Repository navigation
Merge upstream quattro (monitor workspaces, mise shims, OpenClaw move) - #87
Merged
Merged
Conversation
* Push a new language's LANG into Hyprland and the shell uwsm-app runs apps in a scope, which keeps its launcher's environment, so after a switch apps opened from bindings or the launcher still used the login's LANG. Set it in Hyprland and respawn the shell from there. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Push LANG to the session in one step hl.env's third argument also exports the value to the systemd and D-Bus activation environments, so it replaces the separate dbus-update-activation-environment call. The shell is only respawned once Hyprland has the new value, after the choice is persisted. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Drop the duplicate systemd environment import at startup dbus-update-activation-environment --systemd --all already imports the whole environment into systemd as well as D-Bus activation. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Omarchy's default CLI tools (gh, uv, the agent CLIs and the rest) are now lazy tools declared in /etc/mise/conf.d/omarchy-tools.toml, which the settings package ships. mise builds native shims for them and installs each one on first run, replacing the handwritten wrappers, which the migration removes. Agent CLIs update themselves through auto_update at most every four hours. Remove Preinstalls opts a user out through ignored_config_paths. Requires mise 2026.10.11. Co-authored-by: Ryan Hughes <ryan@heyoodle.com>
omacom#15809)" (omacom#15854) This reverts commit f5b2376.
omarchy update runs migrations on a fixed system PATH with no ~/.local/bin, so the installer's closing check that the first 'openclaw' on PATH is the runtime found nothing and failed the update after the move had already succeeded. No PATH the installer can be started with is the one a session uses: the update leaves out /usr/local/bin, mise's shims and ~/.local/bin, the desktop puts the shims first, and a terminal adds mise's tools ahead of those. So the checks no longer ask what comes first. The runtime has to be the only openclaw on the caller's PATH or in any of the directories the update leaves out, which answers the same in every context, catches a mise shim or /usr/local/bin openclaw the update could not see, still catches an openclaw the package brings, and keeps --check meaning the default agent can launch the runtime. The test's PATH no longer reaches the runner's own commands, and its copy of the installer looks in a fixture directory instead of the real /usr/local/bin, so an openclaw installed on the machine running it is never found. Co-Authored-By: Codex XHigh <noreply@openai.com>
The desktop runs mise's shims from MISE_DATA_DIR, or XDG_DATA_HOME/mise, ahead of everything else, so an openclaw shim kept there was missed when the update's fixed PATH hid it and the data directory was not the default. Both variables come from the environment rather than mise's config, and omarchy update passes them through from the session, so the lookup reads them. The default directory stays too, since env-bootstrap puts it on every session's PATH whichever directory mise uses. Co-Authored-By: Codex XHigh <noreply@openai.com>
MISE_SHIMS_DIR takes precedence over MISE_DATA_DIR/shims, so with both set the lookup searched a directory mise no longer used and missed the one it did. It now follows the same precedence, still keeping the default directory env-bootstrap puts on every session's PATH. mise's shims_dir setting in its config file is not read: that would mean running mise during the update, and the default agent's --check still refuses such a shim on the session's real PATH. Co-Authored-By: Codex XHigh <noreply@openai.com>
The OpenClaw migration needs the openclaw package to be the seed, but that package and this release reach a channel separately: the fast ring sent the seed out ahead of the release, and the release train cannot publish the pinned omarchy pair and a package into one channel at the same moment. With the release first, the migration exited 1 and every OpenClaw user's omarchy update stopped there until the package caught up, though their OpenClaw still worked. A migration can now exit 75 (EX_TEMPFAIL) to defer. The runner leaves it unmarked, carries on with the rest of the queue and runs it again on every omarchy update; --pending leaves it out, so the login notifier does not nag and the 4.0 upgrade does not refuse to finish over it. The OpenClaw migration defers while the package is still the runtime, so the release can ship first and the seed can follow whenever it reaches the channel. Co-Authored-By: Codex XHigh <noreply@openai.com>
