thread.

Start: Omarchy fork on the [omarchy-aarch64] edge lane (legacy cohort), GRUB, busybox encrypt, LUKS2 + Btrfs (@, @home), ESP at /boot. The Mac was heavily customized for M3 bring-up: a self-built kernel next to linux-aurora, an /etc/default/update-m1n1 that pinned boot.bin to our own device trees, and the aurora-sep IgnorePkg block (#778).
Tool: unmodified bin/omarchy-mac-migrate, sha256 07c01b67…c25c (matches the README).
End: omarchy 4.5.0rc3 → omarchy update to 4.5.0-3 on rc, omarchy-mac 0.2.3-2, omarchy-mac-boot 20261104-1, linux-aurora 7.2.11.aurora14, m1n1-aurora 1.6.1.aurora22, Limine + UKI, sd-encrypt. The verify unit completed reboot and retire on the first boot. There were no failed units.

Pre-run prep: the same as earlier reports, plus four refusals that came from our own customizations. They are listed in case other bring-up Macs carry similar leftovers:

  • /etc/omarchy-mac/migration-target with channel=rc. Without it, detection picks edge because of the old lane URL. The mirrorlist pointed at arm-rc-mirror.omarchy.org (edge: omarchy/hyprland 0.57.0-2 blocks update after Arch moved to aquamarine 0.16.0 / hyprgraphics 0.6.0 omarchy#17368).
  • GRUB_FONT was commented out (Drop the fork's GRUB_FONT once the ESP leaves /boot #727).
  • New refusal: GRUB_CMDLINE_LINUX="${GRUB_CMDLINE_LINUX} asahi.t8122_start=1" is refused as shell expansion. Merging it into a single plain line fixed this. Because legacy_grub_value reads only the last GRUB_CMDLINE_LINUX line, a second line with cryptkey= had to go as well.
  • New refusal: check refused the boot-file consistency test because our update-m1n1 pin pointed at an unowned m1n1. It passed once the pin file was moved away and update-m1n1 was run against the packaged DTBs.
  • The legacy ESP space check also counts extra vmlinuz-linux-* files. Removing two old self-built kernels left an 18 MiB margin.
  • The IgnorePkg block was reduced to libfprint only.

check → Ready, run → clean in 12 minutes.

One surprise during run: the rc3 etc-overrides removed the omarchy-sudo-passwordless grant mid-transaction. The run itself was not affected, but anyone driving the migration through passwordless sudo should expect to re-grant it.

After migration, J613 on rc (aurora14):

Area Result
Speakers ✅ full sound, j613 DSP graph active
Microphone ✅ clear, background music filtered (j613-mic)
Brightness keys ✅ full range
Keyboard backlight ✅ via Ctrl+F1/F2
Cursor after unlock (#783) ✅ visible at once
Lid closed / s2idle / Wi-Fi after resume ✅
GPU (vkcube, Vulkan "Honeykrisp") ✅ no faults, runtime cap 2 = 618 MHz
Touch ID ✅ after removing leftovers, see below
Clock at boot (#782) ❌ 1970 on every cold boot until NTP
ISO keyboard (#350) ❌ ^/< swapped with hid_apple iso_layout=-1
External displays ❌ expected on aurora14 (dcpext not reserved); next test: aurora16 from edge
Wi-Fi with iwd ❌ BCM4388, details in #781; works with wpa_supplicant

Notes for the maintainers:

  1. Touch ID after aurora-sep: omarchy setup security fingerprint installed libfprint-git and wrote /etc/modprobe.d/omarchy-mac-touchid.conf, then failed with "device not found". Two leftovers were the cause:

    • an unowned /etc/modprobe.d/aurora-sep.conf from aurora-sep's read-only mode (options apple_sep xart_writes=0 provision_keybag=0);
    • the aurora-touchid package's apple-sep.service, which loads apple_sep early.

