[Bug] macOS 26: 16.1.0/16.1.1 IPC socket churn causes Virtual HID outages and stuck/repeated input

[MahdiNazemi]

Symptoms

• Intermittent input freezes, repeated characters after a single key press, among other things.
• In 16.1.x, Core Service and Virtual HID logs continuously alternate receiver: closed / receiver: bound about every two seconds.
• During one incident, the device grabber stopped and Virtual HID was unavailable for about 40 seconds. After recovery, Karabiner logged drop unmatched momentary switch key_up before merged_input_event_queue insertion.

Downgrading to 16.0.0 seems to have solved the problem.

Environment

• macOS 26.5.2 (25F84), Apple Silicon
• Karabiner-Elements 16.1.0, then 16.1.1 beta
• Internal Apple keyboard/trackpad; an external USB keyboard sometimes connected

[tekezo]

I'm not seeing the same issue on my end, so I'd appreciate a bit more information.
Is there a crash report in the Console?

[MahdiNazemi]

Apologies for the delay. Since I downgraded to an older version, I’m not sure whether I can access the crash reports. I’ll upgrade today and report back if the issue happens again.

[buck-openart]

I'm not seeing the same issue on my end, so I'd appreciate a bit more information.

Deterministic repro on Apple Silicon (any sleep longer than ~10 seconds):

sudo pmset relative wake 12 && pmset sleepnow

(The commanded 12 produced an actual sleep of 16s per pmset -g log; sleep entry latency and RTC slack add a few seconds. If it doesn't trigger on your machine, pad the number to 20 so the actual sleep clears 10s.) Tested sleep durations: ~5s is clean, ~16s and ~49s both produce ~11s of dead keyboard after wake. The threshold matters: a quick lid-close test sleeps too briefly to trigger it, which may be why it doesn't reproduce everywhere.

The 10 seconds appears to come from the new IPC library that 16.1.0 switched to. pqrs::unix_domain_stream declares a peer dead after heartbeat_timeout = 10000ms (options.hpp) with no frame received. The deadline runs on asio::steady_timer, and mach_absolute_time keeps advancing during sleep on Apple Silicon, so after any sleep over ~10s every connection's deadline has expired at wake and all peers get declared dead ("Operation timed out") even though they were just suspended. Log from the 16s run:

[2026-07-17 10:43:21.970] [info] system_will_power_on
[2026-07-17 10:43:21.977] [debug] console_user_id_changed_receiver: peer_error_occurred (42286): Operation timed out
[2026-07-17 10:43:31.980] [debug] device_grabber is stopped.
[2026-07-17 10:43:33.063] [debug] device_grabber is started.
[2026-07-17 10:43:34.172] [info] Apple Internal Keyboard / Trackpad (device_id:4294971328) hid queue value monitor is started (grabbed).

[tekezo]

I made some improvements to the related code in v16.1.4. Please give the beta version a try.
How to upgrade to the latest beta version

[tekezo]

I'll continue working on this in #4498.