Any command inside a migration that happened to exit 75 read as a deferral, so a migration now defers only when it also writes why to $OMARCHY_MIGRATION_DEFER; a bare 75 is an ordinary failure. A deferral record also outlived what it described: it hid the migration from --pending even after a direct pacman or another user's update brought the package it was waiting for, leaving nothing to tell that user, and it survived a later real failure. The record now holds the package database's timestamp and stops hiding the migration once any package changes, and every retry clears it first, so only a fresh deferral writes it back. Co-Authored-By: Codex XHigh <noreply@openai.com>
… stale A deferral is pending again once any package changes, but a later migration in the same run can change packages itself, as removing Elsewhen does, and that left the deferral looking pending the moment the run ended: the 4.0 upgrade's final check would refuse to finish and the notifier would ask for a run that had just happened. Each migration that deferred is now run once more at the end if packages changed after it, which also catches the rare later migration that brings the package it waits for. And a deferral only counts as still waiting against a real package database timestamp; without one there is nothing to show the wait holds, so it is pending. Co-Authored-By: Codex XHigh <noreply@openai.com>
With this release out and the channel still serving the openclaw package that is OpenClaw itself, Install > AI installed that package and then failed with "The installed OpenClaw package cannot set up a self-updating OpenClaw. Run 'omarchy update'", which no update could fix until the seed shipped. Choosing OpenClaw as the default agent failed too, its --now refusing the package's own /usr/bin/openclaw as another OpenClaw. Until the package is the seed it is the OpenClaw a machine has, as it was before this release: --check takes it as installed and --now installs it and stops, and the migration moves it once the seed arrives. The web app takes its icon from wherever the installed package keeps it.
The move stops a running gateway before seeding, so a run that then fails, say because nodejs.org or the npm registry is unreachable, leaves it stopped. The next run read that stopped gateway as one the user had stopped, moved it, and left it off: OpenClaw finished moving with its gateway down until the next login or launch. A run now records the services it stops, and the next run that completes starts them again, whether it moved them itself or an earlier run's installer already had.
Taking the package that is OpenClaw itself as the installation is right on a machine that never moved, but on one that had already moved and then got that package back, as the runtime rollback did to edge, it put an older copy ahead of the self-updating one on PATH and called it installed; launching then hangs against the newer gateway. Such a machine is now told the package is in the way until the seed returns, and nothing is touched. The record of a stopped service is written before the stop, so failing to write it stops nothing; each record goes as soon as its own service is running again, so one that moved is not restarted later after the user stops it; a service that will not start again keeps its record and fails the run instead of the migration finishing with it down; and Remove > AI clears the records with the services. Co-Authored-By: Codex XHigh <noreply@openai.com>
The migration runs once per user, and on a machine without the openclaw package it finds nothing to move and is marked done. Anyone who then installs OpenClaw before the seed ships gets the package that is OpenClaw itself, and when the seed arrives nothing moves them: the upgrade takes the old runtime away and leaves them with no openclaw command and a gateway on deleted files. Installing or accepting that package now clears the migration's record, so it waits for the seed again and moves them when it arrives.
