#!/usr/bin/env bash
#
# Scheduled dependency update + vulnerability scan (every ~16 days).
#
# Driven by the systemd user timer `skatespots-deps.timer` (a daily tick; this
# script self-gates to one real run per INTERVAL_DAYS). A second weekly timer
# (`skatespots-deps-scan.timer`) runs this with --scan-only for earlier CVE
# visibility without touching the lockfile. What it does:
#   1. bails if the working tree is dirty (never mixes with your own changes);
#   2. scans `npm outdated` + `npm audit` into a dated report;
#   3. applies NON-breaking updates (`npm update` + `npm audit fix` — never --force);
#   4. runs the full test suite (`npm test`) AND the boot smoke test;
#   5. on success commits + pushes the lockfile/package.json (the deploy timer then
#      ships it); on failure it RESTORES the previous lockfile and reports.
#
# Flags:
#   --scan-only   report only (no apply, no commit)
#   --no-push     apply + test + commit locally, but don't push
#   --force       ignore the interval gate
#
# Env: DEPS_INTERVAL_DAYS (default 20), DEPS_STATE_DIR (default ~/.local/state/skatespots-deps)
set -uo pipefail

REPO="/home/sovereignsk8/SkateSpots"
# Resolve the newest nvm Node so a Node upgrade doesn't silently break the
# timer (the old hard-coded v22.23.1 path did).
if [ -d "$HOME/.nvm/versions/node" ]; then
  NODE_BIN="$(ls -d "$HOME"/.nvm/versions/node/*/bin 2>/dev/null | sort -V | tail -1)"
  [ -n "$NODE_BIN" ] && export PATH="$NODE_BIN:$PATH"
fi
export PATH="$HOME/go/bin:$PATH"
# The job runs the full gate itself; don't make the pre-push hook repeat it.
export SKIP_PREFLIGHT=1
cd "$REPO"

INTERVAL_DAYS="${DEPS_INTERVAL_DAYS:-16}"
STATE_DIR="${DEPS_STATE_DIR:-$HOME/.local/state/skatespots-deps}"
mkdir -p "$STATE_DIR"
LAST="$STATE_DIR/last-run"
REPORT="$STATE_DIR/report-$(LC_ALL=C date +%F).txt"
FAILURE_FILE="$STATE_DIR/last-failure"
FAILURE_DIR_BASE="$STATE_DIR/failures"
OUTDATED_JSON="$STATE_DIR/outdated-$(LC_ALL=C date +%F).json"
# OSV-Scanner (broader than npm audit: queries the OSV database for the lockfile).
# Installed with: go install github.com/google/osv-scanner/v2/cmd/osv-scanner@latest
OSV_BIN="${OSV_BIN:-$HOME/go/bin/osv-scanner}"

SCAN_ONLY=0; NO_PUSH=0; FORCE=0
for a in "$@"; do
  case "$a" in
    --scan-only) SCAN_ONLY=1 ;;
    --no-push) NO_PUSH=1 ;;
    --force) FORCE=1 ;;
  esac
done

log() { echo "[$(date -Is)] $*"; }
notify() {
  command -v notify-send >/dev/null 2>&1 && notify-send -a SkateSpots "Dependency job" "$1" 2>/dev/null || true
}
finish_timer() { date +%s > "$LAST"; }

# ── fallback helpers ────────────────────────────────────────────────────────
# Fast gate for the per-package fallback. The full `npm test` + smoke still runs
# once at the end; this keeps the fallback from taking an hour.
fast_gate() {
  npx tsc --noEmit >/dev/null 2>&1 \
    && npx tsc -p cvm-server/tsconfig.json >/dev/null 2>&1 \
    && npx tsc -p ai-proxy/tsconfig.json >/dev/null 2>&1 \
    && npx tsc -p push-service/tsconfig.json >/dev/null 2>&1 \
    && node scripts/check-i18n.mjs >/dev/null 2>&1 \
    && npx vitest run --reporter=dot --silent >/dev/null 2>&1
}