    Together they loaded the driver read-only, and the kernel refuses a second activation in the same boot. Removing both and rebooting fixed it: "identity keybag provisioned", then enrolment, sudo, polkit and the lock screen all worked. It might help if the migration or enable-fingerprint warned about conflicting apple_sep options.

  2. Package parity: the same as the J413 note above. This Mac still runs asahi-audio, mesa and asahi-bless, and audio and mic work fine on them.

  3. Leftover user unit: ~/.config/systemd/user/omarchy-asahi-mic.service from the fork still has ExecCondition=/usr/bin/omarchy-hw-apple, and that binary no longer exists. The mic works without it, but the unit is dead weight on migrated Macs.

  4. AUR during omarchy update: obsidian-appimage 1.15.0 fails on aarch64 because its split package also builds obsidian-bin from an x86 .deb. With set -e, omarchy-update-boot is then skipped. Running it by hand passed ("installed boot files match"). It might be worth running the boot verification even when the AUR step fails.

Disclosure: the preparation, the analysis and this text were made with an AI coding agent (Claude Code). The migration and all tests ran on the real J613 and were checked by me.

@arikisonfire

Copy link
Copy Markdown

Field report: MacBook Air 13" M2 (apple,j413), MX Mac 4.0.4-mac.1 on linux-asahi → 4.5.0 stable, unencrypted, completed

As far as I can see, this is the first MX Mac in this thread that started on linux-asahi rather than Aurora. It's also the first one after the fork's final package release (omarchy-dev 4.0.4.r7129.g5ec219b, the one with the "no longer maintained" notice).

Start: linux-asahi 7.2.12.asahi3-1 (running 7.2.12-3-1-ARCH), m1n1 1.9.9, uboot-asahi 2026.07.asahi2-3, Limine, Btrfs without LUKS, ext4 /boot, FAT /boot/efi, Snapper. grub was still installed next to Limine, but its GRUB_FONT file exists, so update-grub inside the transaction needed no change. A local IgnorePkg = hyprgraphics hyprland-guiutils had kept Hyprland at 0.56.2.

Tool: unmodified bin/omarchy-mac-migrate at 933d74e4 (sha256 07c01b67…c25c). I used the usual prep from this thread: a target file with format=1, type=repository and channel=stable, the mirrorlist on arm-stable-mirror.omarchy.org, and the hold removed. check → Ready: run moves this Mac (mx-mac, limine boot) onto repository https://pkgs.omarchy.org/stable/$arch (stable), then run → clean. Under 15 minutes from run to the second boot.

New compared to the earlier reports:

  • The kernel downgrade (7.2.12 → 7.2.11) was clean. 11-omarchy-mac-saved-modules.hook from omarchy-mac-boot 20261104 moved the old tree inside the transaction (Moving the running kernel's saved modules (7.2.12-3-1-ARCH) to /usr/lib/modules/.old …), so m1n1 got the Aurora device trees.
  • No verify race. The state was complete after the first boot without re-running the verify unit. Probably because the hook had already moved the only old tree, so linux-modules-cleanup had nothing to move at boot. That fits @enriquebv's analysis.
  • Full fresh-install parity works on J413, boot pieces included. After the first boot I ran pacman -S --needed --ask 4 aurora-audio alsa-ucm-conf-aurora speakerguardd mesa-aurora vulkan-aurora aurora-bless omarchy-aurora-firmware uboot-aurora quickshell. omarchy-apple-silicon-boot-check --boot-chain passed and the next boot was fine. That covers the uboot-aurora/omarchy-aurora-firmware swap @martythedev-tech hadn't done yet.
  • Orphans created by the move. In 4.5, omarchy update always runs omarchy-update-orphan-pkgs -y right after the migrations. So the first update after the move removes, without asking, everything the move turned into an orphan. Nothing here was an orphan before the move. Afterwards python-requests and vulkan-headers were, and I marked them --asexplicit before the first omarchy update. That's the step that took @seanfreiburg's FEX packages. The tool could mark, or at least list, the packages its own transaction orphans.

