v3.4.3 — Persistent Menu for Everyone + Uninstall Fix

AudioRouterNow — Release Notes

How to read this document

Each release contains two sections:

• For Everyone — Plain language. What changed, whether it affects you, and what (if anything) you need to do.
• For Power Users — Technical details: root causes, implementation notes, affected components. Skip this if you just want to know if you should update.

---

v3.4.3 — June 30, 2026

For Everyone

The persistent menu is now on by default — and the uninstaller no longer freezes.

This release makes the improved menu interface (which stays open while you work) the standard experience for all users, including those updating from an earlier version. It also fixes the uninstaller, which could freeze the app for 30+ seconds and sometimes fail to quit afterward.

• Persistent menu for everyone. The menu that stays open while you select devices, change sample rates, or explore settings is now the default for all users — including those who had already installed an earlier version. No configuration needed.
• Uninstaller no longer freezes. Clicking "Uninstall AudioRouterNow…" previously caused a long freeze (macOS spinning wheel, 30+ seconds) because the uninstall process ran on the app's main thread. It now runs in the background — the menu bar shows "Uninstalling…" and the app quits cleanly once done.
• Double-click protection on Uninstall. Since the menu stays open, it was possible to accidentally trigger the uninstaller twice. This is now prevented.

No action required. Update normally.

For Power Users

Fix	Component	Root Cause	Resolution
NSPopover as default	engine/config.py	use_popover_menu defaulted to False; new installs always received the legacy NSMenu.	Default flipped to True in both the AppConfig dataclass field and from_dict().
One-time migration	engine/config.py	Existing users had use_popover_menu: false already persisted in ~/.audiorouter/config.json; a default change alone wouldn't reach them.	popover_migrated flag added to AppConfig (default True for new installs; False sentinel for legacy configs). On first load of a pre-3.4.3 config (key absent → data.get("popover_migrated", False) = False), from_dict() forces use_popover_menu=True and sets popover_migrated=True. load_config() immediately persists via save_config() to survive force-quit/crash. After migration, a manual use_popover_menu: false is respected (no re-override).
Uninstall main-thread freeze	engine/menu_bar_app.py	_uninstall() called first_launch.uninstall_all() synchronously on the main thread. uninstall_all() contains time.sleep(2.0) + subprocess.run(osascript + killall coreaudiod, timeout=60) — together ~30 s of main-thread blocking → macOS spinning wheel, rumps.quit_application() not reached reliably.	_uninstall() now starts a daemon worker thread (arn-uninstall) that runs uninstall_all(). A rumps.Timer (0.3 s interval) in _uninstall_poll_result() collects the result on the main thread and calls quit_application(). os._exit(0) added as a 3 s hard-exit fallback (triggered only if terminate_() somehow doesn't propagate). killall coreaudiod shell command is unchanged.
Reentrancy guard	engine/menu_bar_app.py	NSPopover stays open after a click, making it possible to invoke _uninstall() a second time while the worker is already running — spawning a second thread and a second osascript admin prompt.	self._uninstalling flag checked at entry; set on entry, cleared on cancel/error in _uninstall_poll_result(). Success path quits the app, so no reset needed there.

Audit: All fixes passed dual Opus audit (@root-cause-analyst + @validator, parallel, both Opus) in a single iteration with zero critical and zero major issues. One cosmetic finding (getattr fallback) resolved in cleanup commit.

Commits: 55bf7f0 (NSPopover default), 60331cd (uninstall background thread), 6101587 (reentrancy guard + audit cleanup), 5d43f06 (one-time migration v3.4.3), 6689cfb (getattr simplification)

---

v3.4.2 — June 29, 2026

For Everyone

Audio reliability fixes plus a friendlier, more informative menu.

This release fixes two audio issues that could leave secondary outputs silent, and reworks the menu so it stays open while you work and explains its own status.

Audio fixes:

• All your outputs are now audible after routing. When AudioRouterNow takes over your system audio, it now carries your current volume level across to every device it routes to, instead of leaving them at whatever (sometimes silent) hardware level they were set to. Outputs you've deliberately set loud are left exactly as they are.
• No more brief drop-outs when adding a third output. On setups with three or more audio outputs (common on Mac mini), adding another output could cause the others to cut out for about ten seconds while macOS reconfigured its audio transport internally. AudioRouterNow now rides through that transition smoothly. Genuine audio failures are still recovered automatically.

Menu & status improvements:

• The menu stays open while you set things up. Previously the dropdown closed after every single click, so selecting several outputs or changing sample rates meant reopening it again and again. The menu now stays open as you work and closes when you click away.
• The menu bar icon no longer cries wolf. If one of your configured devices is unplugged but audio is still routing fine to the others, the icon now stays green instead of turning orange. It only turns orange when no outputs are available at all. The menu still notes "(N unavailable)" so you stay informed.
• New: Help → Status Guide. A built-in colour legend explains exactly what 🟢 green, 🟡 orange, and 🔴 red mean and when each appears — no need to guess or look it up.

No action required. Update normally.

For Power Users

Audio fixes

Fix	ID	Component	Root Cause	Resolution
W2-1	H7	engine/audio_device_control.py, engine/menu_bar_app.py	When "Audio Router" becomes the system default, set_default_output_volume() drives only the virtual device's volume_q16. Physical fan-out targets retain their frozen HW volume (often near zero) — silence despite a working IOProc. bogdanw confirmed: pre-selecting the internal speaker (already loud) → audible; U3277WB/Pebble V3 (frozen low) → silent.	get_device_volume_scalar() / set_device_volume_scalar() read/write per-device HW volume via kAudioHardwareServiceDeviceProperty_VirtualMainVolume (falling back to kAudioDevicePropertyVolumeScalar per channel). equalize_volume_after_switch(prev_level, target_uids) called at all three switch sites: reads previous default's level before switch, sets virtual device to that level after switch, raises any physical target below LOW_VOLUME_THRESHOLD (15%) to raise_to (or COMFORTABLE_FALLBACK_VOLUME = 50% if prev level itself was sub-15%). Devices without software volume control are skipped silently (get_device_volume_scalar returns -1.0). Devices ≥15% are left untouched (user intent preserved). All CoreAudio calls on main thread; CFStrings released.
W2-2	H8	engine/healer.py, engine/menu_bar_app.py	Adding BuiltInSpeakerDevice as a 3rd output to the AVAudioSession forces coreaudiod to restart the IOWorkLoop (HALC_ProxyIOContext:1593 "ending the transport"), transiently stalling already-running outputs. The healer's 1600 ms stall window (1000 ms soft-stall + 3×200 ms persist polls) detected this as a real failure and called reconnect_output — producing a visible ~10 s "Output removed/re-added" churn cycle.	notify_output_added() in Healer records time.monotonic() of each genuine output add (detected by set-diff in _apply_active_outputs). Grace-period check in _process_output suppresses reconnect_output for GRACE_PERIOD_S = 2.0 s after any add, provided ioproc_age_ms ≤ HARD_STALL_MS (5000 ms). Hard stalls (device truly gone) still heal immediately even during the window. State is thread-safe under the existing self._lock.

