Repository navigation
Sync upstream quattro through a9cedf93 - #99
Merged
Merged
Conversation
…macom#16233) * Clear the SoundWire routes the retired N1x UCM profiles left behind Omarchy's own MT8901 profiles, replaced by NVIDIA's in 3a20989, routed PDI2 to DP3 and PDI7/PDI8 to codec ports. alsactl saved those routes and restores them every boot, and NVIDIA's profiles only set the routes they use, so DP3 (ProArt P14 speakers) and DP7/DP13 (internal microphones) end up fed by two PDIs. The driver rejects the 2-channel streams with EINVAL, PipeWire drops the HiFi profile, and the card falls back to off: no internal speakers or microphones, with nothing to fall back to when a dock's microphone goes away. * Save the cleared N1x routes on a retry after a failed save A run whose sudo prompt was cancelled had already cleared the live routes, so the retry found nothing stale and the migration completed with the old routes still saved. Check asound.state too, and reprobe the card whenever the migration acts. * Keep the N1x route migration pending when the card can't be opened amixer reports "Invalid card number" for a card the user has no seat ACL on, as during an omarchy update over SSH with nobody logged in at the machine. The migration read that as no card and exited 0, so omarchy-migrate marked it done with the stale routes still saved. Test for the card under /proc/asound, which every user can see, and let amixer fail the run when the card is there but can't be opened. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Drop the N1x route migration's test AGENTS.md and agents/skills/migrations.md no longer allow a test for a migration (omacom#16302): it runs once per machine and is frozen once it ships. * Keep the N1x route migration pending when the saved state can't be read An unreadable asound.state read as clean, so a retry after a failed save could complete with the old routes still saved. A missing file or a missing control still reads as not stale. Co-Authored-By: Codex High (GPT-6.1-sol) <noreply@openai.com> --------- Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Codex High (GPT-6.1-sol) <noreply@openai.com>
* Accept the Lenovo N1x GPU memory reservation layout The Lenovo Pro 9 ships EXPOSED_DISP_FRM with a 0x40 marker and a 32 GiB reservation. Keep that marker, and still shrink the reservation to 4 GiB. * Say NVIDIA only described the XPS 16's EXPOSED_DISP_FRM layout The Lenovo's 0x40 byte is what its firmware ships and was checked to boot with 4 GiB; NVIDIA hasn't said what it means. Also put back the note that efivarfs files start with 4 attribute bytes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Co-Authored-By: Codex High <noreply@openai.com> * Drop the N1x unknown-marker test that passes without the change Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Ryan Hughes <ryan@heyoodle.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Codex High <noreply@openai.com>
) The user transition reset the default browser to Chromium unconditionally, after the retire step had deliberately kept a customized ~/.config/mimeapps.list. Fall back to Chromium only when no default browser resolves. Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Codex Medium <noreply@openai.com>
…y is unplugged (omacom#16310) * Turn a panel-disabled laptop display back on once its external display is unplugged * Leave a panel-disabled laptop display off while a headless output is enabled The DRM check sees only physical connectors, so with a headless output (a streaming display, say) as the only other output, the docked poll sent disabled = false every two seconds and undid the panel's choice. Re-enable only when nothing but FALLBACK is left enabled in Hyprland as well. The test now puts FALLBACK beside the unplugged panel, adds the headless case, and checks the docked case reached the update rather than passing on an empty log. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Co-Authored-By: Codex Medium <noreply@openai.com> Co-Authored-By: muse-spark-1.3-contributor <noreply@meta.com> --------- Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Codex Medium <noreply@openai.com> Co-authored-by: muse-spark-1.3-contributor <noreply@meta.com>
* Keep unresolved FIDO2 migrations pending * Handle FIDO2 recovery and inspection failures safely * Clear the notification log before checking the follow-up's retry text The linked-parent cases just before it leave a notification carrying the same retry text, so the assertion passed whether or not the follow-up raised its own. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Give the FIDO2 recheck migration its own timestamp omacom#16396 adds a different migration as 1792702724.sh. Completion markers are keyed by filename, so sharing the name would let whichever ran first mark the other done. The new name also sorts after the migrations already on quattro. --------- Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Ryan Hughes <ryan@heyoodle.com>
