Judge a service by its program, and name any service left stopped

A unit is Omarchy's or the old package's by the program its ExecStart runs or the script it hands that program, not by a path appearing anywhere in the line, so a gateway that merely names ~/.openclaw in an argument is not taken over. The state directory and config overrides are cleared with the profile selectors, since Omarchy's OpenClaw is the default one. Any failure after a running service was stopped for the move names that service, a partial stop included.

Co-Authored-By: Codex XHigh <noreply@openai.com>

bin/omarchy-install-openclaw-cli

@@ -26,9 +26,10 @@ if (( EUID == 0 )); then
   exit 1
 fi
 
-# These select another profile's unit and command, for upstream's installer
-# and the CLI alike, and everything here is about the default ones.
-unset OPENCLAW_PROFILE OPENCLAW_SYSTEMD_UNIT OPENCLAW_WRAPPER
+# Omarchy's OpenClaw is the default one, in ~/.openclaw with the default unit,
+# whatever another profile, unit or state directory a shell selects; a service
+# installed from here would otherwise carry that selection.
+unset OPENCLAW_PROFILE OPENCLAW_SYSTEMD_UNIT OPENCLAW_WRAPPER OPENCLAW_HOME OPENCLAW_STATE_DIR OPENCLAW_CONFIG_PATH
 
 prefix="$HOME/.openclaw"
 runtime_command="$prefix/bin/openclaw"
@@ -69,10 +70,22 @@ installed() {
   omarchy-pkg-present openclaw && runtime_runs && on_path
 }
 
-# The service a unit runs, read from its ExecStart alone: other lines can name
-# ~/.openclaw paths whichever OpenClaw the unit starts.
+# Whether a unit's ExecStart runs a program from under a path: the program
+# itself, or the script it is handed. Later arguments and other lines can name
+# ~/.openclaw whichever OpenClaw the unit starts.
 unit_runs() {
-  grep -E '^ExecStart=' "$1" | grep -qF -- "$2"
+  local words word
+  read -ra words <<<"$(sed -n 's/^ExecStart=//p' "$1" | head -1)"
+  if [[ ${words[0]:-} == "$2"* ]]; then
+    return 0
+  fi
+  for word in "${words[@]:1}"; do
+    if [[ $word != -* ]]; then
+      [[ $word == "$2"* ]]
+      return
+    fi
+  done
+  return 1
 }
 
 # A gateway service already running some other OpenClaw. Upstream's installer
@@ -103,6 +116,9 @@ stop_legacy_services() {
       echo "Could not stop the OpenClaw $role service, so OpenClaw was not moved to $prefix." >&2
       return 1
     fi
+    if [[ " ${running[*]} " == *" $role "* ]]; then
+      stopped+=("$role")
+    fi
   done
 }
 
@@ -129,11 +145,14 @@ rehome_services() {
   done
 }
 
-# Said whenever a failure leaves a service stopped by this run.
+# Said on any failure that leaves a service this run stopped not running.
 stopped_note() {
-  if (( ${#running[@]} )); then
-    echo "The OpenClaw ${running[*]} service was stopped for this and is not running now." >&2
-  fi
+  local role
+  for role in "${stopped[@]}"; do
+    if ! systemctl --user is-active --quiet "openclaw-$role.service"; then
+      echo "The OpenClaw $role service was stopped for this and is not running now." >&2
+    fi
+  done
 }
 
 case "$mode" in
@@ -184,6 +203,7 @@ fi
 # anything changes.
 moving=()
 running=()
+stopped=()
 for role in gateway node; do
   if [[ -f $(unit_path "$role") ]] && unit_runs "$(unit_path "$role")" "/usr/lib/node_modules/openclaw/"; then
     moving+=("$role")
@@ -193,6 +213,7 @@ for role in gateway node; do
   fi
 done
 
+trap 'status=$?; (( status == 0 )) || stopped_note' EXIT
 stop_legacy_services
 
 if [[ $seeding == "true" ]]; then
@@ -202,7 +223,6 @@ if [[ $seeding == "true" ]]; then
   bash "$seed/install-cli.sh" --install-method npm --prefix "$prefix" --version "$seed/openclaw.tgz" --no-onboard || true
   if ! runtime_runs; then
     echo "OpenClaw setup did not complete. Re-run this command after resolving the installer error." >&2
-    stopped_note
     exit 1
   fi
 fi
@@ -210,10 +230,7 @@ fi
 mkdir -p "${command_path%/*}"
 ln -sfn "$runtime_command" "$command_path"
 