Audit: Both audio fixes passed two independent line-by-line audits (SuperClaude root-cause-analyst + Claude validator, both Opus) in a single iteration with zero critical and zero major issues.

UI / UX changes

Change	ID	Component	Detail
NSPopover menu	use_popover_menu	engine/menu_bar_app.py	The rumps/NSMenu dropdown is replaced by a persistent NSPopover toggled from the NSStatusItem, gated behind a use_popover_menu flag. The popover stays open across clicks (multi-device selection, sample-rate changes, Help navigation) and dismisses only on outside click (transient behavior). All callbacks and dynamic updates preserved; all AppKit calls guarded to the main thread.
NSPopover hardening	3 warnings	engine/menu_bar_app.py	Follow-up audit fixes: (1) status rows are clickable inside the popover — restart_helper / reinstall_driver / switch_audio dispatched via action_key; (2) the status line updates live while the popover is open (refresh() in _update_status_ui(), main-thread safe); (3) flicker guard on icon toggle — if the popover is shown, close only (no reopen), with a 0.15 s monotonic debounce against the popoverDidClose_ transient-dismiss race.
NSPopover checkbox fix	UX	engine/popover_menu.py, engine/menu_bar_app.py	All Row items defaulted to kind="item", which _viewForRow_ rendered as NSSwitchButton (checkbox). Pure action items (Quit, Status Guide, Open documentation, Check for Updates, Save Diagnostic Report, Uninstall, Support, System Audio toggle, footer) do not have a persistent checked state and looked wrong with checkboxes. Added kind="action" row type rendered as borderless NSButton (plain text, no checkbox). Toggle items (output devices, sample rates, safe mode) keep kind="item" and their checkbox.
Channel pairs always visible	UX	engine/menu_bar_app.py	Channel-pair sub-rows (Ch 1-2, Ch 3-4, Ch 5-6) for multi-channel devices previously only appeared after the device was activated, making the channel-selection feature invisible to users who hadn't yet tried a multi-channel device. Sub-rows now render at all times, greyed-out (enabled=False) when the parent device is inactive. Activating the device enables them. selected flag guarded with is_active and to prevent stale checkmarks.
Icon colour logic	UX	engine/menu_bar_app.py	Menu bar icon stays green when ≥1 output is active and routing works, even if a configured device is unavailable. Turns orange only at zero available outputs. Status text keeps the (N unavailable) counter for transparency. Fixes the misleading orange icon that alarmed users while audio was working fine.
Status Guide	UX	engine/menu_bar_app.py	New Help → Status Guide item opens a native NSAlert colour legend describing all three icon states (🟢 routing active, 🟡 warning, 🔴 error) with the specific scenarios that trigger each.

Build & assets

Change	Component	Detail
Single-source version	engine/version.py, installer/AudioRouterNow.spec, installer/build_local.sh, driver/resources/Info.plist	Four independent hardcoded version strings replaced by true single-source derivation from engine/version.py (APP_VERSION = "3.4.2"). The spec reads APP_VERSION via exec(); build_local.sh reads it via sed and adds a version-gate that fails on divergence; the driver Info.plist is set to 3.4.2. Found and fixed during the Wave-2 local build audit (2 audit iterations, both PASS).
Logo assets	assets/logo/	Brand logo set added: Inline and Stacked variants in Black and White, each as SVG + PNG.
Config merge-save	engine/config.py	save_config() previously overwrote the JSON file with only the fields known to AppConfig, silently dropping any keys added by newer app versions. If the older installed app saved config (e.g. after a sample-rate change), use_popover_menu and future flags were lost. Fix: read existing file first, spread unknown keys, then overwrite known fields — forward-compatible with all future flag additions.

Commits: 722ee69 (W2-1 + W2-2), 6b05285 (case analysis + fix plan), da2b5d8 (icon colour), 00e4084 (Status Guide), 767d147 (NSPopover), 390bc53/dda961c (NSPopover hardening), 3b57f14 (single-source version), 7901afc (NSPopover checkbox fix), db5db60 (config merge-save), 83ef0d1 (logo assets), 730e4a1 (channel pairs always visible)

---

v3.4.1 — June 25, 2026

For Everyone

Improved feedback, status display, and diagnostics.

This release addresses issues discovered from user feedback: the app's status indicator was showing "Routing active" even when no audio was actually routing, device names could silently disappear from the menu, and some error messages were shown in German regardless of system language.

• Accurate routing status — The status indicator now reflects whether audio is actually flowing, not just whether a device was selected. If routing fails, you'll see a clear error state instead of a misleading green icon.
• Missing devices are now visible — If a device you had selected is no longer available (e.g. unplugged), it now appears marked as unavailable in the menu instead of silently disappearing.
• Diagnostic report improved — The built-in diagnostic report now includes system audio state and fan-out status, making it easier to identify audio issues.
• English error messages — All user-visible error messages are now in English.

For Power Users

Fix	ID	Component	Root Cause	Resolution
H2	Status-UI	engine/menu_bar_app.py	Status string read _active_device_names (saved selection) not resp['active'] (real IOProc state). Showed "Routing active — 2 devices" even when fan-out was dead.	7-state matrix reading status['active'], is_audio_router_default(), and ioproc_calls progression.
H5	Stale-Config	engine/menu_bar_app.py	Devices missing from current HAL scan were silently skipped at menu_bar_app.py:1230.	Missing devices marked ⚠ unavailable in menu + N/M counter in status line.
i18n		audio_device_control.py, first_launch.py, diagnostic.py	User-facing error strings were hardcoded in German.	All user-facing strings translated to English.
Diagnostic	Fan-out	diagnostic.py	Diagnostic report had no visibility into whether fan-out IOProcs were actually running.	New SYSTEM AUDIO STATE and FAN-OUT sections added; reports active[], ring_frames, ioproc_calls.

Commits: 7521f0b, a7265bd, 7115a60, 68ec5ec

---

v3.4.0 — June 13, 2026

For Everyone

Critical fix: No audio after fresh installation.