* Never overwrite a shell.json the shell could not read When shell.json fails to parse, or lacks version: 1, the shell runs on the shipped defaults, and every bar change persisted them over the user's file. Keep applying changes in memory but leave the file untouched until it loads again. Co-Authored-By: Codex Medium <noreply@openai.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Let the shell finish its startup writes before breaking shell.json The elsewhen widget saves its default zones into shell.json about half a second after the shell loads, so dropping the brace straight after the position check raced that write and lost on every tree. Wait for the file to settle first, and check that the blocked change still applies in memory, as the fix says it does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Check the unreadable shell.json guard in the runtime smoke test The separate test started a second shell to break shell.json, and on today's quattro Elsewhen's first-run save could land between the break and the shell's reload, writing valid JSON back before the guard ran. The runtime smoke test already runs a scratch shell with curl stubbed, and by the time it reaches its shell.json checks that save is long done. AGENTS.md asks for the area's existing test file over a new one. Co-authored-by: Codex High (GPT-6.1-sol) <noreply@openai.com> * Wait for Elsewhen's first-run save before breaking shell.json Co-authored-by: Codex High (GPT-6.1-sol) <noreply@openai.com> --------- Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com> Co-authored-by: Codex Medium <noreply@openai.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Ryan Hughes <ryan@heyoodle.com>
* Remove shell tests that only regex QML source * Drop Bar.qml source regexes from the bar notch test * Drop source greps from the click-clipping and button border tests * Drop Menu.qml and AppLibrary.qml regexes from the app search test * Drop clock widget source regexes and shipped preset lists from the clock test * Drop ImagePicker.qml source regexes from the image picker test * Drop source and package-list greps from the video background test * Drop Clipboard.qml regexes and the setpriv fixture from the clipboard test * Drop Bar.qml and shell.qml source regexes from the bar test * Drop Menu.qml regexes and shipped menu pins from the menu test * Stop asking to bump the menu test's font charset range The menu test no longer pins the icon font's coverage. * Drop QML and script source-text checks from shell tests Keep the model, VM, runtime fixture and command-output checks. The overlay no-animation check now covers every OverlayWindow plugin instead of a fixed list, and the dock default-pins check reads the shipped defaults instead of a copy. * Remove source-text tests of monitor, network, Limine, theme and launcher scripts * Trim source-text checks from the upgrade-to-quattro test * Trim Snapper test to the configure script run * Drop source greps from the Hyprland default config test * Drop restated titlebar config and QML source checks from the floating workspaces test * Drop source and template greps from the window transparency test * Drop restated bindings and QML source greps from the shell shortcuts test * Drop unit, skill, router and menu greps from the crash capture test * Drop default layout pins and source greps from the config test * Drop n1x.sh source greps from the N1x hardware test * Drop source greps and exact error copy from the Windows key test * Drop omarchy-update step greps covered by the update sequence test * Drop QML source greps from the power profile test * Drop trailing source greps from recorder, docker, profile, battery, browser, nightlight and Wi-Fi tests * Assert the display fields that matter instead of a JSON snapshot in the monitor state test * Drop all.sh and source greps from about, language, locale and platform setup tests * Drop all.sh, syntax and menu guard checks from the GB10 sleep test * Drop syntax, all.sh and source greps from the DGX Spark boot test * Drop the repeated hibernate case and drop-in grep from the fprintd resume hook test * Replace exact language counts in the i18n test with the registry size * Drop the result-by-result copy table from the fingerprint package test * Drop removed-marker and exact setsid line greps from first-run and desktop entry tests * Drop the menu when-string restatement from the Apple Silicon quirks test * Keep display sizes in the monitor state test's display check The monitor panel's scale picker reads width and height from that line, so dropping them from the comparison let the panel break without a failure. * Check the menu's glyphs against the Omarchy font's own charset The pinned e900-e91f range had to be bumped with every glyph. Reading the ranges from fc-query