Result: omarchy/omarchy-settings 4.5.0-3, omarchy-mac 0.2.3-2, omarchy-mac-boot 20261104-1, linux-aurora 7.2.11.aurora14-1, m1n1-aurora 1.6.1.aurora22-1, uboot-aurora 2026.07.aurora6-1. omarchy update -y ran all pending migrations (none left). There are no failed system or user units, and hyprctl configerrors is clean. Wi-Fi (brcmfmac, NM + iwd) came up right away. This Mac already used wlan0, so the wld0 pins from #781 didn't apply. Speakers, internal mic, brightness keys, ISO keyboard keys, lid sleep and KVM all work. I haven't tested a USB-C display yet.

+1 on @seanfreiburg's -Syuu suggestion. The fork tracked live ALARM until the end, so 136 packages were newer than the frozen stable snapshot. omarchy update still passed here, 1791403310 (titlebars) included, because the hold had kept Hyprland at 0.56.2. Two more data points:

  • The fork's final package snapshot (04ae09fa, Nov 12) ships Hyprland 0.57.2 and aquamarine 0.16.0. An MX Mac that installed it without a hold starts above stable's Hyprland, like @seanfreiburg's edge Mac did. I haven't tested that path.
  • Later, installing a package that pulls poppler-qt6 (here signed) failed. Stable's poppler-qt6 26.08.0-1 needs poppler=26.08.0, while the fork's poppler-glib 26.10.0-2 needs poppler=26.10.0. Adding poppler poppler-glib to the same pacman -S (i.e. downgrading them to stable) fixed it.

Unrelated to the tool: on the first linux-aurora boots the RTC was 13.8 days behind (not 1970 as in #782), so the tool's complete timestamp says 2026-10-30. hwclock --systohc --utc after the NTP sync fixed it.

I ran the tool through a small wrapper script. It:

  • does the prep above, and undoes it if check refuses
  • runs run as a transient systemd unit under systemd-inhibit
  • restarts the verify unit if the first boot isn't complete
  • handles the parity swap and the orphans

It's tailored to this Mac, so I haven't published it, but I'm happy to share it. The wrapper and this report were written with Claude (Claude Code). All versions and messages above come from the logs of the real run.

Thanks for the tool!

@bibstha

bibstha commented Nov 13, 2026

Copy link
Copy Markdown

Field report: MacBook Pro 16" M1 Max (apple,j316c), quattro legacy (GRUB + busybox encrypt, ESP at /boot) → 4.5.0 stable, encrypted, completed with no manual step during run

This is the same cohort as @seanfreiburg's j316c and @Knutsi's preflight. It's the first run in this thread from a 4.0.3rc4-2 quattro install that had followed the [omarchy-aarch64] edge lane until today, and it started on linux-asahi.

Start: omarchy/omarchy-settings 4.0.3rc4-2, linux-asahi 7.2.12.asahi3-1, GRUB with cryptdevice=, LUKS2 + Btrfs, Snapper (config present), ESP (vfat) at /boot. The edge lane had taken Hyprland to 0.57.1, aquamarine to 0.16.0 and Firefox to 159.0.1. There was no IgnorePkg, and no aurora-sep, Steam/FEX or self-built kernels.
Tool: unmodified bin/omarchy-mac-migrate at 933d74e4, sha256 07c01b67…c25c (matches the README). Installed root-owned and run as a transient systemd unit under systemd-inhibit.

Prep (the usual set from this thread):

  • GRUB_FONT=/boot/grub/fonts/unicode32.pf2 commented out (Drop the fork's GRUB_FONT once the ESP leaves /boot #727). This install had it too.
  • /etc/omarchy-mac/migration-target: format=1 / type=repository / channel=stable
  • /etc/pacman.d/mirrorlist: Server = https://arm-stable-mirror.omarchy.org/$arch/$repo
  • A Snapper snapshot before starting.

check → Ready: run moves this Mac (legacy, grub boot, encrypted) onto repository https://pkgs.omarchy.org/stable/$arch (stable). It may remove linux-asahi m1n1 omarchy-mac-keyring. The harmless line 550: /sys/class/power_supply/hidpp_battery_3/capacity: No such file or directory from @Knutsi's report showed up too (Logitech receiver).

run: all 11 steps passed on the first attempt, 12m53s wall clock: about 6 min prefetch and conflict scan, about 5.5 min transaction, then boot-chain, loader, defaults, verify, unpin in under a minute. The harmless database file for 'omarchy-aarch64' does not exist warning appeared as reported.