# If the blanket `npm update` fails, retry only patch-level bumps one package at
# a time; keep each one that passes the fast gate and skip the rest. This keeps
# safe updates instead of reverting the whole batch. Requires OUTDATED_JSON
# captured before the full update.
patch_only_fallback() {
  local tmp list pkg kept=() skipped=()
  tmp="$(mktemp -d)"
  cp package.json package-lock.json "$tmp"/ 2>/dev/null || true

  list="$(node -e '
    const fs = require("fs");
    let j = {};
    try { j = JSON.parse(fs.readFileSync(process.argv[1], "utf8")); } catch {}
    for (const [name, v] of Object.entries(j)) {
      const cur = String(v.current || ""), want = String(v.wanted || "");
      if (!cur || !want || cur === want) continue;
      const a = cur.split("."), b = want.split(".");
      if (a.length < 3 || b.length < 3) continue;
      if (a[0] === b[0] && a[1] === b[1] && Number(b[2]) > Number(a[2])) console.log(name);
    }
  ' "$OUTDATED_JSON" 2>/dev/null || true)"

  if [ -z "$list" ]; then
    log "fallback: no patch-only updates available"
    rm -rf "$tmp"
    return 1
  fi

  while IFS= read -r pkg; do
    [ -n "$pkg" ] || continue
    log "fallback: trying $pkg"
    if npm update "$pkg" >/dev/null 2>&1 \
       && npm dedupe >/dev/null 2>&1 \
       && fast_gate; then
      kept+=("$pkg")
      # New baseline for the next package.
      cp package.json package-lock.json "$tmp"/ 2>/dev/null || true
    else
      log "fallback: $pkg failed the fast gate — reverting it"
      cp "$tmp/package.json" "$tmp/package-lock.json" . 2>/dev/null || true
      npm i --silent >/dev/null 2>&1 || true
      skipped+=("$pkg")
    fi
  done <<< "$list"

  rm -rf "$tmp"
  log "fallback: kept ${#kept[@]} (${kept[*]:-none}), skipped ${#skipped[@]} (${skipped[*]:-none})"
  [ "${#kept[@]}" -gt 0 ]
}

# ── interval gate ───────────────────────────────────────────────────────────
if [ "$FORCE" -eq 0 ] && [ "$SCAN_ONLY" -eq 0 ] && [ -f "$LAST" ]; then
  last=$(cat "$LAST" 2>/dev/null || echo 0)
  now=$(date +%s)
  elapsed=$(( now - last ))
  if [ "$elapsed" -lt $(( INTERVAL_DAYS * 86400 )) ]; then
    log "not due yet ($(( (INTERVAL_DAYS * 86400 - elapsed) / 86400 ))d left)"
    exit 0
  fi
fi

# ── scan ────────────────────────────────────────────────────────────────────
{
  echo "=== SkateSpots dependency scan — $(date -Is) ==="
  echo
  echo "=== npm outdated ==="
  npm outdated 2>&1 || true
  echo
  echo "=== npm audit ==="
  npm audit 2>&1 || true
  echo
  echo "=== osv-scanner (OSV database) ==="
  if [ -x "$OSV_BIN" ]; then
    "$OSV_BIN" scan source -L package-lock.json 2>&1 || true
  else
    echo "osv-scanner not installed (expected $OSV_BIN)"
  fi
} | tee "$REPORT"

# The OSV summary line (e.g. "Total 27 packages affected by 59 known ...") is the
# number worth surfacing in notifications; npm audit's counts are narrower.
OSV_SUMMARY="$(grep -E '^Total [0-9]+ packages affected' "$REPORT" | tail -1)"
[ -n "$OSV_SUMMARY" ] || OSV_SUMMARY="OSV scan: no summary (scanner missing?)"

