From a4465f37c6f12bd1b5a659539bce73224c55b2f3 Mon Sep 17 00:00:00 2001
From: "Oliver Surke (Gitea)" <gitea@surke.de>
Date: Mon, 11 May 2026 18:02:31 +0200
Subject: [PATCH] Improve schedule and mackup reliability

- bin/ssr-schedule: use `launchctl bootstrap`/`bootout` (modern API) with
  legacy `load -w` fallback. The previous `load -w`-only flow silently
  no-op'd on macOS 10.15+ in some configurations, leaving `ssr status`
  reporting `inactive` after a successful `ssr schedule on`.
- bin/ssr-status: query the service via `launchctl print`/`list <label>`
  for consistent active/inactive detection.
- lib/mackup.sh: pre-materialize iCloud placeholders via `brctl download`
  and retry `mackup -f backup` up to three times to dodge the recurring
  `OSError [Errno 66] Directory not empty` race between mackup's rmtree
  and iCloud's filesystem provider. Soft-fails so the rest of `ssr sync`
  still runs even if mackup never settles; logs actionable hints.

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 bin/ssr-schedule | 51 +++++++++++++++++++++++++++++++++++++------
 bin/ssr-status   |  4 +++-
 lib/mackup.sh    | 56 +++++++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 103 insertions(+), 8 deletions(-)

diff --git a/bin/ssr-schedule b/bin/ssr-schedule
index 572206b..8b0224a 100755
--- a/bin/ssr-schedule
+++ b/bin/ssr-schedule
@@ -17,6 +17,20 @@ INTERVAL_HOUR="${SSR_SCHEDULE_HOUR:-9}"
 INTERVAL_MIN="${SSR_SCHEDULE_MINUTE:-0}"
 SSR_BIN="$SSR_ROOT/bin/ssr"
 
+# Modern launchd target. `launchctl load -w` has been deprecated since 10.11
+# and is a silent no-op on recent macOS — use bootstrap/bootout against the
+# GUI domain of the current user.
+GUI_DOMAIN="gui/$(id -u)"
+SERVICE_TARGET="${GUI_DOMAIN}/${LABEL}"
+
+# ssr::schedule::is_loaded — exit 0 iff launchd knows the job.
+# `launchctl print` is the modern check; `launchctl list <label>` is the
+# legacy fallback that still works on every macOS we care about.
+is_loaded() {
+    launchctl print "$SERVICE_TARGET" >/dev/null 2>&1 \
+        || launchctl list "$LABEL"       >/dev/null 2>&1
+}
+
 render_plist() {
     ssr::ensure_dir "$LOG_DIR"
     ssr::ensure_dir "$(dirname "$PLIST_DST")"
@@ -31,18 +45,43 @@ render_plist() {
     chmod 600 "$PLIST_DST"
 }
 
+# Boot the agent into the user's GUI domain, captures stderr for diagnostics.
+# Falls back to the legacy `load -w` only if bootstrap fails entirely.
+load_agent() {
+    local err
+    # Always bootout first so re-runs pick up plist changes.
+    launchctl bootout "$SERVICE_TARGET" >/dev/null 2>&1 || true
+
+    err="$(launchctl bootstrap "$GUI_DOMAIN" "$PLIST_DST" 2>&1)" || {
+        ssr::warn "launchctl bootstrap failed: ${err:-no stderr}"
+        ssr::warn "Falling back to legacy 'launchctl load -w'"
+        launchctl load -w "$PLIST_DST" 2>&1 || true
+    }
+    # `enable` makes the job survive logout/reboot (parity with old `load -w`).
+    launchctl enable "$SERVICE_TARGET" >/dev/null 2>&1 || true
+}
+
+unload_agent() {
+    launchctl bootout "$SERVICE_TARGET" >/dev/null 2>&1 \
+        || launchctl unload "$PLIST_DST" 2>/dev/null || true
+}
+
 cmd_on() {
     [[ -f "$PLIST_SRC" ]] || ssr::die "launchd template missing: $PLIST_SRC"
     render_plist
-    # Reload to pick up changes if it was already loaded.
-    launchctl unload "$PLIST_DST" 2>/dev/null || true
-    launchctl load -w "$PLIST_DST"
-    ssr::ok "Scheduled sync enabled (${INTERVAL_HOUR}:$(printf '%02d' "$INTERVAL_MIN") daily) → $PLIST_DST"
+    load_agent
+    if is_loaded; then
+        ssr::ok "Scheduled sync enabled (${INTERVAL_HOUR}:$(printf '%02d' "$INTERVAL_MIN") daily) → $PLIST_DST"
+    else
+        ssr::die "launchd did not register the job. Diagnose with:
+    launchctl bootstrap $GUI_DOMAIN $PLIST_DST
+    launchctl print $SERVICE_TARGET"
+    fi
 }
 