  • omarchy_hooks.conf was not a trap here. It was owned by omarchy-settings 4.0.3rc4-2 and pacman -Qii showed it [unmodified], so the same-name upgrade replaced it with the new sd-encrypt-compatible file, and no .pacnew was left. This confirms @Knutsi's correction: only modified copies, or older installs where it wasn't a backup= file (@seanfreiburg's 4.0.2-2), hit it.
  • The only new .pacnew files on the whole system were two tpm2-tss FAPI profiles.
  • hid_apple.conf with fnmode=1 was kept (unlike @simon-nicholls' fnmode=2 case).

First boot: Limine, the sd-encrypt passphrase prompt, and the verify unit completed reboot and retire on the first boot. No verify race. The single old module tree had already been moved during the transaction, as in @arikisonfire's run (also a 7.2.12 → 7.2.11 kernel downgrade). No failed units.

Aligning with stable (needed, as @seanfreiburg found): the migration's transaction made 37 package changes (4 downgrades) and kept everything the edge lane had made newer. So before the first omarchy update, I ran pacman -Rdd clang-libs, then OMARCHY_ALLOW_DIRECT_PACMAN=1 pacman -Syuu (snapshot first). It finished cleanly in 2.5 min and rebuilt the initramfs and UKI with sd-encrypt. Notable downgrades: Hyprland 0.57.1 → 0.56.2, aquamarine 0.16.0 → 0.15.1, clang 23.1.4 → 23.1.3, poppler 26.10 → 26.08 (which avoids @arikisonfire's poppler-qt6 clash), dua-cli 3.0 → 2.46, Firefox 159.0.1 → 158.0.

  • Firefox: after the downgrade, Firefox refused the existing profile ("Using an older version of Firefox can corrupt bookmarks…") and offered a new one. Backing up the profile and starting it once with firefox --allow-downgrade fixed it, and the profile opened normally (one major version back). Worth a line in the README for edge-lane Macs going to stable.
  • After the alignment, asar, go and simdutf were orphans. I marked them --asexplicit before the first update, because omarchy update runs omarchy-update-orphan-pkgs -y (+1 to @arikisonfire's suggestion that the tool list the orphans its own transactions create).

First omarchy update -y: all migrations ran, including 1791403310 (titlebars), thanks to the alignment. It then failed at the AUR step on obsidian-appimage 1.15.0, the same as @GehrmannSolutions' note 4 (package_obsidian-bin(): bsdtar: Error opening archive: Failed to open '…/data.tar.xz'), which also skipped the boot-file check. Stable's [omarchy] repo now ships a native obsidian 1.15.1 for aarch64 that conflicts with obsidian-appimage. Swapping (pacman -R obsidian-appimage && pacman -S obsidian) and running omarchy update -y again made it pass. Suggestion: the move (or a post-move migration) could replace obsidian-appimage/obsidian-bin with [omarchy]'s obsidian on aarch64, since every migrated Mac with Obsidian installed hits this.