keeps the one link between the menu and the font without the pin. * Make the media-key fallback check match the quoted command the helper writes helpers.lua quotes the fallback's arguments (omarchy-shell 'media' 'playPause'), so the unquoted pattern could never match, and dropping a media shortcut passed. The notifications half goes: Control notifications with the keyboard falls back on purpose. * Keep running the Suspend and Hibernate menu guards on the GB10 fixtures This block ran the real guards on fixture machines, not source text. Only the string match on the guard goes. * Keep the lock, polkit and idle plugins loaded across plugin reloads A rescan recreates every service not kept loaded, which for these drops a lock, an authentication or the timer that locks the session. The check reads the manifests' JSON, not source text. * Allow the docker group grant only in the sudoless Docker setup The removed check only read the Quattro upgrade. Docker is root-equivalent, so the rule now covers everything that ships. * Run the Quattro upgrade's step block to check the order that keeps it bootable Each step logs its name and the block runs as written: the snapshot comes before the channel change, root= is pinned before the kernel install and checked after Limine is rewritten, and --reboot never reboots when root= was not confirmed. This replaces the line-order greps without reading the script's text. * Keep the lock's read-the-current-state check as the one source guard on locking It is the only guard on a fix for machines resuming unlocked, and the session lock cannot be driven offscreen to test it another way. The rest of the stranded-lock source checks stay removed. Co-authored-by: Mark Groves <mark-groves@users.noreply.github.com>
* Remove migration tests Migrations run once per machine and are frozen once they ship, so their tests stop guarding anything. Drop the migration test files and the migration blocks inside other tests, and update the testing docs to say migrations get no suite test. * Remove live-desktop pixel tests They read the developer's real screen with grim and skip whenever a window is open; one dispatches DPMS and another nests Hyprland and moves the real pointer. Checks like these belong in the acceptance VM. Drop the require_uncovered_background helper only they used. * Keep the lock auto-repeat test, which runs LockView's key filter rather than reading the screen It loads the real LockView offscreen and checks that held Return and Escape are dropped while Backspace and Delete still repeat. Nothing else covers that predicate.
* Remove migration tests Migrations run once per machine and are frozen once they ship, so their tests stop guarding anything. Drop the migration test files and the migration blocks inside other tests, and update the testing docs to say migrations get no suite test. * Merge the audio pactl tests into one file with a shared stub * Fold the Kimi usage test into its regression file and drop the wall-clock heartbeat wait * Collect the omarchy-shell IPC tests into shell-ipc-test.sh * Merge the keybindings menu tests and make the stale-cache check able to fail * Merge the dev-link and dev-unlink tests * Merge the mise node and python dev-env tests into one table * Fold the parallel menu guard tests into menu-guards and run the generated script instead of matching its text * Fold the migrate wrapper test into migrate-scope * Make the migrate-notify shared /tmp lock check able to fail and drop its copy greps * Fold the Apple mkinitcpio drop-in tests into mkinitcpio-hooks * Fold the factory reset Limine hash tests into factory-reset-limine * Fold the external monitor tests into hw-display * Merge the Hyprland session lock tests * Loop the hook and state bad-name matrix once * Fold the stale screenrecording audio tests in and drop source greps * Point the guard budget comment at the merged guards test * Fold the sudoers tests into one table-driven sudoers-test * Fold the per-quirk hardware tests into one table-driven hardware-quirks-test * Check Chromium extension ids against their native hosts in one loop * Share one Panel.qml staging harness between the network panel runtime tests * Name the merged sudoers test in the locale grant comment * Write the sudo stub's listing in one go omarchy-dns reads it through grep -q under pipefail; a second write after grep stopped reading died of SIGPIPE under load and sent the helper to polkit. * Hold each Chromium native host to the extension that messages it, and keep the shipped ids Set membership passed with the copy-url and yt-dlp grants swapped, and a regenerated key, which orphans an extension's settings, passed too. * Probe the root locale.gen with a code the packaged registry knows With only the unknown code zz, root exited before it read any locale.gen, so a root half that honoured OMARCHY_LOCALE_GEN still