After a fresh install from the DMG, AudioRouterNow appeared to run correctly (green icon, output device selected, system audio set to Audio Router) but produced no sound. This release fixes the root cause, and also prevents "zombie helper" processes (leftover old helper binaries from previous installations) from silently blocking audio on update.

• Audio now works after fresh installation — A macOS-specific limitation made it impossible for the audio driver to access the shared memory channel that carries audio between components. The driver was silently dropping all audio frames.
• Audio clock deadlock resolved — Even after the access issue, a circular dependency in the audio timing mechanism prevented audio from flowing. The clock now runs freely, as recommended by Apple's CoreAudio documentation.
• Zombie helper prevention — If an old helper process from a previous version is still running after an update, the app now detects it, shuts it down, and starts a fresh one automatically.
• No action required beyond updating. All fixes are automatic on first launch.

For Power Users

Fix	Component	Root Cause	Resolution
I-1	Helper (C)	fchown()/fchmod() on POSIX SHM FDs return EINVAL on macOS — SHM kept gid=staff/0660. Driver host (_coreaudiod) not in staff group → shm_open(O_RDWR) EACCES → gSHMRing=NULL → all frames dropped silently.	umask(0) + shm_open(O_CREAT, 0666). fchown/fchmod/getgrnam removed. Security preserved by magic/version/size ring integrity checks (C-1).
I-2	Driver (C)	GetZeroTimeStamp derived sampleTime from gFramesWritten. Circular deadlock: HAL won't call WriteMix until clock advances; clock only advances when WriteMix writes frames.	Freely running mach_absolute_time() clock from gAnchorHostTime (set at StartIO). Matches Apple NullAudio reference. P0-C ticksPerFrame fallback and H-4/H-5 timeline seed preserved.
I-3	Helper (C)	Makefile set ARN_HELPER_VERSION "3.2.0"; fallback #define was "3.1.2". Neither matched APP_VERSION = "3.3.1".	Both updated to "3.3.1".
I-4	Engine (Python)	_find_helper_binary() tried PyInstaller bundle before HAL path. After a sudo cp binary update the bundle still held the old binary → split-brain: old helper read stale SHM segment, new driver wrote to new segment → no audio.	HAL path (/Library/Audio/Plug-Ins/HAL/…) now has priority. Bundle is fallback only. HAL_HELPER constant introduced.
I-5	Engine (Python) / Helper (C)	No version negotiation between app and helper. Old helpers had no version field in get_status — running a stale helper after an update was undetectable.	ensure_running() calls _check_helper_version() before accepting a live helper. get_status now always includes "version": ARN_HELPER_VERSION (both ready/not-ready branches). Helpers below MIN_HELPER_VERSION = "3.3.0" are shut down and re-spawned. g_helper_version[] constant and #ifndef ARN_HELPER_VERSION fallback moved to compile-unit start so the build succeeds without -D flag (e.g. Driver Makefile path).
I-6	All components	First DMG build shipped all binaries still reporting 3.3.1 (Helper, App CFBundleShortVersionString) and 1.0.0 (Driver Info.plist). Version was never bumped before the build. Verified by post-install runtime audit (Fable agent, Jun 12).	Version bumped to 3.4.0 in all four canonical sources: helper/Makefile (-DARN_HELPER_VERSION), helper/AudioRouterNowHelper.c (both #ifndef fallback and legacy define), driver/resources/Info.plist (CFBundleShortVersionString + CFBundleVersion), installer/AudioRouterNow.spec (version, CFBundleVersion, CFBundleShortVersionString). I-5 zombie-detection now unambiguously distinguishes v3.3.x from v3.4.0 at runtime.

Post-Install Verification (Jun 12, 2026)

Runtime audit performed after fresh DMG install confirmed all fixes active:

Component	Verified
SHM permissions	0666 world-rw — I-1 confirmed
IOProc clock	33 k+ stable calls, ioproc_age_ms: 8, zero deadlock — I-2 confirmed
Helper spawn path	HAL path used, SHA-256 HAL = Bundle — I-4 confirmed
Version negotiation	Socket returns "version": "3.4.0" ≥ MIN — I-5 confirmed
Zombie processes	None (Z-status check clean)
Error logs	helper.err 0 bytes, helper.log clean
Audio routing	2 active routes (Komplete Audio 6 MK2 Ch1-2 + Ch3-4), underruns: 0, stalls: 0

---

v3.3.1 — June 11, 2026

For Everyone

• Better security: The audio data channel between AudioRouterNow's components is now strictly private — other user accounts on the same Mac can no longer access it.
• Stability: Fixed a theoretical race condition in the self-healing module that could have caused crashes under very specific timing conditions.
• No action required. Update normally.

For Power Users

Fix	Component	Description
H-1	Helper (C)	Removed dead coreaudiod_watchdog_tick() function — unreachable since F6 removed the call site in v3.3.0. Also removed find_coreaudiod_pid(), read_proc_cpu_ns(), outputs_stop_all_thread() and three unused includes (sys/sysctl.h, sys/proc_info.h, libproc.h).
H-2	Helper (C)	POSIX SHM now uses 0660 + gid _coreaudiod (resolved dynamically via getgrnam(), fallback 202) instead of gid 61 (localaccounts). Only the owner process and _coreaudiod can access the audio ring buffer.
H-4/H-5	Driver (C)	GetZeroTimeStamp outSeed now uses an atomic counter gTimelineSeed (incremented on StartIO and PerformDeviceConfigurationChange) instead of a hardcoded 1. Correct per macOS HAL spec for multi-device timeline sync.
H-6	Engine (Python)	Healer class now uses threading.Lock in all public methods. Fixes a race condition between process() (health-poll thread, 200 ms) and reset_all() (UI timer thread, 500 ms).

---

v3.3.0 — June 11, 2026

For Everyone

Freeze-Prevention & Security release — prevents system freezes, fixes an audio startup deadlock, and hardens the app against malformed shared memory.

After a hard reboot caused by the app freezing the entire system during teardown, a full root-cause analysis identified 15 issues. All are fixed.

