Recover shell restarts when the replacement launch is rejected

bin/omarchy-restart-shell

@@ -67,9 +67,16 @@ while timeout 5 quickshell kill -p "$CONFIG_DIR" --any-display >/dev/null 2>&1;
 
 # Spawn from Hyprland so the shell inherits the canonical session environment,
 # not transient variables from a terminal, SSH connection, or development tool.
-hyprctl dispatch 'hl.dsp.exec_cmd("omarchy-launch-shell")' >/dev/null
+launch_shell() {
+  hyprctl dispatch 'hl.dsp.exec_cmd("omarchy-launch-shell")' >/dev/null
+}
+
+launch_shell
 
-for (( attempt = 0; attempt < 20; attempt++ )); do
+# Plugin discovery and first-load QML compilation can take several seconds.
+ready_deadline=$((SECONDS + 30))
+next_launch=$((SECONDS + 2))
+while (( SECONDS < ready_deadline )); do
   if OMARCHY_PATH="$session_omarchy_path" OMARCHY_SHELL_IPC_TIMEOUT=0.5s omarchy-shell shell ping >/dev/null 2>&1; then
     # The session stays compositor-locked after the old lock client died, so
     # re-acquire the lock and let the user authenticate out of it.
@@ -86,6 +93,14 @@ for (( attempt = 0; attempt < 20; attempt++ )); do
     systemctl --user try-restart 'omarchy-*-invitation.service' 2>/dev/null || true
     exit 0
   fi
+
+  # Quickshell can release its IPC target before its singleton lock. A launch
+  # in that gap exits immediately, so retry until one replacement stays up.
+  if (( SECONDS >= next_launch )); then
+    launch_shell
+    next_launch=$((SECONDS + 2))
+  fi
+
   sleep 0.1
 done
 

test/shell.d/restart-shell-test.sh

@@ -73,6 +73,7 @@ restart_state="$test_tmp/restart-pids"
 restart_log="$test_tmp/restart.log"
 restart_env_log="$test_tmp/restart-env.log"
 dispatch_log="$test_tmp/dispatch.log"
+dispatch_count="$test_tmp/dispatch-count"
 ipc_log="$test_tmp/ipc.log"
 runtime_dir="$test_tmp/runtime"
 mkdir -p "$restart_root/shell" "$restart_bin" "$runtime_dir"
@@ -89,9 +90,19 @@ printf '%s\n' "$*" >>"$OMARCHY_TEST_IPC_LOG"
 
 case "$*" in
   *'shell ping')
-    [[ $* == *"-p $OMARCHY_TEST_SESSION_PATH/shell"* ]] &&
+    if [[ -n ${OMARCHY_TEST_QS_READY_AFTER:-} ]]; then
+      attempts=$(<"$OMARCHY_TEST_QS_READY_COUNT")
+      attempts=$((attempts + 1))
+      printf '%s\n' "$attempts" >"$OMARCHY_TEST_QS_READY_COUNT"
+    fi
+
+    if [[ $* == *"-p $OMARCHY_TEST_SESSION_PATH/shell"* ]] &&
       grep -Fx '303' "$OMARCHY_TEST_QS_STATE" >/dev/null &&
+      (( ${attempts:-1} >= ${OMARCHY_TEST_QS_READY_AFTER:-1} )); then
       printf 'ok\n'
+    else
+      exit 1
+    fi
     ;;
   *'lock lock')
     touch "$OMARCHY_TEST_QS_STATE.locked"
@@ -141,8 +152,15 @@ if [[ ${1:-} == "-j" && ${2:-} == "monitors" ]]; then
   fi
 elif [[ ${1:-} == "dispatch" && ${2:-} == hl.dsp.exec_cmd* ]]; then
   printf '%s\n' "${2:-}" >>"$OMARCHY_TEST_DISPATCH_LOG"
-  OMARCHY_PATH="$OMARCHY_TEST_SESSION_PATH" \
-    env -u OMARCHY_TEST_TRANSIENT_ENV omarchy-launch-shell
+  if [[ -n ${OMARCHY_TEST_DISPATCH_COUNT:-} ]]; then
+    attempts=$(<"$OMARCHY_TEST_DISPATCH_COUNT")
+    attempts=$((attempts + 1))
+    printf '%s\n' "$attempts" >"$OMARCHY_TEST_DISPATCH_COUNT"
+  fi
+  if (( ${attempts:-1} > ${OMARCHY_TEST_DISPATCH_FAILS:-0} )); then
+    OMARCHY_PATH="$OMARCHY_TEST_SESSION_PATH" \
+      env -u OMARCHY_TEST_TRANSIENT_ENV omarchy-launch-shell
+  fi
   printf 'ok\n'
 elif [[ ${1:-} == "dispatch" ]]; then
   exit 1
@@ -214,6 +232,52 @@ grep -F 'hl.dsp.exec_cmd("omarchy-launch-shell")' "$dispatch_log" >/dev/null ||
 grep -F "ipc -n -p $restart_root/shell call -- shell ping" "$ipc_log" >/dev/null || fail "restart checks readiness in the session checkout"
 pass "restart replaces duplicate shell instances from the session checkout"
 