passed. * Check only the timezone rule's arguments against the matcher The dns, browser and locale rules are pinned whole above, so their rows only exercised the test's own matcher. The timezone rule is a regex, and its rows check what that regex admits. * Check the browser theme setter hands its writer bare hex The sudoers test that held this folded away, and the setter's own test only checked that the writer ran. * Drop the shell IPC trailing-newline check, which $() made unable to fail Command substitution strips trailing newlines whatever the client prints. * Check that n1x.sh does nothing on other machines Its next step drops every kernel but linux-omarchy, and the source grep that held its gate is gone. * Check every config and user setup step is reached, and platform setup runs last The per-step all.sh greps in the snapper, locale, language, GB10, DGX and platform tests are gone; one loop over the real dispatchers covers them all. * Let the sudoers test's stub sudo ignore SIGPIPE grep -q closes the pipe at its first match, and under the parallel runner's load the stub's printf still landed a write after that: 41 of 300 elevation probes fell back to pkexec. With SIGPIPE ignored the stub's exit status no longer depends on when the reader leaves; 0 of 300 under the same load. * Skip the root locale-gen probe when the test itself runs as root A user namespace maps root to real root, so locale-gen would rebuild this machine's locale archive from the scratch locale.gen. * Count only real references when checking a hardware leaf is reached nvidia.sh names hardware/n1x.sh in a comment and a message, so n1x.sh passed as reached with its all.sh line removed. * Accept either quote style in the native host sender lookup
) * Remove migration tests Migrations run once per machine and are frozen once they ship, so their tests stop guarding anything. Drop the migration test files and the migration blocks inside other tests, and update the testing docs to say migrations get no suite test. * Wait for the clipboard stub in the emoji insert test instead of racing it * Check the elsewhen zone table against itself, not the host's tzdata * Gate the stale-marker screenrecording case on the lock being held * Drive background intro waits from stub probes instead of sleeps and wall-clock bounds * Make OpenClaw onboarding wizard stubs outlast wrapper polls, not wall-clock sleeps * Wait for the lid watcher's syncs and assert it stays quiet once held, not exact call counts after a sleep * Replace sleep lock wall-clock bounds with observed behaviour and fail if the scaling sed misses * Skip rather than pass the OCaml sudo proof without namespaces, and stop racing its detached poller * Pin the Codex scanner test to UTC, refresh the cache before the reuse case, and share one opencode insert helper * Keep checking that the emoji insert helper copies plain text Splitting the exact wl-copy argument match into per-flag checks dropped the --type text/plain assertion, so a helper offering the emoji under another MIME type passed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Co-Authored-By: Codex Medium <noreply@openai.com> * Say what the signal case's fresh-attack wait actually failed to see wait_for_fresh_attack fails when the detached poller stopped trying sudo, but the message said it kept polling, which points whoever reads a red run at the opposite problem. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Hold the screenrecording lock from one process so a failed wait leaks nothing flock's bash child stayed blocked opening the release fifo after cleanup killed flock. A subshell that locks its own descriptor dies with the kill and frees the lock. --------- Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Codex Medium <noreply@openai.com>
* Remove tests that can't fail Relands omacom#16335, which merged into t3/prune-brittle-tests after that branch had already been squashed into quattro, so none of it landed. Since then quattro added real cases to three of these files: hibernation-setup and mise-provision keep theirs, and the clamshell cases from omacom#16310 move onto the compacted harness. * Keep errexit on inside the NVIDIA apps setup run The outer set +e carried into the setup's subshell, so a refused Dashboard start only failed the case while it was the setup's last line. * Keep the unit test for the acceptance notification parser input-methods-test.sh asserts that no input-method notification appeared, so a parser that stops recognising one passes there silently. The backup and restore cases stay removed: they guard a disposable VM. * Keep checking that every locate indexing path passes the service's updatedb options The install and AUR paths call updatedb directly with options that must match the scheduled service's drop-in. The checks on updatedb and plocate's own behaviour go. * Check the NVIDIA apps setup does nothing without its optional packages With every package stubbed present, removing either package guard passed, and a machine without Workbench or Dashboard would run their missing commands. * Keep the non-login shell locale test It runs default/bash/envs the way SSH sessions do and is the regression test for 6d7826d: without the LANG block those shells land in the C locale and print unicode escapes literally. Co-authored-by: Mark Groves <mark-groves@users.noreply.github.com>