…hey install it Clearing the migration's record from --now only reached installs that went through Omarchy's installer, and left the migration pending straight away, so the login notifier asked for a run that could only wait. omarchy-migrate now has --rearm, which clears a finished migration's record and records it as waiting, the way a deferral does, so it runs on the next update without being pending before a package changes; the installer rearms through it. An install by another route, or another user on the same machine, is still never rearmed, so the launcher sets OpenClaw up when the package is installed but no openclaw command is, which is the state the seed leaves such a user in. And Remove > AI drops each service's restart record as soon as that service is gone, so a removal that fails later does not leave a record for a unit that no longer exists. Co-Authored-By: Codex XHigh <noreply@openai.com>
…there omarchy-agent refused a default agent whose command was missing before dispatching it, so a user whose OpenClaw was left unmoved by the seed, with the package installed and no openclaw command, got "openclaw is not installed" from the agent key and never reached the launcher that now sets OpenClaw up. With the openclaw package present, a missing command is the launcher's to fix. Co-Authored-By: Codex XHigh <noreply@openai.com> Co-authored-by: Mark Groves <mark-groves@users.noreply.github.com>
The agent opens OpenClaw's TUI in a terminal that closes when the launcher exits, so when the launcher's fallback setup failed the terminal vanished with the installer's message. It now waits for a key, as Omarchy's other terminals do after a failure. Co-Authored-By: Codex XHigh <noreply@openai.com>
…a change A deferral is hidden from --pending while packages are unchanged, and it took that from the package database's mtime in whole seconds, so a package that arrived in the same second the migration deferred, the seed among them, left it hidden and the login notifier quiet until the next unrelated package change. The stamp is now the list of installed packages, which any install, upgrade or removal changes however close in time; a finer mtime would only shrink the window to the filesystem's clock tick. It is listed in the C locale's order, so a terminal and the login notifier compute the same stamp, and a database that cannot be listed gives none, so the deferral counts as pending instead of matching a stamp that never changes. The end-of-run second look also ran a deferral again whenever no stamp could be read, so without a package database every deferred migration ran twice a run; it now runs again only when a change is shown. Co-Authored-By: Greptile <165735046+greptile-apps[bot]@users.noreply.github.com> Co-Authored-By: Codex XHigh <noreply@openai.com>
A run that fails after stopping the gateway records it so the next run starts it again, but if the user then removed the gateway, say with openclaw gateway uninstall, every later run tried to start a unit that no longer existed and failed, telling them to start it by hand, which failed too. A record whose unit file is in none of the directories the running user manager loads units from is now dropped without starting anything. The directories are the manager's own UnitPath, read over D-Bus as JSON so a directory with a space in its name stays one directory, which means a unit moved to another of them is still started and a caller with a different XDG_CONFIG_HOME still finds it; and the files decide rather than the manager's loaded state, which still calls a unit deleted without a daemon-reload loaded and one it cannot read not-found. Only a complete answer counts: when the manager cannot be asked, or the lookup fails partway or comes back empty, the service is tried as before. Co-Authored-By: Greptile <165735046+greptile-apps[bot]@users.noreply.github.com> Co-Authored-By: Codex XHigh <noreply@openai.com>
…ed arrives While the openclaw package is still OpenClaw itself, --check took it as installed and --now accepted it without asking what a session would run, so a user's own openclaw ahead of /usr/bin, a mise shim from an npm install say, passed both: OpenClaw could be chosen as the default agent and the agent key then ran that other copy. The package's command is now held to the rule the runtime is, that it is the only openclaw a session might run, and anything else is named and refused. The migration is still re-armed first, so the move happens once the other copy is gone and the seed arrives. Co-Authored-By: Greptile <165735046+greptile-apps[bot]@users.noreply.github.com>
4.0.5 shipped the move as 1790397381 before the seed package, so on every machine with the openclaw package it fails and stops the update (omacom#15804). It is already marked done on every machine that followed the workaround on omacom#15804, and on dev machines that updated while omacom#15809 had it retired on quattro, and omarchy-migrate never runs a migration marked done again. A fix to that file would never reach them, and once the seed package replaced /usr/bin/openclaw they would be left without OpenClaw. So 1790397381 now finishes everywhere without changing anything, and the move that waits for the seed is 1792334754.sh, which every machine runs once whatever became of the old one; the package's install rearms that one.