-if ! rehome_services; then
-  stopped_note
-  exit 1
-fi
+rehome_services
 
 if ! on_path; then
   echo "$runtime_command is ready, but 'openclaw' on PATH is $(type -P openclaw || echo missing), which is what Omarchy runs." >&2

test/shell.d/openclaw-cli-test.sh

@@ -22,13 +22,14 @@ printf 'pkg-add %s\n' "$*" >>"$OMARCHY_TEST_ROOT/events"
 touch "$OMARCHY_TEST_ROOT/package-installed"
 SH
 # The user manager, as far as these tests need one: a service is active while
-# a marker says so. Stopping clears it; OMARCHY_TEST_STOP_FAIL refuses.
+# a marker says so. Stopping clears it, except for the unit named in
+# OMARCHY_TEST_STOP_FAIL.
 cat >"$mock_bin/systemctl" <<'SH'
 #!/bin/bash
 case "$2" in
   stop)
     printf 'systemctl %s\n' "$*" >>"$OMARCHY_TEST_ROOT/events"
-    [[ -z ${OMARCHY_TEST_STOP_FAIL:-} ]] || exit 1
+    [[ ${OMARCHY_TEST_STOP_FAIL:-} != "$3" ]] || exit 1
     rm -f "$HOME/active-$3"
     ;;
   is-active) [[ -e $HOME/active-$4 ]] ;;
@@ -166,7 +167,7 @@ pass "--check needs the package too, so --now never has a password to ask for un
 # just made, so a gateway running another OpenClaw stops the seeding first.
 new_home foreign-gateway
 mkdir -p "$test_home/.config/systemd/user"
-printf 'Environment=OPENCLAW_CONFIG_PATH=%s/.openclaw/openclaw.json\nExecStart=/opt/node %s/openclaw/dist/index.js gateway\n' "$test_home" "$test_home" \
+printf 'Environment=OPENCLAW_CONFIG_PATH=%s/.openclaw/openclaw.json\nExecStart=/opt/node %s/openclaw/dist/index.js gateway --config %s/.openclaw/openclaw.json\n' "$test_home" "$test_home" "$test_home" \
   >"$test_home/.config/systemd/user/openclaw-gateway.service"
 run omarchy-install-openclaw-cli --now && fail "a gateway running another OpenClaw stops the install"
 grep -q "runs another OpenClaw" "$test_tmp/output" || fail "a gateway running another OpenClaw is named" "$(cat "$test_tmp/output")"
@@ -193,12 +194,23 @@ pass "a service the old package installed moves to the runtime, and only that on
 new_home stop-fails
 mkdir -p "$test_home/.config/systemd/user"
 printf 'ExecStart=/usr/bin/node /usr/lib/node_modules/openclaw/dist/index.js gateway --port 18789\n' >"$test_home/.config/systemd/user/openclaw-gateway.service"
-OMARCHY_TEST_STOP_FAIL=1 run omarchy-install-openclaw-cli --now && fail "a gateway that will not stop stops the install"
+OMARCHY_TEST_STOP_FAIL=openclaw-gateway.service run omarchy-install-openclaw-cli --now && fail "a gateway that will not stop stops the install"
 grep -q "Could not stop the OpenClaw gateway service" "$test_tmp/output" || fail "a gateway that will not stop is named" "$(cat "$test_tmp/output")"
 ! grep -q '^install-cli' "$events" && [[ ! -e $test_home/.openclaw ]] ||
   fail "a gateway that will not stop leaves the runtime unseeded" "$(cat "$events")"
 pass "a gateway that will not stop stops the install before anything is seeded"
 
+new_home stop-partial
+mkdir -p "$test_home/.config/systemd/user"
+for role in gateway node; do
+  printf 'ExecStart=/usr/bin/node /usr/lib/node_modules/openclaw/dist/index.js %s\n' "$role" >"$test_home/.config/systemd/user/openclaw-$role.service"
+  touch "$test_home/active-openclaw-$role.service"
+done
+OMARCHY_TEST_STOP_FAIL=openclaw-node.service run omarchy-install-openclaw-cli --now && fail "a node host that will not stop stops the install"
+grep -q "gateway service was stopped for this and is not running now" "$test_tmp/output" ||
+  fail "a gateway stopped before a later stop failed is named as stopped" "$(cat "$test_tmp/output")"
+pass "a service this run stopped is named whenever the run then fails"
+
 # Upstream's installer rewrites the gateway itself and only warns when it will
 # not start again, so a gateway that was running has to be running afterwards.
 new_home start-fails