* Remove migration tests Migrations run once per machine and are frozen once they ship, so their tests stop guarding anything. Drop the migration test files and the migration blocks inside other tests, and update the testing docs to say migrations get no suite test. * Remove live-desktop pixel tests They read the developer's real screen with grim and skip whenever a window is open; one dispatches DPMS and another nests Hyprland and moves the real pointer. Checks like these belong in the acceptance VM. Drop the require_uncovered_background helper only they used. * Share one QML fixture runner across the shell tests and fold the password overflow check into the fingerprint indicator fixture * Share one scratch-shell harness across the shell budget and smoke tests, and time login through the launcher in the bar startup budget * Document the shared shell test harnesses and the shell budget guards * Keep the lock auto-repeat test, which runs LockView's key filter rather than reading the screen It loads the real LockView offscreen and checks that held Return and Escape are dropped while Backspace and Delete still repeat. Nothing else covers that predicate. * Run the lock auto-repeat fixture through run_qml_fixture * Keep checking that default bar widgets show at a real size The section check alone passes a widget that renders hidden or at zero size. Whatever shows must have a size, and the menu, workspaces and clock must show. * Clear a QML fixture's result before running the next one A second run_qml_fixture in one file read the first fixture's result and passed without waiting for its own. * Keep runtime-smoke's collision listing from aborting the test With more than 20 collisions, head closed the pipe and pipefail killed the test before it reported the failure.
Drop N1x settings that linux-omarchy 7.3-3 now handles in the kernel
* Keep the clock still when right click cycles its format The bar's press feedback shrank the label for every mouse button, so a right click on the clock scaled the next format down as it landed and sprang it back. A widget can now say which buttons its mark gives under, and the clock leaves out right click, whose new label is feedback enough. The bar's press handler and an open panel's overlay report which button holds the widget, so the choice holds whichever of them takes the press. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Hold every button still down through an open panel's overlay The overlay told the widget only the button just pressed, so pressing right while left held the clock replaced the mask with right alone and the label sprang back while left was still down; releasing one of two buttons cleared the hold entirely. It now passes the full set of buttons down on press and keeps the hold while any remains on release. The bar's own press handler accepts only the left button, so it cannot chord. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Co-Authored-By: Codex Medium <noreply@openai.com> --------- 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>
* Bound wallpaper resource use when sampling bar colors * Test the configured-color fallback when a wallpaper sample fails or times out * Sample 6K and 8K wallpapers within the bar color memory bound The 256 MiB pixel-cache limit made many legitimate wallpapers fall back to the configured text color: Arch's ImageMagick is Q16-HDRI, so a 6K image alone needs about 230 MiB, and resizing to a 4K or 6K screen added a second full-size buffer. 31 of 92 stock theme backgrounds fell back at 1080p, 46 at 4K. Raise the limit to 512 MiB, let JPEG decode at about canvas size, sample on a canvas at most 1920 px wide, and point-sample to it instead of resizing. Only the bar strip's mean color is needed. All 92 stock backgrounds now make the same choice as unbounded sampling, 6K and 8K JPEG and PNG are sampled, and a 96 MP PNG still falls back cheaply. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Average wallpaper detail when scaling the bar color sample Point sampling to the canvas could land on one phase of fine stripes and pick the wrong text color. Box scaling averages every source pixel at about the same memory cost. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Make the stripe test fail when the wallpaper sample does not finish The stripe assertion expected the configured text color, which is also what the helper prints when ImageMagick gives up, so a sample that never ran passed it. Swap the colors so only a finished, averaged sample produces the expected one; point sampling and a failed sample both still fail it. Co-Authored-By: Codex Medium <noreply@openai.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com> Co-authored-by: Codex Medium <noreply@openai.com>
