Unstick migration 1786643346 when it insists a browser window is open
Checked all three claims against the shipped migration on this workstation, which runs omarchy 4.0.2-1, and re-read both issues and the pull request today. Confirmed on this machine that profile_open() at lines 129 to 133 of /usr/share/omarchy/migrations/1786643346.sh tests only -L $1/SingletonLock, -e $1/SingletonLock and -S $1/SingletonSocket, with no liveness check on the PID the symlink names, and that the same function is byte-identical on quattro and at tag v4.0.3. Confirmed from the GitHub API that PR 7026 is still open on 2026-09-11, so the fix section stands. Confirmed on this machine that grep -rn SingletonCookie /usr/share/omarchy/ returns nothing, so deleting SingletonCookie is harmless and is not what unsticks the migration, exactly as the record says. Also confirmed on this machine: omarchy-migrate sets set -euo pipefail at line 6 and runs each migration with bash -euo pipefail at line 93, touching the marker only after it succeeds at lines 94 to 95, and omarchy-update sets set -e at line 8 and calls omarchy-migrate at line 48, so the symptom's blast radius is right. The marker path and name in the verify block are right, ~/.local/state/omarchy/migrations/1786643346.sh, and the record's ls globs cover all thirteen profile roots listed at lines 80 to 94. Both issues support the claim. Issue 7019 is the mechanism thread, and issue 7453 is the version and browser evidence, reporting Omarchy 4.0.0-1 after an Omarchy 3 to Quattro upgrade with Brave, with a second reporter on Vivaldi and Chromium. Three things needed correcting. The cause said the manual removal is the fix through 4.0.2, which went stale when v4.0.3 shipped on 2026-09-08 with the same unchanged function, and it did not mention PR 6872, the competing open PR, or upstream's own finding that 7026's kill -0 test still misreads an EPERM PID, a reused PID, a foreign hostname and a socket-only profile. The danger was incomplete: it named SingletonLock only, although the fix also deletes SingletonSocket and the check accepts the socket alone, and it omitted the quiet failure the migration's own comments at lines 109 to 112 and 186 to 196 describe, where removal under a live browser gets the migration marked done while the browser reverts the repair on exit. I added the profile-directory warning because a reporter in 7019 deleted ~/.config/BraveSoftware wholesale. The fix also claimed the two rm commands were what reporters confirmed, and they are not verbatim: both reporters' commands included SingletonCookie, so I kept the commands and described their provenance accurately. Not exercised: I did not delete any Singleton file, did not run omarchy-migrate or omarchy update, and this user has no Singleton files in any of those profile roots, so the unstick itself was not reproduced here and rests on both threads. Worth knowing but not a defect: the migration never removes Preferences.omarchy-copy-url-repair.bak, the only references being lines 75, 117 and 148, so a repaired profile stays eligible for the same gate if the completion marker is ever lost.
Only remove these markers while no Chromium-family browser is running, and the same caution covers SingletonSocket as well as SingletonLock, because the shipped check accepts either one on its own.
Removing them under a live browser does two kinds of damage. It lets a second instance open the same profile, which can corrupt that profile. It also defeats the prompt silently: the migration then repairs Preferences under the running browser, its own post-repair re-checks read the profile as closed too, omarchy-migrate writes the completion marker, and the browser writes its stale in-memory Preferences back when it exits. You are left with the migration recorded as done, the Copy URL shortcut still broken, and no prompt to tell you.
Do not delete a whole profile directory to clear the lock. One reporter removed ~/.config/BraveSoftware outright, which loses that profile's bookmarks, history, saved passwords and extensions.
Symptom
omarchy update stops at migration 1786643346, "Repair the Copy URL shortcut for profiles that predate its pinned extension id", and asks you to close the browser:
Close the browser windows to repair the Copy URL shortcut, then continue
It asks again however many times you confirm, with every browser closed, after pkill -f on all of them, and after several reboots. Declining prints:
A running browser would undo the Copy URL shortcut repair.