Other changes after the update, for anyone with customised configs:

  • Migration 1791382872 ("native lazy shims") removed the generated wrappers from ~/.local/bin (gh, claude, codex, …). That's correct, but a git credential.helper that hardcoded ~/.local/bin/gh broke (could not read Username for 'https://github.com'). Pointing it at ~/.local/share/mise/shims/gh fixed it.
  • tensaku was replaced by Omasnap (1788129995), and power-profiles-daemon by omarchy-power-profiles-daemon.
  • Symlinked ~/.config/hypr and other dotfile symlinks were left as symlinks. The migrations appended to bindings.lua and looknfeel.lua through the symlink, and changed the VS Code theme to Omarchy.
  • Keybindings: the fork's Apple top-row screenshot binds (SUPER+F10/F11/F12, SUPER+mute/volume keys, and their screen-recording variants) are gone. New defaults include ALT+SPACE / hold Right Alt for dictation and CTRL+SPACE as the terminal prefix, which users of Option-key compose may want to know about. 4.5 adds a SUPER+A default, so a user SUPER+A bind needs hl.unbind("SUPER + A") first.
  • The lock-stops-working-after-first-unlock bug I had patched locally on 4.0.3rc4 is fixed upstream in 4.5. Service.qml now decides with currentlyLocked() instead of the stale locked binding.

Leftovers (not harmful): the fork's ~/.config/systemd/user/omarchy-asahi-mic.service with ExecCondition=/usr/bin/omarchy-hw-apple (binary gone, unit inactive), the same as @GehrmannSolutions' note 3. Also the unowned /etc/mkinitcpio.conf.d/apple_hid_modules.conf, and grub plus /boot/grub (as @AmmonMatthews noted).

Package parity: I swapped to the fresh-install stack afterwards with pacman -S --needed --ask 4 aurora-audio alsa-ucm-conf-aurora speakerguardd mesa-aurora vulkan-aurora aurora-bless. The transaction was clean: the asahi counterparts were removed, there were no failed units and no orphans. I'll follow up after the reboot on audio, mic and GPU. I haven't swapped the boot pieces (uboot-aurora, omarchy-aurora-firmware) yet.

End state: omarchy/omarchy-settings 4.5.0-3, omarchy-mac 0.2.3-2, omarchy-mac-boot 20261104-1, linux-aurora 7.2.11.aurora14-1, limine 12.9.3-1, Hyprland 0.56.2, ESP at /boot/efi, rd.luks.name= unlock. No failed system or user units, and hyprctl configerrors is clean.

Thanks for the tool. Its preflight-then-journal design made a legacy GRUB + LUKS move uneventful.

Disclosure: the preparation, the checks and this report were done with Claude Code. Every version, message and timing above comes from the real run on this Mac.

@bibstha

bibstha commented Nov 13, 2026

Copy link
Copy Markdown

Follow-up to my j316c report above, after the aurora audio and graphics swap and a reboot:

  • GPU: vulkaninfo reports Apple M1 Max (G13C C0), Mesa 26.3.0-devel (git-a196911147), Borealis, derived from Honeykrisp (mesa-aurora/vulkan-aurora 26.3.0_devel.20261031). The desktop runs normally and there are no failed units.
  • Audio: the j316-convolver filter chain gives the "MacBook Speakers" sink, and speakersafetyd is active. As expected, speakerguardd is inactive on J316. Headphones through the jack work. WirePlumber kept the user's saved default sink across the swap. I haven't listened to the speakers yet, because headphones were plugged in.
  • Mic: j316-mic (effect_output.j316-mic) is the default source, and a 3 s capture gave real signal (room noise around −55 dBFS RMS).
  • Correction and addition to the leftover note (and @GehrmannSolutions' note 3): the fork's ~/.config/systemd/user/omarchy-asahi-mic.service isn't just dead weight. It has the same name as omarchy-mac's /usr/lib/systemd/user/omarchy-asahi-mic.service, so the user copy shadows the packaged unit. Its ExecCondition=/usr/bin/omarchy-hw-apple points to a binary that's gone, so the mic mapper (omarchy-audio-asahi-mic-map --watch) is skipped on every boot on migrated Macs. Removing the user copy (rm ~/.config/systemd/user/omarchy-asahi-mic.service && systemctl --user daemon-reload && systemctl --user enable --now omarchy-asahi-mic.service) brings up the packaged unit, with ExecCondition=/usr/bin/omarchy-hw-apple-silicon. Suggestion: retire could remove that user-level copy when it matches the fork's template, or a post-move migration could.
  • gkr-pam: couldn't unlock the login keyring at an autologin is harmless here. It was there before the move too.
  • Also seen at boot, not caused by the move I think: Mount point '/boot/efi' which backs the random seed file is world accessible (the new /boot/efi fstab entry has fmask=0022,dmask=0022; 0077 would silence it).