What changed for you:

• System freeze on force-quit no longer possible — killing the app while audio was routing could cause the audio subsystem to spin at 100% CPU, freezing the entire machine. The audio helper now has a hard 5-second exit deadline with a guaranteed kill path.
• Audio starts reliably after launch — a logic error introduced in the previous patch caused the health monitor to silently never contact the Helper on startup, leaving all auto-recovery inactive from the first second.
• Menu bar never freezes when Helper is unresponsive — the 0.5s UI timer now reads from a background cache instead of making live system calls.
• Installation is safer on multi-user Macs — the driver install script now uses a unique temp file name (no predictable path an attacker could pre-create).
• SHM size validation — the audio driver now validates the shared memory segment size before mapping it. A malformed or too-small segment previously could cause a system-wide audio crash (coreaudiod SIGBUS). Now the driver safely skips invalid segments.

Who is affected: All users. Update by reinstalling from the new DMG.

---

For Power Users

Root Cause of the System Freeze (11. Juni 2026):

1. Python app killed → Helper orphaned (start_new_session=True)
2. SIGTERM → pthread_join hung: volume_poll_thread blocked in Mach-IPC to degraded coreaudiod
3. Additional kills had no effect (handler only set g_running=0)
4. sudo killall coreaudiod → HAL clients froze; new coreaudiod loaded plugin with device-clock race
5. GetZeroTimeStamp returned rate≈0 → coreaudiod RT-thread busy-spun → system freeze

Fix	Component	Root Cause	Resolution
F1	Helper (C)	No kill-path for stuck pthread_join	Double-SIGTERM → _exit(1); SIGALRM + alarm(5) as hard 5s backstop
F2	Helper (C)	outputs_stop_all() called before flag-file → Watchdog never fired	Flag-file → g_watchdog_tripped=1 → outputs_stop_all in detached pthread
F3	Driver (C)	GetZeroTimeStamp returned rate≈0 when no IO → coreaudiod RT busy-spin	Fallback to mach_absolute_time() when gFramesWritten=0; hybrid-clock clamp
F4	Python	Helper survived after SIGTERM timeout	SIGKILL escalation: terminate→2s→proc.kill()
F5	Helper (C)	Double-munmap race between signal handler and watchdog	atomic_exchange(&g_ring, NULL) before munmap
F6	Helper (C)	proc_pid_rusage() in volume_poll_thread could block → pthread_join hung	Entire coreaudiod CPU-poll section removed
F7/M1	Helper (C)	set_rt_priority() Mach-IPC call could hang	Function removed entirely
F8	Python	ping() on rumps main thread → 5s freeze when Helper unresponsive	Replaced with _cached_status(max_age=1.5)
F9	Python	Lock file open had TOCTOU race	os.open(O_RDWR|O_CREAT) + os.fdopen + seek/truncate
N6	Driver (C)	gDeviceIsRunning stuck at 0 after Zombie-Helper StopIO	Guard removed; gSHMRing != NULL is sole guard; self-heal on WriteMix
MC-5	Python	~/.audiorouter created with mode 0755; C-Helper requires 0700	_ensure_secure_base_dir(): mkdir(0o700) + os.chmod(0o700)
K1	Python	_health_poll_loop guard prevented get_status() when alive=False → permanent deadlock	Loop restructured: get_status() always called first
H1	Helper (C)	process_hotplug_removals / sr_reinit_all_outputs raced with in-flight output_add	Both skip active=false slots; local new_proc_id committed under lock
H3	Python	log_out/log_err FDs never closed after Popen()	try/finally closes parent FDs immediately after spawn
H4	Python	Healer + trip notifications not reset on Helper respawn	healer.reset_all() + _notified_trips.clear() on alive False→True
H5	Python	Shell-injection in install/uninstall paths	All path interpolations replaced with shlex.quote()
H6	Build	HELPER_DST hardcoded to Contents/Frameworks/ (wrong for PyInstaller 6.x)	find(1) locates binary dynamically
H7	Python	Install script at predictable /tmp/.arn_install.sh without O_EXCL	tempfile.mkstemp() — unique temp file, O_CREAT|O_EXCL internally
H8	Python	is_audio_router_default() (CoreAudio Mach-IPC) on 0.5s UI timer tick	Cached in health-poll thread (_router_is_default); timer reads cache
C1	Driver (C)	mmap without fstat size check → SIGBUS in coreaudiod if SHM too small	fstat guard at all 3 mmap sites; close+skip if st_size < ARN_SHM_SIZE
P12	Build	PyInstaller 6.x corrupts FAT→arm64 thinning (magic 0xCF→0xAA)	target_arch=None in spec; lipo -thin arm64 post-process guard

Commits: ecffc53 7e2d3a0 50166d4 b6c7228

---

v3.2.1 — June 11, 2026

For Everyone

Post-release hotfix — fixes installation failure and no-audio on first launch.

• First-launch "Helper Error" fixed — ~/.audiorouter permissions 0755 vs required 0700
• No audio on first use fixed — gDeviceIsRunning race after kill-9
• Installation wizard no longer hangs at 95% — PyInstaller FAT-binary corruption

---

For Power Users

Fix	Component	Root Cause	Resolution
MC-5	Python	~/.audiorouter created 0755 by umask; Helper requires 0700	_ensure_secure_base_dir()
N6	Driver	gDeviceIsRunning stuck at 0 after Zombie-StopIO	Guard removed, self-heal added
P12	Build	PyInstaller 6.20.0 corrupts arm64 magic byte	target_arch=None + lipo post-process

---

v3.2.0 — June 10, 2026

For Everyone

Stability & Security release — smoother hot-plug, tighter internals, and a cleaner codebase.

If you've ever noticed a half-second of silence on your primary audio output when another device was unplugged, this release eliminates that completely. All other changes are under the hood — the app should feel identical, but behave better under edge-case conditions.

What's new:

• Hot-plug silence eliminated — removing or adding any audio device no longer interrupts outputs that weren't involved. Previously, every device change could briefly restart all active IOProcs (≤85 ms silence). Now each output only restarts when it personally needs to.
• Pre-roll halved to 43 ms — the brief silent buffer that fills before a new device starts playing was mistakenly calculated as 85 ms due to a samples-vs-frames mix-up. Corrected to the intended 43 ms.
• No more stale audio on sudden clock drift — if a device's audio clock drifts sharply beyond tolerance, the engine now outputs clean silence instead of replaying old stale frames.
• Shared memory hardened — the inter-process memory segment that carries audio data is now created with tighter OS-level access permissions. The correct Unix group (localaccounts, which includes both your user account and the macOS audio subsystem) is set at creation time.
• Lag auto-recovery — if an output falls more than 90% of the ring buffer behind (e.g. after a system sleep or extreme CPU load), it now automatically re-syncs instead of staying stuck behind.
• Numerous small fixes — mute toggle correctness, sample-rate selection, circuit breaker logic, health check startup grace period, and more (see Power Users section for full list).