The clock is text to read rather than an icon to press, so shrinking it under left and middle clicks looked as wrong as it did under right. Its press feedback is now off for every button. Co-authored-by: David Heinemeier Hansson <david@hey.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
…ts behind an info mark (omacom#16795) * Show agents panel sections only for subscriptions, keeping stale limits with a quiet note An agent with only local history no longer gets an empty section; it still counts toward the summary line. Paused or unreachable limits stay on show with a dim note saying why, instead of a red error or a bare placeholder. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Turn the agents panel's stale-limits note into an info mark with a tooltip Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Show the agents info mark for failed refreshes that kept their limits Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Count history-only agents as set up in the agents panel Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Mark Groves <mark-groves@users.noreply.github.com>
…tate writes (omacom#8307) * Stop re-sampling the wallpaper for the transparent bar on unrelated state writes The bar watched ~/.local/state/omarchy/current for background changes. A directory path never sits in QFileSystemWatcher's file set, so Quickshell's FileView re-emitted fileChanged for every entry created, renamed or removed in ~/.local/state/omarchy as well. Any plugin writing there atomically (tmp + rename) re-ran omarchy-bar-text-color, and on a 5K display that ImageMagick pass cost ~2.6 core-seconds across every thread. One plugin saving every 10 s held ~40% of a core for the whole session, invisible in quickshell's own CPU figure because magick is a reaped grandchild. Watching the symlink directly does not work: inotify follows it to the old image and ln -nsf never touches that inode. Keep the directory watch, resolve the link with readlink -f and only refresh when the wallpaper actually moved. Sample at a reduced scale in omarchy-bar-text-color too. The strip is averaged to one pixel, so shrink the geometry by an integer factor (never upscale) and let the JPEG decoder skip the detail: 2.58 -> 0.065 core-seconds per sample on a 2912x1632 JPEG at 5120x2160, same result. * Resolve the background link at startup and cover the watcher in the runtime smoke test FileView reports no change for its initial load, so without a baseline the first unrelated write in the state directory still compared against "" and sampled once. Run the probe when the bar starts. The smoke test now stubs omarchy-bar-text-color and counts calls in the running shell: turning transparency on samples once, an atomic write of an unrelated file in ~/.local/state/omarchy does not sample again, and repointing current/background does. Against the previous Bar.qml the unrelated write registers as a second sample and the test fails. * Drop the reduced-scale helper tests, which omacom#14281 covers The merge took quattro's omarchy-bar-text-color, whose 1920px canvas, JPEG size hint and resource caps from omacom#14281 already bound the sample this branch shrank, so its tests for the old downscale go with it. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Watch inside current/ so unrelated state writes do not re-sample the bar FileView watches its path's parent directory too, and re-emits fileChanged for every entry created, renamed or removed there whenever the path itself is not a watched file. Watching ~/.local/state/omarchy/current made that parent ~/.local/state/omarchy, which plugins write to constantly, so the transparent bar re-ran ImageMagick on every unrelated save. Watching current/theme moves the parent to current/, where omarchy-theme-set swaps the theme and omarchy-theme-bg-set repoints the background link, and both still arrive. This replaces a readlink probe that resolved the link on every event: no extra process, and a theme switch that lands on a same-named background still re-samples. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Co-Authored-By: Codex Medium <noreply@openai.com> --------- Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Omarchybot <omabot@omarchy.org> Co-authored-by: Codex Medium <noreply@openai.com>