 cmd_off() {
     if [[ -f "$PLIST_DST" ]]; then
-        launchctl unload "$PLIST_DST" 2>/dev/null || true
+        unload_agent
         rm -f "$PLIST_DST"
         ssr::ok "Scheduled sync disabled"
     else
@@ -51,7 +90,7 @@ cmd_off() {
 }
 
 cmd_status() {
-    if [[ -f "$PLIST_DST" ]] && launchctl list | grep -q "$LABEL"; then
+    if is_loaded; then
         ssr::ok "Scheduled sync: active (label=$LABEL)"
     else
         ssr::log "Scheduled sync: inactive"
diff --git a/bin/ssr-status b/bin/ssr-status
index 2fbbb16..8e2520c 100755
--- a/bin/ssr-status
+++ b/bin/ssr-status
@@ -31,7 +31,9 @@ else
 fi
 
 label="${SSR_LAUNCHD_LABEL:-de.surke.ssr.sync}"
-if launchctl list 2>/dev/null | grep -q "$label"; then
+service_target="gui/$(id -u)/${label}"
+if launchctl print "$service_target" >/dev/null 2>&1 \
+   || launchctl list "$label"        >/dev/null 2>&1; then
     echo "  schedule:   active ($label)"
 else
     echo "  schedule:   inactive"
diff --git a/lib/mackup.sh b/lib/mackup.sh
index 995478a..70a15b4 100644
--- a/lib/mackup.sh
+++ b/lib/mackup.sh
@@ -32,9 +32,63 @@ ssr::mackup::configure() {
     ssr::ok "Wrote $cfg"
 }
 
+# Force iCloud Drive to materialize placeholder (.icloud) files under the
+# Mackup storage directory before mackup walks them. The most common ssr sync
+# crash is `OSError [Errno 66] Directory not empty` raised by shutil.rmtree
+# inside mackup when iCloud lazily re-creates dataless placeholders between
+# listdir() and rmdir(). Pre-downloading sidesteps the race.
+ssr::mackup::_materialize_icloud() {
+    local mackup_root="$1"
+    [[ -d "$mackup_root" ]] || return 0
+    case "$mackup_root" in
+        *Mobile\ Documents*) ;;            # only worth doing for iCloud
+        *) return 0 ;;
+    esac
+    # `brctl download` is the documented way to request file materialization.
+    # Available on macOS 10.12+; silently skip if missing on older systems.
+    if command -v brctl >/dev/null 2>&1; then
+        brctl download "$mackup_root" >/dev/null 2>&1 || true
+    fi
+    # Belt-and-braces: prune any stub files that survived the download
+    # request (sometimes left over from interrupted prior runs).
+    find "$mackup_root" -name '*.icloud' -type f -delete 2>/dev/null || true
+}
+
+# Try `mackup -f backup` with a small retry budget. iCloud-induced races are
+# transient: a 2-second sleep is usually enough for the provider to settle.
+# We deliberately do NOT abort `ssr sync` on persistent failure — the user
+# still benefits from brew dump, cleanup, and the unmanaged-apps scan that
+# run afterwards. The failure is surfaced loudly via ssr::warn and syslog.
 ssr::mackup::backup() {
     ssr::log "mackup backup"
-    mackup -f backup
+
+    # Derive mackup root from ~/.mackup.cfg (path + directory).
+    local cfg="$HOME/.mackup.cfg" mackup_root="" path dir
+    if [[ -f "$cfg" ]]; then
+        path="$(awk -F= '/^path/{gsub(/^[[:space:]]+|[[:space:]]+$/,"",$2); print $2}' "$cfg")"
+        dir="$( awk -F= '/^directory/{gsub(/^[[:space:]]+|[[:space:]]+$/,"",$2); print $2}' "$cfg")"
+        [[ -n "$path" && -n "$dir" ]] && mackup_root="$path/$dir"
+    fi
+    [[ -n "$mackup_root" ]] && ssr::mackup::_materialize_icloud "$mackup_root"
+
+    local attempt rc=0
+    for attempt in 1 2 3; do
+        if mackup -f backup; then
+            return 0
+        fi
+        rc=$?
+        ssr::warn "mackup backup failed (attempt $attempt/3, rc=$rc)"
+        if [[ $attempt -lt 3 ]]; then
+            [[ -n "$mackup_root" ]] && ssr::mackup::_materialize_icloud "$mackup_root"
+            sleep 2
+        fi
+    done
+
+    ssr::warn "mackup backup did not complete after 3 attempts (rc=$rc)."
+    ssr::warn "Sync will continue without it; rerun 'mackup -f backup' manually,"
+    ssr::warn "or switch ~/.mackup.cfg to a non-iCloud storage path if this persists."
+    ssr::syslog "mackup backup failed after retries (rc=$rc)"
+    return 0   # soft-fail so the rest of ssr sync still runs
 }
 
 ssr::mackup::restore() {