Who is affected: All users benefit. No action required.

---

For Power Users

Architecture: Tombstoning replaces Swap-Remove

The most significant change is in output_remove_locked() in the C Helper. Previously, removing a device output used swap-remove — the last slot was moved into the freed slot. This forced all moved IOProcs to be destroyed and re-registered with their new pointer, causing every uninvolved output to re-arm its pre-roll (85 ms silence). The new approach marks the freed slot as a tombstone (uid[0] = '\0'), leaving all other slots in-place. IOProcs for uninvolved outputs never see any interruption.

Key counters: g_n_outputs is now a monotone high-water-mark; g_n_active_outputs tracks the actual live count. This eliminates bugs KC-1, HC-1, HC-2, and ARC-2 entirely.

Full list of fixes applied in this release:

Fix	Component	Summary
KC-1	Helper	Swap-remove forced moved-IOProc restart → eliminated by tombstoning
KC-2	Helper	g_deferred_unmap_timestamp_ns: 150 ms grace before munmap prevents SIGBUS when IOProc holds old g_ring pointer
KC-3	Helper	frac_ridx_reset_pending (bool) → frac_ridx_reset_gen (uint32 generation counter) — eliminates lost-update race between volume thread and IOProc
HC-1	Helper	Tombstoning: IOProcs for uninvolved outputs never restart on device change
HC-2	Helper	src_frac_ridx of moved output no longer force-reset (no more move = no more reset)
HC-4	Helper	Generation counter (KC-3) closes HC-4 race variant — same fix, two races addressed
H1	Python	Mute toggle: set_muted(current > 0.0) → set_muted(not get_default_output_muted()) — reads actual mute state instead of volume proxy
H3	Python	Sample-rate selection: picks the rate with maximum device coverage instead of hardcoded 48 kHz fallback
H4	Python	Drift persistence: 3 consecutive polls required before persisting (grace period eliminates transient drift false-positives)
M1	Python	Circuit breaker: failures counter only incremented on actual failure (was also incrementing on success); eviction after 2 consecutive misses
M4	Python	Media-key volume changes use a worker thread + queue — no more per-keypress thread spawns
M5	Python	Health monitor: 3-iteration startup grace period before emitting "critical" state
M6	Python	Trip notifications use breaker_name(uid, ch_offset) API instead of searching sh.outputs
M7	Helper	shutdown() — proc.wait()/terminate()/kill() moved outside lock; only socket send runs under lock
ARC-2	Helper	Eliminated by tombstoning (no more output moves → no more force src_frac_ridx reset)
ARC-3	Python	New device debounce: reported only after 2 consecutive scans (~4 s); removals still immediate
ARC-5	Helper	Lag-eviction: output > 90% ring capacity behind → forced re-sync via generation mechanism
N2	Python/C	Unused imports removed across healer.py, health.py, helper_client.py, diagnostic.py
N3	Python	_notified_trips: set initialized in __init__ (was lazy hasattr init)
N4	Python	device_manager.refresh() now fires on_devices_changed callback when changes detected
N5	Python	_format_sample_rate() extracted as @staticmethod — deduplicates kHz formatting across menu
Batch 6	Helper	write_full() partial-write loop; parse_int_strict() replaces all atoi(); sigaction with SA_RESTART; shm_connect() validates capacity+channels
Batch 7	Helper	SHM: shm_open(…, 0660) + fchmod(0660) + fchown(fd, -1, 61) — gid 61 = localaccounts group (contains both user uid 501 and _coreaudiod uid 202)
Batch 8	Helper	active_buf 4096→16384, resp 8192→24576; JITTER_TOLERANCE path outputs silence not stale data; non-interleaved nFrames uses mNumberChannels; g_outputs_generation counter for watchdog-race detection
Pre-roll	Helper	ARN_PREROLL_FRAMES = ARN_RING_CAPACITY / 8u (2048 frames = 43 ms @ 48 kHz) — was /4u = 4096 frames = 85 ms due to samples-vs-frames confusion
Infra	Python	engine/version.py — single source of truth for version number; all modules import from here

Commits: 3206ee3 (M-fixes nachtrag) — followed by version bump commit

Full technical documentation: DOKUMENTATION.md, Kapitel 46.

---

v3.1.2 — June 9, 2026

For Everyone

New: Send a diagnostic report to the developer with one click.

If you're experiencing an issue — audio not routing, unexpected behaviour, anything unexpected — you can now send a detailed report directly from the app. No manual log hunting, no copy-pasting.

What's new:

• Help → Save Diagnostic Report… — generates a structured text file with your system info, current audio status, and recent log events. Mail.app opens automatically with the file already attached and the email pre-filled. Add a short description of your problem and click Send.
• If Mail.app isn't available, the file is saved to your Desktop and revealed in Finder, with a notification showing where to send it.

Who is affected: Everyone who ever wanted to report a bug but wasn't sure how to attach the logs.

What you need to do: Nothing — the menu item appears in Help automatically.

---

For Power Users

New module: engine/diagnostic.py

• Collects: platform.mac_ver(), sysctl -n hw.model, CPU arch, Helper status (get_status_quick()), last 1 MB of helper.err, last 3 MB of helper.log (event tokens only — polling blocks removed via regex)
• Log extraction strategy: poll blocks (Ring: N Frames | Outputs: N | IOProc-Calls: …) removed via re.sub, then event tokens extracted by known prefixes (Helper:, AudioRouterNow Helper, Warte auf SHM, SHM:, Helper laeuft). Simpler and more reliable than a single event regex with lookahead + DOTALL.
• generate_report(helper_client) → Path saves ~/Desktop/AudioRouterNow_DiagReport_{timestamp}.txt
• open_mail_with_report(path) → bool opens Mail.app via AppleScript (timeout: 20 s). Escapes backslashes and double-quotes in path.
• reveal_in_finder(path) — fallback via open -R

engine/menu_bar_app.py changes:

• import diagnostic added
• Help menu: new "Save Diagnostic Report…" item (separator before Uninstall)
• _save_diagnostic_report() starts a daemon thread → main thread never blocked (sysctl + 3 MB read + osascript ≤ 23 s worst-case)
• All outcome notifications via rumps.notification (thread-safe) rather than rumps.alert

Commits: 317f531 (feature) · Branch: main

Full technical documentation: DOKUMENTATION.md, Kapitel 45.

---

v3.1.1 — June 9, 2026

For Everyone

Fixes a rare audio glitch that occurred every ~12.5 hours of continuous play.