A gateway set up by an older OpenClaw keeps its agent database at that release's schema, and the runtime the move seeds refuses it: the gateway starts, then exits 78 a few seconds later asking for `openclaw doctor --fix`. systemctl had already called it active, so the move finished with OpenClaw's gateway down, on exactly the 4.0.4 users whose OpenClaw the move is for. A gateway that was running now has to answer before the run counts it back, wherever it comes back: when the move reinstalls it, and when a later run starts one a failed run stopped. The move's own check only asks whether the unit left the old package; whether the gateway is up is this step's to judge, through systemd's restart wait or after an exit. It keeps its record until it answers, so another service failing after it moved cannot leave it unrepaired, and a record only repairs a gateway the user manager runs from Omarchy's runtime, by its effective command, drop-ins included and whichever directory its unit came from, not one the user has since pointed at an OpenClaw of their own. When it does not answer, the run stops it, and only once it has stopped runs OpenClaw's automatic repairs (`doctor --fix --non-interactive`, which backs the databases up before migrating them), then starts it again; one that still does not answer fails the run and keeps its record for the next update. Only the local gateway's own `"ok": true` counts, asked on the port its service is started with, which can differ from the configured one. health also exits 0 while a gateway is still starting, and honours a caller's OPENCLAW_GATEWAY_URL, port and credentials, which could name another gateway, so those are cleared for the probe. A unit in systemd's restart wait is waited for, up to two minutes, rather than doctored: the unit restarts on any exit but 78. Co-Authored-By: Codex XHigh <noreply@openai.com>
* Move OpenClaw only when the seed is there, without deferring migrations 4.0.6 and the seed package reach each channel together, so the OpenClaw move does not need a migration that waits for another repository's package. 1792334754 now moves OpenClaw when the package is the seed and otherwise finishes with nothing to do, as a machine without the package already did. omarchy-migrate, its test and agents/skills/migrations.md go back to what they were before omacom#14212, and installing the package no longer rearms the move: a machine that gets the seed without the move has its OpenClaw set up by the launcher the agent key and the web app go through. * Drop the OpenClaw test's copy of omarchy-migrate and its package database Only the deferral and rearm cases used them, and those are gone. Co-Authored-By: Codex XHigh <noreply@openai.com> --------- Co-authored-by: Codex XHigh <noreply@openai.com>
- A monitor keeps its identity while plugged in, so unplugging one of two identical monitors no longer hands its block to the other - Renumbering a workspace clears a stale return record left under its new id - The bar reads the block file from XDG_STATE_HOME like Hyprland writes it
…rameter out (omacom#15852) A plain KERNEL_CMDLINE[default]= in /etc/default/limine, which limine-entry-tool's own README suggests, replaces everything omarchy-defaults.conf appends, and a per-kernel key replaces the default. On such a machine the rebuild can never add systemd.tty.term.console=dumb, so the check after it aborted every update. When the rebuilt boot menu still lacks the parameter, the migration now asks limine-entry-tool for the command line of each entry limine-mkinitcpio builds: kernels whose modules.builtin a package owns, named after that package, plus their fallbacks when MKINITCPIO_FALLBACK, read through Limine's own config loader, builds them. It fails only when one of those entries is configured with the parameter, which means the rebuild itself went wrong. Asking after the rebuild rather than before means a misread config can at worst keep the migration pending, never skip a rebuild that would have added the parameter. Command lines, from the running kernel, the boot menu and limine-entry-tool alike, are split the way systemd, which reads this parameter, splits them: on spaces, tabs and newlines outside single or double quotes, with the last assignment of the key winning. A quoted value counts; text inside another parameter's value, or a value a later assignment replaces, does not. Co-authored-by: David Heinemeier Hansson <david@hey.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Codex Medium <noreply@openai.com>