+# Plugin discovery and first-load QML compilation can take longer than the old
+# 20-attempt readiness budget. Keep polling long enough for a slow shell that
+# eventually becomes IPC-responsive.
+ready_count="$test_tmp/ready-count"
+printf '0\n' >"$ready_count"
+printf '303\n' >"$restart_state"
+
+PATH="$restart_bin:$PATH" \
+OMARCHY_PATH="$restart_root" \
+XDG_RUNTIME_DIR="$runtime_dir" \
+OMARCHY_TEST_QS_STATE="$restart_state" \
+OMARCHY_TEST_QS_LOG="$restart_log" \
+OMARCHY_TEST_QS_ENV_LOG="$restart_env_log" \
+OMARCHY_TEST_DISPATCH_LOG="$dispatch_log" \
+OMARCHY_TEST_IPC_LOG="$ipc_log" \
+OMARCHY_TEST_SESSION_PATH="$restart_root" \
+OMARCHY_TEST_QS_READY_AFTER=25 \
+OMARCHY_TEST_QS_READY_COUNT="$ready_count" \
+  timeout 8 "$ROOT/bin/omarchy-restart-shell" || fail "restart waits for a slow shell to become ready"
+
+(( $(<"$ready_count") >= 25 )) || fail "restart keeps polling beyond the old readiness budget"
+pass "restart waits for a slow shell to become ready"
+
+# Quickshell can release its IPC target before releasing its singleton lock.
+# When that happens the first replacement exits with "instance already
+# running", so waiting for readiness alone can never recover the bar.
+printf '0\n' >"$dispatch_count"
+: >"$restart_state"
+
+PATH="$restart_bin:$PATH" \
+OMARCHY_PATH="$restart_root" \
+XDG_RUNTIME_DIR="$runtime_dir" \
+OMARCHY_TEST_QS_STATE="$restart_state" \
+OMARCHY_TEST_QS_LOG="$restart_log" \
+OMARCHY_TEST_QS_ENV_LOG="$restart_env_log" \
+OMARCHY_TEST_DISPATCH_LOG="$dispatch_log" \
+OMARCHY_TEST_DISPATCH_COUNT="$dispatch_count" \
+OMARCHY_TEST_DISPATCH_FAILS=1 \
+OMARCHY_TEST_IPC_LOG="$ipc_log" \
+OMARCHY_TEST_SESSION_PATH="$restart_root" \
+  timeout 8 "$ROOT/bin/omarchy-restart-shell" || fail "restart retries a replacement that loses the singleton-lock race"
+
+(( $(<"$dispatch_count") >= 2 )) || fail "restart retries shell launch after the first replacement exits"
+[[ $(<"$restart_state") == 303 ]] || fail "retried launch leaves one fresh shell instance"
+pass "restart retries a replacement that loses the singleton-lock race"
+
 : >"$restart_log"
 printf '303\n' >"$restart_state"
 touch "$restart_state.locked"