From d1bd6aae488404d7e85663137bf4a38960cc1c7d Mon Sep 17 00:00:00 2001
From: Josue Zamudio <josuezamudio1@gmail.com>
Date: Tue, 27 Oct 2026 18:32:33 -0700
Subject: [PATCH] fix(shikane): detect silent deadlocks; re-apply profile on
 sway reload

watchdog: covers both observed deadlock flavours -- the loud one that repeats
"Received invalid input AtomicChangeDone at state Discarded", and the silent
one that logs nothing and just never leaves VariantInProgress.

shikane-recover-on-reload.sh (new, previously referenced by sway/config but
never committed): shikanectl reload -> sleep 2 -> watchdog check, falling
back to a service restart when reload cannot recover.

clamshell-mode.sh: documents why the laptop panel is no longer enabled
unconditionally on lid-open, which raced shikane's "External Display" profile
(eDP-1 disabled by design) and produced "Atomic commit failed: Device or
resource busy" churn.
---
 .config/sway/scripts/clamshell-mode.sh        | 10 ++++++-
 .../sway/scripts/shikane-recover-on-reload.sh | 13 +++++++++
 .config/sway/scripts/shikane-watchdog.sh      | 29 ++++++++++++++-----
 3 files changed, 43 insertions(+), 9 deletions(-)
 create mode 100755 .config/sway/scripts/shikane-recover-on-reload.sh

diff --git a/.config/sway/scripts/clamshell-mode.sh b/.config/sway/scripts/clamshell-mode.sh
index 98de4c3..9237008 100755
--- a/.config/sway/scripts/clamshell-mode.sh
+++ b/.config/sway/scripts/clamshell-mode.sh
@@ -1,4 +1,12 @@
 #!/bin/sh
+# Lid-driven laptop panel handling.
+#
+# shikane is authoritative for output layout/scaling
+# (~/.config/shikane/config.toml + shikane-watchdog.timer). This script used to
+# run `swaymsg output eDP-1 enable` whenever the lid was open, which on every
+# `swaymsg reload` raced the shikane "External Display" profile (eDP-1 off) and
+# contributed to "Atomic commit failed: Device or resource busy" churn.
+# It now only ever disables.
 
 LAPTOP_OUTPUT="eDP-1"
 LID_STATE_FILE="/proc/acpi/button/lid/LID/state"
@@ -6,7 +14,7 @@ LID_STATE_FILE="/proc/acpi/button/lid/LID/state"
 read -r LS < "$LID_STATE_FILE"
 
 case "$LS" in
-*open)   swaymsg output "$LAPTOP_OUTPUT" enable ;;
 *closed) swaymsg output "$LAPTOP_OUTPUT" disable ;;
+*open)   : ;;   # leave to shikane
 *)       echo "Could not get lid state" >&2 ; exit 1 ;;
 esac
diff --git a/.config/sway/scripts/shikane-recover-on-reload.sh b/.config/sway/scripts/shikane-recover-on-reload.sh
new file mode 100755
index 0000000..5ebacfa
--- /dev/null
+++ b/.config/sway/scripts/shikane-recover-on-reload.sh
@@ -0,0 +1,13 @@
+#!/bin/sh
+# shikane-recover-on-reload: run after every sway startup/reload
+# (wired up via exec_always in the sway config).
+#
+# Re-applies the shikane profile, then verifies the daemon came out
+# the other side: if the re-apply hit the commit-failure race, the
+# watchdog check (2s later) restarts the service to recover.
+
+set -u
+
+shikanectl reload 2>/dev/null
+sleep 2
+exec /home/work/.config/sway/scripts/shikane-watchdog.sh
diff --git a/.config/sway/scripts/shikane-watchdog.sh b/.config/sway/scripts/shikane-watchdog.sh
index 494b91c..5b55899 100755
--- a/.config/sway/scripts/shikane-watchdog.sh
+++ b/.config/sway/scripts/shikane-watchdog.sh
@@ -3,8 +3,11 @@
 #
 # Bug (shikane 1.1.1): if a variant application is discarded mid-way
 # (e.g. by a failing DP atomic commit), the daemon gets stuck in
-# VariantInProgress/Discarded and keeps logging
-#   [WARN shikane::variant] Received invalid input AtomicChangeDone at state Discarded
+# VariantInProgress/Discarded. Two flavors observed:
+#  loud:   keeps logging
+#           [WARN shikane::variant] Received invalid input AtomicChangeDone at state Discarded
+#  silent: no WARN at all; the variant state ends at "Discarded" and the
+#          daemon state machine simply never leaves VariantInProgress
 # `shikanectl reload` does not recover it; a service restart does
 # (re-applies the active profile in ~1s).
 #
@@ -25,21 +28,31 @@ if [ -n "$latest3" ]; then
         *VariantApplied*)
             : ;;  # healthy
         *)
-            # InProgress/Discarded within the last 3 min: only a deadlock
-            # if the WARN signature is present (a legitimate slow transition
-            # has no WARN)
+            # Daemon still InProgress. Deadlock signatures (a legitimate
+            # slow transition has neither):
+            #  loud:   the WARN from the discarded-variant race
+            #  silent: variant discarded but daemon never processed it
             printf '%s\n' "$logs3" | grep -q 'invalid input' && stuck=1
+            if [ "$stuck" -eq 0 ]; then
+                lastvariant=$(printf '%s\n' "$logs3" \
+                    | grep 'shikane::variant' | grep 'New variant state' | tail -1)
+                case "$lastvariant" in
+                    *Discarded*) stuck=1 ;;
+                esac
+            fi
             ;;
     esac
 else
     # No state change in the last 3 min — check the last known state
-    last10=$(journalctl --user -u "$unit" -q --since "-10min" 2>/dev/null \
+    # (unbounded: a stuck daemon may have gone silent long ago,
+    # e.g. across a laptop suspend)
+    lastline=$(journalctl --user -u "$unit" -q -n 500 2>/dev/null \
         | grep 'daemon::state_machine' | tail -1)
-    case "$last10" in
+    case "$lastline" in
         '' | *VariantApplied*)
             : ;;  # stable and applied, or nothing to check
         *)
-            stuck=1 ;;  # stuck in-progress/discarded for >3 min
+            stuck=1 ;;  # stuck in-progress/discarded, any age
     esac
 fi