Fixing watchdog

systemd/streamdeck-go-watchdog.sh

@@ -106,14 +106,52 @@ prev=""
 
 curr="$(current_addr)"
 
-# Always update the state file so the next run sees a fresh baseline.
+# Only update the state file when the device is present. If we overwrote with
+# an empty string while the device was absent (e.g. mid-KVM-swap), the very
+# next run would see prev="" and miss the address change on return.
+if [[ -n "$curr" ]]; then
     printf '%s' "$curr" > "$STATE_FILE"
+fi
 
 # No device present — nothing to do. Don't touch the service.
 if [[ -z "$curr" ]]; then
     exit 0
 fi
 
+# Linux-only: detect a stale hidraw fd held by the daemon. When the device
+# unplugs, hidraw's open fd survives but its /dev node is removed; procfs
+# marks the symlink "(deleted)". hid_read_timeout on this fd silently returns
+# zero bytes, so the daemon's 3-error reconnect path never trips.
+stale_fd_detected() {
+    [[ "$OS" != "Linux" ]] && return 1
+    local pid
+    pid="$(systemctl --user show -p MainPID --value streamdeck-go.service 2>/dev/null || true)"
+    [[ -z "$pid" || "$pid" == "0" ]] && return 1
+    [[ ! -d "/proc/$pid/fd" ]] && return 1
+    ls -la "/proc/$pid/fd/" 2>/dev/null | grep -qE 'hidraw[0-9]+ \(deleted\)'
+}
+
+# Linux-only: detect that the system resumed from suspend after the daemon
+# started. On resume, the xhci controller may reset the deck's USB device
+# in place (same bus address, same hidraw node, fd not deleted). The kernel
+# reset leaves the existing fd's input queue dead — buttons no longer reach
+# userspace — but no externally visible signal flags the failure. Restarting
+# the daemon is cheap and reliably fixes it.
+#
+# Idempotent by construction: once we restart, the daemon's ActiveEnterTimestamp
+# moves past the resume event, so this check stops firing until the next sleep.
+resumed_since_start() {
+    [[ "$OS" != "Linux" ]] && return 1
+    local started
+    started="$(systemctl --user show -p ActiveEnterTimestamp --value streamdeck-go.service 2>/dev/null || true)"
+    [[ -z "$started" || "$started" == "n/a" ]] && return 1
+    local started_epoch
+    started_epoch="$(date -d "$started" +%s 2>/dev/null || true)"
+    [[ -z "$started_epoch" ]] && return 1
+    journalctl -k --since "@$started_epoch" --no-pager 2>/dev/null \
+        | grep -qE 'PM: suspend exit|PM: Finishing wakeup'
+}
+
 reason=""
 
 if [[ -z "$prev" ]]; then
@@ -127,6 +165,10 @@ elif [[ "$curr" != "$prev" ]]; then
     reason="device address changed: $prev → $curr (likely unplug/replug)"
 elif ! service_active; then
     reason="device present at $curr but service is not active"
+elif stale_fd_detected; then
+    reason="daemon holds a deleted hidraw fd (post-unplug stale handle)"
+elif resumed_since_start; then
+    reason="system resumed from suspend since daemon started (USB reset may have invalidated input queue)"
 fi
 
 if [[ -n "$reason" ]]; then