Close the browser windows, then run: omarchy-migrate
and exits 1. Because omarchy-migrate runs under set -euo pipefail and omarchy-update calls it under set -e, every migration sorting after it never runs and the rest of the update is skipped: the post-update hook, AUR packages, mise and the orphan cleanup. A pending-migration notification then appears at every login.
Reported on Omarchy 4.0.0-1 stable and after an Omarchy 3 to Quattro upgrade, with Brave, and confirmed by a second reporter with Vivaldi and Chromium.
Cause
Established in the threads and confirmed against the shipped migration on an omarchy 4.0.2-1 install, where profile_open() at line 129 reads:
profile_open() {
[[ -L $1/SingletonLock || -e $1/SingletonLock || -S $1/SingletonSocket ]]
}
The migration never looks at processes or windows. It decides a Chromium-family profile is open purely from those filesystem markers in the profile's user-data-dir. Chromium-family browsers create SingletonLock as a symlink to <hostname>-<pid> and unlink it on a clean exit, but leave it behind after a crash, a SIGKILL or a hard power-off. It lives under ~/.config/, so a reboot never clears it and pkill has nothing left to signal. The guard is a while loop, so a marker that is permanently present re-prompts forever.
affected_profile_open() gates on two sets of profiles rather than on browsers in general: the profiles whose Preferences still need the repair, and any profile root holding a Preferences.omarchy-copy-url-repair.bak from an attempt the migration has not yet verified. So the stale marker can sit in a profile you stopped using long ago. The second reporter's was an old Brave profile.
PR 7026 makes profile_open() read the SingletonLock symlink target and treat the lock as stale when kill -0 on the PID it names fails. It is still open on 2026-09-11, and so is the competing PR 6872 against the same function. Upstream's review of 7026 is that its test is incomplete: kill -0 returns EPERM for a live PID this user cannot signal, a PID reused after a reboot still reads as open, the hostname half of the target is discarded, and a lock left as a plain file or a profile with only SingletonSocket left behind still blocks. profile_open() is byte-identical on quattro today and at tag v4.0.3 (published 2026-09-08), so the manual removal below is still the fix on every shipped release up to and including 4.0.3.
Fix
With every Chromium-family browser genuinely closed, find the stale markers:
ls -l ~/.config/{chromium,google-chrome*,BraveSoftware/Brave-Browser*,microsoft-edge*,vivaldi,opera,helium}/Singleton* 2>/dev/null
Those globs cover all thirteen profile roots the migration scans, including the beta and nightly variants. A dangling symlink to a <hostname>-<pid> that no longer exists is the signature. Delete SingletonLock, and SingletonSocket if one is there, in whichever profile roots turn up. The shipped check accepts the socket on its own, so leaving it behind keeps the migration stuck. For the two combinations reporters were stuck on, Brave plus Chromium and Chromium plus Vivaldi:
rm -vf ~/.config/BraveSoftware/Brave-Browser/Singleton{Lock,Socket} ~/.config/chromium/Singleton{Lock,Socket}
rm -vf ~/.config/chromium/Singleton{Lock,Socket} ~/.config/vivaldi/Singleton{Lock,Socket}
Both reporters also deleted SingletonCookie, which sits beside them. That is harmless, and it is not what unstuck the migration: SingletonCookie appears nowhere in /usr/share/omarchy, so no shipped check looks at it.
Then run the migrations again, which also finishes the rest of the interrupted update:
omarchy-migrate
omarchy update
Verify
omarchy-migrate # runs to the end without the prompt
omarchy-migrate --pending # prints nothing
ls ~/.local/state/omarchy/migrations/ | grep 1786643346 # the completion marker is written
No migration notification appears at the next login. Both reporters confirmed the migration ran through once the stale lock files were gone.
Sources
- https://github.com/omacom/omarchy/issues/7019
- https://github.com/omacom/omarchy/issues/7453
- https://github.com/omacom/omarchy/pull/7026
- https://github.com/omacom/omarchy/blob/quattro/migrations/1786643346.sh
- https://github.com/omacom/omarchy/pull/6872
- https://github.com/omacom/omarchy/blob/v4.0.3/migrations/1786643346.sh
- https://github.com/omacom/omarchy/releases/tag/v4.0.3