…macom#12540) * Restore bar-widget entries on enable when a plugins[] record lingers setEnabled keyed its bar-widget restore off findEntryLocation, which also matches a residual plugins[] entry left by a sibling panel/service kind. Disabling such a widget removed only its bar.layout entry; re-enabling then found the plugins[] record and skipped reinserting the widget, so the icon never came back (radio-atlas, anishfn.widgets). Key the restore off bar presence instead, and pin a regression in the plugin-registry contract test. * Drop a clone's plugins[] record when restoring its source Enabling a multi-kind clone that carries a plugins[] record now replaces its source in the bar, so disabling it can find the clone in both places. restoreCloneSource removed only the layout entry when it found one, and the surviving plugins[] record kept the clone active and routing its source's calls. It now clears both. Co-Authored-By: Codex Medium <noreply@openai.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Cover a plain widget's plugins[] entry, bar put and a stale bar.id The registry fixture only exercised the residual plugins[] record through a panel + bar-widget plugin. A plain bar widget reaches the same state when disabled, and omarchy bar put goes through the same guard, so these cases from omacom#15018 pin both: enabling a widget listed only in plugins[] places it and leaves its other settings alone, a disable and re-enable round trip brings it back, put places it with an explicit section and with none, and a bar.id naming a widget already on the bar does not insert it a second time. The no-section put expects centerEnd, which is where a widget without a default section lands since the bar sections were reworked. Co-Authored-By: Omarchybot <317366263+omarchybot@users.noreply.github.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Omarchybot <317366263+omarchybot@users.noreply.github.com> Co-authored-by: Codex Medium <noreply@openai.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The About box's OS Age is the age of /'s birth time. The 4.5 installer writes a root filesystem image built with the ISO and makes it @, so / is now born when the ISO was built: the age was too old by the ISO's age, or negative where the installed machine's clock ran behind the build host's. 4.5.0rc2 showed "-10 days" on a fresh install. On an install from today's quattro ISO, / was born at 13:48:21 (the image build) and /var/log at 14:27:35 (the install). /var/log is the @log subvolume both installers create during the install, archinstall's before 4.5 and the image's since, so OS Age now reads its birth time. On a system set up by hand it dates from pacstrap, as / does.
The power profile test still drives the busctl write, and a reset window in the agents panel still carries elapsed. Co-authored-by: Mark Groves <mark-groves@users.noreply.github.com>
|
Important Review skippedBot user detected. To trigger a single review, invoke the
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Upstream
quattromoved fromb8706a3f(26 Oct, already on this fork via #98) toa9cedf93(27 Oct, OS Age from the install omacom#16828). This replays all 22 first-parent commits, including merge omacom#16758. Two commits touched the fork overlay and were resolved so Face, Grok Bot, and Cursor stay.What changed
N1x and ProArt P14
1792634594clears the stale routes and stays pending when the card or the saved state cannot be read (Clear the SoundWire routes the retired N1x UCM profiles left behind omacom/omarchy#16233).EXPOSED_DISP_FRMmarker is0x40, and the reservation is still shrunk to 4 GiB (Accept the Lenovo Pro 9 N1x GPU memory reservation layout omacom/omarchy#16423).Quattro upgrade
Displays
Security
1793069744is the recheck (Keep unresolved FIDO2 migrations pending omacom/omarchy#16421).Shell settings
shell.jsonthe shell could not parse, or that is not version 1, is left on disk. Bar changes still apply in memory and are not written over the user's file (Never overwrite a shell.json the shell could not read omacom/omarchy#14534).plugins[]record is still there, so the icon comes back (Restore bar-widget entries on enable when a plugins[] record lingers omacom/omarchy#12540).Clock and bar
~/.local/state/omarchychanges (Stop re-sampling the wallpaper for the transparent bar on unrelated state writes omacom/omarchy#8307).Agents
About
/var/log, which is created during install. On a 4.5 image install,/is born when the ISO was built, so the About box was showing the image's age, including a negative age when the machine's clock was behind the build host (Count OS Age from the install, not the root image's build omacom/omarchy#16828).Tests
busctlwrite.Fork overlay
Face, Grok Bot, and Cursor stay.
FaceChromeCanvasstays immediately beforeFilterInput. Cursor stays the official tarball. The power-profile menu provider stays out and the write staysbusctl. English catalogs keep the face strings and still omitagents.helpStartToSeeLimitsand the power-profile menu strings. Fork delta vsa9cedf93is 220 paths.Topic draft #94 (
cursor/factory-reset-gum-argv-421a) was left alone. No other open sync.