@martythedev-tech

Copy link
Copy Markdown

Small correction to @arikisonfire's note: I did do the boot-piece swap on my M2 Air (apple,j413) on 11-04, after the audio/mesa step. uboot-asahi → uboot-aurora 2026.07.aurora6-1 and omarchy-aurora-firmware 20261014-7, then a UKI rebuild. Reboots have been fine since, and the initrd loads the vendor firmware. So that's two J413s with full boot-piece parity.

@aabulkhairov

Copy link
Copy Markdown

Preflight-only report from an M1 MacBook Air (apple,j313). I haven't seen a j313 report in this thread yet. Migration not run.

Setup: quattro install from the fork, encrypted, GRUB, ESP at /boot, edge.

  • omarchy/omarchy-settings 4.0.3-1 (checkout at 4.0.3rc4), linux-asahi 7.2.12.asahi3-1, grub 2:2.16-3, Hyprland 0.57.2
  • Root: LUKS (nvme0n1p5) → btrfs @. Busybox encrypt hook via cryptdevice=
  • ESP: nvme0n1p4 vfat mounted at /boot
  • pacman.conf: [omarchy-aarch64] edge lane above [omarchy] (pkgs.omarchy.org/edge), plus [asahi-alarm], [alarm] and an extra [aur] section
  • Local holds: IgnorePkg = qt6-* poppler poppler-glib poppler-qt6 quickshell. Qt 6.12 made the Omarchy bar invisible here (the 'Color' clash, upstream Bar invisible on Qt 6.12: QtQuick's built-in Color shadows qs.Commons Color singleton omarchy#18012), so qt6-base is held at 6.11.2-3 and quickshell at 0.3.1-1

Command: sudo bash omarchy-mac-migrate check, using bin/omarchy-mac-migrate at 933d74e4 (sha256 07c01b67…c25c, which matches the README; the mac-migrate-v1 release asset still returns 404)

Result: exit 0

gpg: next trustdb check due at 2026-12-31
Ready: run moves this Mac (legacy, grub boot, encrypted) onto repository https://pkgs.omarchy.org/edge/$arch (edge).
It installs: omarchy-dev omarchy-settings-dev omarchy-mac omarchy-mac-boot linux-aurora m1n1-aurora uboot-asahi limine-mkinitcpio-hook aether avd-fw cliamp ghostty ghostty-shell-integration ghostty-terminfo herdr libva-v4l2_request-avd localsend mise-bin obs-studio omacalc omacut omarchy-keyring omawrite openai-codex-desktop tensaku ttf-ia-writer ttf-jetbrains-mono-nerd-basic ttfx tzupdate ufw-docker xdg-terminal-exec yaru-icon-theme yay asahi-alarm-keyring
It may remove: linux-asahi m1n1 omarchy omarchy-settings

Questions:

  1. The plan doesn't list quickshell or any qt6-* package, so it looks like the IgnorePkg hold is being respected. Is it safe to run with the hold in place, and will 4.5's shell work on quickshell 0.3.1 / Qt 6.11? Or does 4.5 already avoid the Qt 6.12 Color clash, so I can drop the hold?
  2. Does the extra [aur] repo section get in the way of the core pacman.conf swap?
  3. The plan installs tensaku, but earlier reports here say 4.5 replaced it with Omasnap. Is that expected on edge right now?
  4. Should I wait for the merge, or is a manual run on j313 useful to you as a tester data point?

Disclosure: Claude Code helped me prepare these checks and this report. Every version and the output above come from this Mac.

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.