If you ever heard a brief (2–3 second) high-pitched constant tone followed by normal audio resuming, this was it. It happened on a fixed timer — roughly every 12 hours and 26 minutes, regardless of what was playing. The glitch was most noticeable during live streams (like WWDC) where the audio demands attention.

What's new:

• 12.5-hour periodic stall eliminated — the root cause was a floating-point counter in the audio engine that was never reset. After ~12 hours of continuous play, the counter exceeded the valid range for a 32-bit integer conversion, causing undefined behavior and a momentary position reset in the audio pipeline.
• No other changes — this is a pure bugfix with no behavioral changes under normal conditions.

Who is affected: Anyone who uses AudioRouterNow for extended continuous listening sessions (12+ hours). The fix is invisible during normal operation.

What you need to do: Install the update if you experienced the periodic tone glitch.

---

For Power Users

Root cause: src_frac_ridx (a double field in DeviceOutput) accumulates += ratio (≈1.0) for every output frame in device_ioproc(). At 47,872 frames/s (512 frames × 93.5 calls/s @ 48 kHz), after 2^31 / 47,872 ≈ 44,739 seconds ≈ 12h 26min, src_frac_ridx * 2.0 exceeds UINT32_MAX = 4,294,967,295. The cast (uint32_t)(src_frac_ridx * 2.0) is then Undefined Behavior under C11 §6.3.1.4 — on ARM64, this typically produces 0, making behind = widx - 0 = widx >> ARN_RING_CAPACITY, which fires the Overflow-Guard and issues a Pending-Reset.

Log evidence: 16 HARD-STALL events in helper.err, appearing in pairs exactly every 8,388,486–8,389,016 IOProc-calls (expected: 2^32/512 = 8,388,608; measured deviation ±0.005%). Fully deterministic, source-independent.

Pair structure: The stall fires twice because the Overflow-Guard resets src_frac_ridx = widx/2. At the overflow moment, widx ≈ 2^32, so widx/2 ≈ 2^31 — immediately triggering UB again on the next IOProc calls. The second stall (376 calls ≈ 2s later) occurs when widx finally wraps to a small uint32_t value, making widx/2 small and safe.

Fix (3 lines, commit ff7556e, helper/AudioRouterNowHelper.c):

dev->src_frac_ridx += ratio;   // line 896 (existing)

// P16: Fold to prevent float→uint32_t UB after ~12h
if (dev->src_frac_ridx >= (double)(1u << 31)) {
    dev->src_frac_ridx -= (double)(1u << 31);
}

Why the fold is safe: 2^31 is a multiple of ARN_RING_CAPACITY = 8192 = 2^13, so all three downstream invariants are preserved:

• frac_as_samp = (uint32_t)(ridx*2): fold changes value by 2^32 ≡ 0 (mod 2^32) → behind unchanged
• Ring index (idx0*2) & RING_MASK: (2^31 * 2) mod (2 × 8192) = 2^32 mod 16384 = 0 → physical position unchanged
• Interpolation frac = ridx - idx0: integer fold, fractional part invariant

All invariants verified with full Python simulation: all MATCH=True.

Full technical documentation: DOKUMENTATION.md, Kapitel 44.

---

v2.9.0 — June 1, 2026

For Everyone

AudioRouterNow now monitors itself and recovers from audio problems automatically.

If your audio ever crackles, drops out, or a device freezes — the app now detects this within seconds and tries to fix it on its own, without you having to restart anything.

What's new:

• Health indicator in the menu bar icon: 🟢 all good, 🟡 something looks off, 🔴 a problem was detected
• Automatic recovery: if a device stalls, the app reconnects it automatically (up to 5 attempts, with increasing wait times between each try)
• Smarter audio start: a brief 43ms buffer fills before audio plays to prevent the "stuttering first second" on new devices
• Safe mode: a new menu option Safe mode (no auto-healing) — when activated, the app only monitors but never touches anything. Recommended for live recordings or concerts where any interruption is worse than the original problem
• Better long-term stability: an improved audio clock drift compensator (PI controller) smooths out tiny timing differences between devices over time — completely inaudible

Who is affected: All users benefit. No action required. The Safe mode toggle is in the menu if you need it.

---

For Power Users

Complete implementation of the Self-Healing Layer (3 Tranches). All changes pass an independent Opus validator audit. ABI remains v4. RT thread (device_ioproc) has only one new line of code.

Tranche A — Telemetry + Health Indicator

New atomic counters in DeviceOutput: recovery_count (stall→healthy transitions), g_reconnect_count (SHM reconnects), g_last_ioproc_call_ns (timestamp of last real IOProc call).

get_status JSON extended with reconnect_count, ioproc_age_ms (ms since last device_ioproc call; 9999 if none yet), recovery_count per output, fill_ewma per output (Tranche C).

engine/health.py — new module: OutputHealth, SystemHealth, HealthMonitor. Three-level classification with hysteresis:

• Degradation: 2 consecutive samples (400ms)
• Recovery: 5 consecutive samples (1s)
• Critical triggers: ioproc_age_ms > 500 (when audio flowing), stalled=1, new reconnect
• Degraded triggers: new underruns, SRC drift > 350 ppm, ring fill < 10% or > 95%

menu_bar_app.py: new health-poll daemon thread (200ms), icon ampel integrated in _compute_status() final branch.

Tranche B — Soft Out-of-RT Healing

Pre-Roll High-Water-Mark (device_ioproc, RT-safe):

if (atomic_load(&dev->preroll_armed, relaxed)) {
    if (behind_p / 2u < hwm) { /* output silence, don't move position */ return noErr; }
    atomic_store(&dev->preroll_armed, 0u, release);  // self-clear at HWM
}

Default HWM: ARN_RING_CAPACITY/4 = 4096 frames ≈ 43ms @48kHz. Re-armed on every output_add, SHM-reconnect, and slot-swap IOProc restart.

reconnect_output socket command: output_remove_locked under lock → output_add outside lock (3-phase design, same as H1). Guards: g_safe_take (→ safe_take error), g_shm_ready=0 (→ shm_reconnecting error).

engine/healer.py — new module:

• STALL_PERSIST_SAMPLES = 3 (600ms) before first attempt
• Backoff: [0.5, 1.0, 2.0, 4.0, 8.0] seconds
• MAX_ATTEMPTS = 5 → tripped=True → rumps.notification
• Breaker fully resets on output recovery
• Python-side safe_take_getter() check (double guard)

Safe-Take-Modus: g_safe_take atomic_int global in Helper. set_safe_take socket command. config.safe_take_mode: bool (persisted). App syncs state on start. Double guard: C (g_safe_take) + Python (Healer.process()).

