Attempt to improve the watchdog because it has failed on some systems.

scripts/i3_watchdog.sh

@@ -17,9 +17,40 @@
 checkInterval=3  # Check every 3 seconds
 timeoutSeconds=2  # Consider i3 locked if it doesn't respond within 2 seconds
 pidFile="${HOME}/.config/i3/i3_watchdog.pid"
+logFile="/tmp/i38-i3-watchdog-${UID}.log"
+maxLogBytes=262144
+
+rotate_log() {
+    local logSize
+
+    [[ -f "$logFile" ]] || return 0
+    logSize=$(wc -c < "$logFile")
+    if ((logSize >= maxLogBytes)); then
+        mv -f -- "$logFile" "${logFile}.old"
+    fi
+}
 
 log_message() {
-    echo "[i3-watchdog] $*"
+    local message
+
+    message="[i3-watchdog] $* - $(date '+%Y-%m-%d %H:%M:%S')"
+
+    rotate_log
+    printf '%s\n' "$message"
+    printf '%s\n' "$message" >> "$logFile"
+}
+
+check_health() {
+    ipcWorks=0
+    focusWorks=0
+
+    if timeout "$timeoutSeconds" i3-msg -t get_version > /dev/null 2>&1; then
+        ipcWorks=1
+    fi
+
+    if timeout 1 xdotool getwindowfocus > /dev/null 2>&1; then
+        focusWorks=1
+    fi
 }
 
 # Kill any existing watchdog instances
@@ -50,17 +81,7 @@ while true; do
     fi
 
     # Try to communicate with i3 IPC AND check X focus
-    ipcWorks=0
-    focusWorks=0
-
-    if timeout "$timeoutSeconds" i3-msg -t get_version > /dev/null 2>&1; then
-        ipcWorks=1
-    fi
-
-    # Also check if X focus is working (use current DISPLAY, not hardcoded)
-    if timeout 1 xdotool getwindowfocus > /dev/null 2>&1; then
-        focusWorks=1
-    fi
+    check_health
 
     # Both must work for system to be healthy
     if [[ $ipcWorks -eq 1 ]] && [[ $focusWorks -eq 1 ]]; then
@@ -101,7 +122,7 @@ while true; do
             fi
 
             # Try to get i3 socket info
-            i3Socket=$(i3 --get-socketpath 2>/dev/null)
+            i3Socket=$(timeout "$timeoutSeconds" i3 --get-socketpath 2>/dev/null)
             log_message "i3 socket path: $i3Socket"
 
             if [[ -n "$i3Socket" ]] && [[ -S "$i3Socket" ]]; then
@@ -112,43 +133,54 @@ while true; do
             fi
 
             # Check if this is a focus issue (Wine keyboard grab bug)
-            focusCheck=$(xdotool getwindowfocus 2>&1)
+            focusCheck=$(timeout 1 xdotool getwindowfocus 2>&1)
             log_message "Focus check result: $focusCheck"
 
             # Try to reset focus to i3
             log_message "Attempting to reset X focus..."
-            xdotool key --clearmodifiers Super_L 2>/dev/null
+            timeout 1 xdotool key --clearmodifiers Super_L 2>/dev/null
             sleep 0.5
 
             # Try to focus on i3's root window
-            i3RootWindow=$(xdotool search --class "i3" | head -1)
+            i3RootWindow=$(timeout 1 xdotool search --class "i3" 2>/dev/null | head -1)
             if [[ -n "$i3RootWindow" ]]; then
                 log_message "Focusing i3 root window: $i3RootWindow"
-                xdotool windowfocus "$i3RootWindow" 2>/dev/null
+                timeout 1 xdotool windowfocus "$i3RootWindow" 2>/dev/null
             fi
 
             # Try i3-msg to focus something
             log_message "Using i3-msg to focus workspace..."
-            i3-msg workspace number 1 >/dev/null 2>&1
+            timeout "$timeoutSeconds" i3-msg workspace number 1 >/dev/null 2>&1
             sleep 0.5
-            i3-msg focus output primary >/dev/null 2>&1
+            timeout "$timeoutSeconds" i3-msg focus output primary >/dev/null 2>&1
             sleep 1
 
-            # Check if focus is actually fixed now
-            if timeout 1 xdotool getwindowfocus >/dev/null 2>&1; then
-                log_message "Focus recovery successful!"
+            # Recovery is successful only if both i3 IPC and X focus work.
+            check_health
+            if [[ $ipcWorks -eq 1 ]] && [[ $focusWorks -eq 1 ]]; then
+                log_message "Recovery successful"
+                lastSuccessTime=$(date +%s)
+                consecutiveFailures=0
             else
-                log_message "Focus still broken - restarting i3..."
-                i3-msg -t run_command restart
-                log_message "Restart command sent"
+                log_message "Recovery incomplete - IPC:$ipcWorks Focus:$focusWorks; restarting i3"
+                if timeout "$timeoutSeconds" i3-msg -t run_command restart >/dev/null 2>&1; then
+                    log_message "Restart command accepted"
+                else
+                    log_message "Restart command failed or timed out"
+                fi
+
+                # Wait for restart to complete, then verify it actually recovered.
+                sleep 5
+                check_health
+                if [[ $ipcWorks -eq 1 ]] && [[ $focusWorks -eq 1 ]]; then
+                    log_message "i3 healthy after restart"
+                    lastSuccessTime=$(date +%s)
+                    consecutiveFailures=0
+                else
+                    log_message "i3 still unhealthy after restart - IPC:$ipcWorks Focus:$focusWorks"
+                fi
             fi
 
-            # Wait for restart to complete
-            sleep 5
-
-            # Reset counters
-            lastSuccessTime=$(date +%s)
-            consecutiveFailures=0
         fi
     fi
 done