* Keep a non-Latin vconsole.conf out of the initramfs again Since 26.134.222-3, Arch's plymouth hook copies /etc/vconsole.conf into the initramfs itself. The guard added for omacom#6229 only leaves the file out of FILES, so it no longer stops it: with a first XKB layout that types no Latin letters, Plymouth asks for the LUKS passphrase in that layout and the passphrase cannot be typed (omacom#14246). omarchy_vconsole.conf refuses the file in add_file unless FILES lists it, which is how it worked before. It is a drop-in of its own because pacman keeps omarchy_hooks.conf as configuration, so a machine whose copy was edited or predates the package never receives a change to it. Nothing rebuilds the boot image when a drop-in changes, so a migration rebuilds it once where the installed drop-ins now leave the file out and the plymouth hook copies it. * Read FILES when vconsole.conf is added, not when the guard loads The guard checked FILES once, when its drop-in was sourced, so a drop-in sorting after it that lists /etc/vconsole.conf was refused as well. It now checks FILES inside add_file: a hook runs with FILES local and empty, so its copy is still refused, and parse_config adds the configuration's own FILES once the hooks have run. The migration read only Omarchy's two drop-ins, so a mkinitcpio.conf or later drop-in listing the file got a rebuild that changed nothing. It now sources the main file and every drop-in as one, as mkinitcpio does. * Keep a Latin layout at the prompt when omarchy_hooks.conf predates the package An omarchy_hooks.conf written before the package (3.0 to 3.8.2) never lists vconsole.conf in FILES, so on such a machine Plymouth's own copy is what brings a Latin layout like German to the LUKS prompt since 26.134.222-3. Refusing it there switched the prompt to US and rejected a passphrase typed in German. Refuse the file only when the first layout is in the non-Latin list omarchy_hooks.conf already keeps out of FILES. --------- Co-authored-by: Emir Beganovic <beganovic.emir@gmail.com>
The failed-install check switches the default to Copilot, so the package launch check has to run while OpenClaw is still selected. Co-authored-by: Mark Groves <mark-groves@users.noreply.github.com>
|
Important Review skippedBot user detected. To trigger a single review, invoke the ⚙️ Run configuration
You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
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 from4e059acf(grammar fix already on the fork via #85) to60f5b496(2026-10-18 23:46 +0200). 32 upstream commits, applied on top of forkquattro. Cursor stays on the official tarball, not a mise tool. Face, Grok Bot, and Cursor menu entries stay. The power-profile menu provider stays out.Each monitor has its own workspaces (omacom#15655, omacom#15867)
Workspaces are per monitor now, including after an unplug: a workspace goes back to the monitor it belongs to instead of landing on whichever screen is left. Closing every window focuses that monitor's first workspace.
A language switch reaches apps launched from Hyprland (omacom#15791)
uwsm-appkeeps the launcher's environment, so after a language change, apps opened from bindings or the launcher were still on the loginLANG. The new language is set in Hyprland (and the session activation environment) and the shell is respawned from there.Reboot to Boot Menu is gone (omacom#15795)
The menu item is removed. Other languages still have the old catalog key until the translation sync drops it.
Default CLI tools are mise lazy shims (omacom#9596)
gh,uv, and the agent CLIs are declared in/etc/mise/conf.d/omarchy-tools.toml, which the settings package ships. mise builds native shims and installs each tool on first run. Agent CLIs check for an update at most every four hours. Handwritten wrappers are removed by migration1791382872. Remove Preinstalls opts a user out throughignored_config_paths. Needs mise 2026.10.11.Cursor is the exception kept on this fork:
cursor-agentis not in that toml. Choosing it still uses acursor-agentalready on PATH, or installs Cursor's official tarball. The migration still deletes an old mise wrapper and still leaves the official symlink alone.Pacman refresh keeps the current channel (omacom#15210)
Refreshing the pacman config no longer snaps the machine back to a different channel.
OpenClaw's move no longer stops every update
4.0.5 shipped the move to
~/.openclawbefore the seed package existed, and a missing package aborted the update (omacom#15809, then restored and replaced). Migration1790397381now finishes without changing anything. The move is1792334754: it runs only when theopenclawpackage is the seed (/usr/share/openclaw). If an OpenClaw of the user's own is in the way, the migration finishes and says so (omacom#15875); launching OpenClaw from Omarchy completes the move once that copy is out of the way. A failed setup stays pending and stays on screen. The agent key reaches OpenClaw's launcher when the package is installed even if the command is not on PATH yet.The detour that let any migration wait on another repository's package (omacom#14212) is reverted (omacom#15864).
omarchy-migrateis back to what it was before that.LUKS prompt stays typable on a non-Latin layout (omacom#14885)
Since plymouth 26.134.222-3, Arch's hook copies
/etc/vconsole.confinto the initramfs on its own, so the oldFILESguard no longer keeps a non-Latin layout out of the LUKS prompt. A new mkinitcpio drop-in refuses that file unlessFILESlists it, and migration1791621765rebuilds the boot image once where that changes the image.The console migration finishes when Limine replaces the command line (omacom#15852)
A
KERNEL_CMDLINE[default]=that replaces everything Omarchy appends can never gainsystemd.tty.term.console=dumb, and the check after the rebuild was aborting every update. The migration now fails only when an entry Limine actually builds was configured with the parameter and the rebuild still left it out.Reviews
Greptile is pointed at
AGENTS.mdand the task guides (omacom#15891).