Tranche C — PI Controller + EWMA SRC

Extends the existing P controller on src_ratio_q20 (volume_poll_thread, 50ms, non-RT) with an I term and EWMA smoothing:

fill_ewma = 0.1 × fill_frames + 0.9 × fill_ewma         (τ ≈ 500ms)
error     = fill_ewma − target_frames
correction = Kp×error + Ki×Σ(error×dt)
           where Ki=0.0005, dt=0.05s, I clamped to ±300ppm
src_ratio_q20 = (base_ratio + correction) × 2²⁰          (total ±500ppm)

Non-atomic fill_ewma and integ_error fields in DeviceOutput — single writer (volume_poll_thread under g_outputs_lock). get_status reads under same lock. Zero RT-path exposure.

State resets at all discontinuities: stall-set, stall-recovery, SHM-reconnect, SR-change (both branches of sr_reinit_all_outputs).

Defensive clamp if (ratio_f < 0.0f) ratio_f = 0.0f before uint32_t cast (prevents UB under pathological base_ratio).

All 6 bugprone-integer-division clang-tidy warnings resolved ((double)(u/u) → (double)u / 2.0).

Commits: 628b719 → fd3d0a5 → f87dfa4 → 8283ffd → 481c33c → c904a62 → 301adcb (merge) → docs/version bump
Full technical documentation: DOKUMENTATION.md, Kapitel 28.

---

v2.8.1 — June 1, 2026

For Everyone

Fixes crackling/clicking audio when routing to multiple channel pairs on the same device.

If you were using AudioRouterNow with a multi-channel interface — for example, routing to channels 1–2 and channels 3–4 of the same audio interface simultaneously — you may have heard noticeable crackling or clicking. This version fixes that completely.

Who is affected: Anyone using two or more channel pairs on the same physical device (e.g. a Komplete Audio 6, Focusrite Scarlett with 6+ channels, or any other multi-channel USB interface).

What you need to do: Nothing. Install the update and the problem goes away. No settings to change.

Upgrade path: Replace your existing AudioRouterNow.app with the new one from the DMG, or let the app update itself on next launch.

---

For Power Users

Two independent bugs combined to cause the crackling, both introduced or exposed by the v2.8.0 multi-output rework.

Bug 1 — Slot-Swap Invalidated IOProc inClientData Pointer

File: helper/AudioRouterNowHelper.c — output_remove_locked()

When an output is removed, the function fills the vacated slot by moving the last element of g_outputs[] into it (g_outputs[slot] = g_outputs[n-1]). However, the moved output's IOProc was registered with inClientData = &g_outputs[n-1] — the old address. After the move, g_outputs[n-1] is zeroed with memset.

On the next CoreAudio callback, the moved IOProc read from a fully zeroed DeviceOutput struct:

• src_ratio_q20 = 0 → ratio = 0.0
• src_frac_ridx = 0.0 → no advancement per call
• local_ridx stays at 0 → K2 Stall Detection fires after 1000 ms
• underruns counter = 0 (no underrun path hit — the behind + JITTER_TOLERANCE < needed check with needed = 0 never triggers)

Once stalled, the output was excluded from update_global_read_idx(), causing the ring to back up. The Producer (HAL driver) dropped frames. The healthy output on channels 1–2 received silence-corrupted frames → audible crackling.

Fix: Before swapping, stop and destroy the moved output's IOProc. After copying the struct to its new slot (&g_outputs[slot]), re-register it via AudioDeviceCreateIOProcID with the now-stable heap address, restart it, and issue a Pending-Reset (frac_ridx_reset_pending) so the IOProc initializes its read position correctly.

// Before: moved IOProc silently reads from zeroed memory
g_outputs[slot] = g_outputs[g_n_outputs - 1];

// After: stop → copy → re-register at new stable address → restart
AudioDeviceStop(moved_src->dev_id, moved_src->proc_id);
AudioDeviceDestroyIOProcID(moved_src->dev_id, moved_src->proc_id);
g_outputs[slot] = g_outputs[g_n_outputs - 1];
AudioDeviceCreateIOProcID(moved->dev_id, device_ioproc, moved, &moved->proc_id);
AudioDeviceStart(moved->dev_id, moved->proc_id);
// + Pending-Reset to write_idx

Bug 2 — SRC Boundary Instability at Sample-Rate Mismatch

File: helper/AudioRouterNowHelper.c — device_ioproc()

When the output device runs at a different sample rate than the ring (e.g. KA6 at 44100 Hz, ring at 48000 Hz), ratio = 48000 / 44100 = 1.0884. The IOProc computes:

needed_samples = floor(512 × 1.0884 × 2) = floor(1114.57) = 1114

Per KA6 IOProc cycle, the Producer (running at 48000 Hz) writes approximately 1114.5 samples into the ring. Due to integer truncation and hardware timing jitter:

• Sometimes behind = 1113 < needed = 1114 → Underrun (local_ridx not updated)
• Sometimes behind = 1115 ≥ needed = 1114 → Normal

Alternating underruns meant local_ridx effectively stalled → K2 Stall Detection would fire → Pending-Reset set local_ridx = write_idx, giving behind = 0 immediately → perpetual underrun → stall loop every 1000 ms.

Fix — 4-sample jitter tolerance:

// Before — exact boundary, no tolerance:
if (behind < needed_samples) { /* underrun */ }

// After — 4-sample (2 stereo-frame) tolerance:
const uint32_t JITTER_TOLERANCE = 4u;
if (behind + JITTER_TOLERANCE < needed_samples) { /* underrun */ }

At the exact boundary (behind = 1110, needed = 1114): 1110 + 4 = 1114 ≥ 1114 → accepted. The 4 missing samples at the tail are filled with silence in memset(temp_buf, 0, ...) — at 44100 Hz this represents ≈ 0.09 ms, completely inaudible.

Stall timeout also increased from 300 ms to 1000 ms to give the SRC P-controller more settle time after a rate-mismatch reconfiguration.

Combined Failure Sequence

1. App configures: KA6 Ch 1–2 (slot 0), BenQ Ch 1–2 (slot 1), KA6 Ch 3–4 (slot 2)
2. BenQ is removed → slot swap: KA6 Ch 3–4 moves from slot 2 → slot 0
3. Bug 1: IOProc for KA6 Ch 3–4 reads from zeroed slot 2 (ratio=0, no progress)
4. K2 Stall fires → output excluded from read_idx aggregation
5. Ring backs up → Producer drops frames → crackling on Ch 1–2
6. Bug 2 (independently): At 44100 Hz / 48000 Hz boundary, alternating underruns cause secondary stall loops even after Bug 1 is fixed alone

Commit: 651b9fb

---

v2.8.0 — May 31, 2026

For Everyone

Major stability and reliability update. This release completes a full security and correctness audit of all components. Most changes are under the hood — you should notice cleaner behavior on device hotplug, smoother volume control, and fewer edge-case audio dropouts.

Highlights:

• Volume keys no longer cause the menu to freeze briefly
• Plugging/unplugging devices is handled more reliably
• The app's config file is now written safely (a crash during save no longer corrupts your settings)
• A second accidental Helper process can no longer start in the background and fight with the first

Who is affected: All users benefit. No action required.

---

For Power Users

Complete implementation of all CRITICAL, HIGH, and MEDIUM findings from the v2.7 deep audit. 12 fixes across 7 commits. Risk score after v2.8.0: CRITICAL 0, HIGH 0, MEDIUM 0.

Fix	Component	Summary
K1+K2	Helper	Stall-Detection: frozen local_ridx excluded from read_idx aggregation; stalled output reset to write_idx via Pending-Reset
H1	Helper	output_add() 3-phase: USB SR-settle (400 ms) now lock-free; only Create+Start run under lock (<20 ms hold)
H2	Helper	g_ring → _Atomic(ARNSharedRing*) + deferred munmap on live-reconnect (SIGBUS prevention)
H3	Helper	Hot-plug listener sets atomic flag only; actual O(N) scan runs in volume-poll thread
H6	Helper	json_escape_into() — RFC 8259-compliant escaping for UID and name in get_status
H7	Helper	Config socket moved from /tmp to ~/.audiorouter/; umask(0177) before bind() eliminates TOCTOU window
H8	Python	Volume polling + media-key osascript calls moved to daemon threads; rumps event loop never blocked
M1	Ring	arn_ring_frames_available: read_idx loaded with acquire before write_idx
M2	Helper	g_hotplug_registered: volatile int → atomic_int
M3	Helper	Socket listen backlog: 4 → 16
M4	Helper	find_device_by_uid: NULL-uid short-circuit skips unnecessary device_output_channels() call
M6	Helper	ch_offset validation: must be even, max_ch ≥ 2, ch_offset + 2 ≤ max_ch
M7	Helper	SRC: 3-tap box-average pre-filter when ratio > 1.005 (downsampling) to reduce aliasing
M8	Helper	Single-instance lock via flock(/tmp/audiorouter.helper.lock)

Full technical documentation: DOKUMENTATION.md, Kapitel 24.

---

v2.7.0 — May 31, 2026

For Everyone

Critical bug fix release. Fixes audio silence after the Helper process restarts, and resolves rare audio timing glitches.

• Helper restart (e.g. after coreaudiod reload) now reliably reconnects without silence
• Volume control timing is more accurate
• App settings survive a crash during save

---

For Power Users

8 fixes from the first audit pass, all targeting RT-correctness and thread-safety:

Fix	Component	Summary
K3	Driver + Ring	instance_id field in SHM header (ABI v4); Watch-Thread compares instance IDs instead of inodes — macOS recycles POSIX SHM inodes, making inode comparison unreliable after Helper restart
K5	Driver	gAnchorHostTime: UInt64 → atomic_ullong; eliminates data race between StartIO (write under gStateMutex) and GetZeroTimeStamp (RT-thread, no lock)
K6	Helper	src_frac_ridx data race: Pending-Reset pattern — only the IOProc writes the field directly; Volume-thread signals via frac_ridx_reset_pending + frac_ridx_reset_widx (both atomic)
K7	Helper	nFrames clamped to ARN_RING_CAPACITY/2 before SRC loop — prevents BSS overflow in temp_buf[16384] at large buffer sizes
H4	Driver	pthread_join (in arn_shm_cleanup) moved outside gStateMutex in ARN_Release — eliminates latent deadlock
H5	Ring	arn_ring_set_sample_rate: also resets read_idx = 0 (seq_cst) — previously only write_idx was reset, causing uint32 underflow → Producer blocked
M5	Helper	base_ratio sanity check (> 0, < 10) in output_add and sr_reinit_all_outputs — prevents NaN/Inf when device SR query fails
M9	Python	save_config(): write to .tmp + fsync + atomic Path.replace() — crash during write no longer corrupts config.json

Full technical documentation: DOKUMENTATION.md, Kapitel 23.

---

v2.6.0 — May 31, 2026

For Everyone

Fixes app freezing for several minutes on restart, and eliminates background Helper processes that kept running after you quit the app (causing fan noise and CPU usage).

---

For Power Users

• Keep-Alive IOProc migrated from Python ctypes callback to C Helper: eliminates stale function pointer crash on app restart (mach_msg2_trap deadlock in HALSystem::InitializeDevices)
• _quit_app() now calls helper.shutdown() before exit — Helper no longer orphans after app quit
• Python stub ensure_router_keepalive() kept for API compatibility

---

v2.5.0 — May 2026

For Everyone

Fixes audio not working on macOS 26 (Tahoe) unless you manually toggled the output device in System Settings.

---

For Power Users

• GetZeroTimeStamp: pre-StartIO fallback (anchor = now when gAnchorHostTime == 0) prevents coreaudiod from classifying the virtual device as "in the future"
• Persistent Keep-Alive IOProc registered on virtual device; AudioDeviceStart called directly via ctypes (later migrated to C in v2.6)

---

v2.1 – v2.4 — May 2026

For Everyone

Series of stability fixes establishing the current architecture:

• v2.4: macOS 26 compatibility (audio now works without manual device toggle)
• v2.3: Volume keys work reliably; audio recovers correctly after device changes
• v2.2: First-run wizard, status display, one-click uninstall
• v2.1: Fixed: no audio after every reboot (sandboxing fix — Helper now creates the shared memory segment, not the sandboxed driver)

---

v2.0 — May 2026

For Everyone

Complete rewrite of the audio routing engine. Replaced the Python socket-based approach with a native C Helper using POSIX shared memory, delivering:

• Dramatically lower latency
• Eliminated audio dropouts on busy systems
• Multi-device, multi-channel routing (up to 8 simultaneous outputs)
• Adaptive sample-rate conversion (SRC) for devices at different rates

---

For architecture diagrams and detailed implementation notes, see the project's internal documentation.