# Rotate diagnostics so the state dir cannot grow forever.
ls -1t "$STATE_DIR"/report-*.txt 2>/dev/null | tail -n +13 | xargs -r rm -f
ls -1dt "$FAILURE_DIR_BASE"/* 2>/dev/null | tail -n +9 | xargs -r rm -rf

if [ "$SCAN_ONLY" -eq 1 ]; then
  log "scan-only; report: $REPORT"
  echo "$OSV_SUMMARY"
  PREV_REPORT="$(ls -1t "$STATE_DIR"/report-*.txt 2>/dev/null | sed -n '2p' | head -1 || true)"
  PREV_OSV=""
  if [ -n "$PREV_REPORT" ]; then
    PREV_OSV="$(grep -E '^Total [0-9]+ packages affected' "$PREV_REPORT" | tail -1 || true)"
  fi
  if [ "$OSV_SUMMARY" != "$PREV_OSV" ]; then
    notify "Vulnerability scan changed: $OSV_SUMMARY (report: $REPORT)"
    timeout 90 "$REPO/scripts/send-ops-alert.sh" "Security scan changed. $OSV_SUMMARY. Check the server report for details." || true
  else
    log "scan unchanged since the previous report — not alerting"
  fi
  exit 0
fi

# ── never mix with the user's uncommitted work ──────────────────────────────
if [ -n "$(git status --porcelain)" ]; then
  log "working tree is dirty — skipping (commit or stash first)"
  exit 0
fi

cp package.json "$STATE_DIR/package.json.bak" 2>/dev/null || true
cp package-lock.json "$STATE_DIR/package-lock.json.bak" 2>/dev/null || true

# Snapshot the pre-update outdated list for the patch-only fallback.
npm outdated --json --depth=0 > "$OUTDATED_JSON" 2>/dev/null || true

# ── apply non-breaking updates ──────────────────────────────────────────────
log "applying non-breaking updates (npm update + npm dedupe + npm audit fix)…"
npm update 2>&1 | tail -3
# Collapse duplicate transitive copies (e.g. @tanstack/query-core, @nostrify/*).
# Without this a minor bump can nest a second copy and break type identity in tsc.
npm dedupe 2>&1 | tail -2
npm audit fix 2>&1 | tail -3
{
  echo; echo "=== post-update npm audit ==="; npm audit 2>&1 || true
  echo; echo "=== post-update osv-scanner ==="
  if [ -x "$OSV_BIN" ]; then "$OSV_BIN" scan source -L package-lock.json 2>&1 || true; fi
} | tee -a "$REPORT"

# ── gate: full tests + boot smoke ───────────────────────────────────────────
log "running tests + smoke…"
if npm test >"$STATE_DIR/test.log" 2>&1 && node scripts/smoke-test.mjs >"$STATE_DIR/smoke.log" 2>&1; then
  if [ -n "$(git status --porcelain package.json package-lock.json)" ]; then
    git add package.json package-lock.json
    git commit -m "chore(deps): scheduled dependency update + audit ($(LC_ALL=C date +%F))" >/dev/null 2>&1 || true
    if [ "$NO_PUSH" -eq 0 ]; then
      # Bound the push: the git remote signs via the remote signer and can stall.
      if timeout 180 git push origin main >"$STATE_DIR/push.log" 2>&1; then
        log "pushed dependency update"
        notify "Dependency update applied, tests pass, pushed. ${OSV_SUMMARY}"
      else
        log "push failed/skipped (see $STATE_DIR/push.log)"
        notify "Dependency update committed locally but the push failed. ${OSV_SUMMARY}"
      fi
    else
      log "committed locally (--no-push)"
      notify "Dependency update applied and committed locally. ${OSV_SUMMARY}"
    fi
  else
    log "no dependency changes to commit"
    notify "Dependency scan complete: nothing to update. ${OSV_SUMMARY}"
  fi
  finish_timer
  rm -f "$FAILURE_FILE"
  log "done"
else
  log "full update FAILED — saving diagnostics and trying the patch-only fallback"
  FAIL_DIR="$FAILURE_DIR_BASE/$(date +%F-%H%M)"
  mkdir -p "$FAIL_DIR"
  cp "$STATE_DIR/test.log" "$STATE_DIR/smoke.log" "$FAIL_DIR"/ 2>/dev/null || true
  cp package.json package-lock.json "$FAIL_DIR"/ 2>/dev/null || true
  printf '%s\n' "full update failed at $(date -Is)" "see $FAIL_DIR/test.log" > "$FAIL_DIR/summary.txt"
  echo "$(date -Is) full update failed — $FAIL_DIR" > "$FAILURE_FILE"

  # Restore the original tree before trying the fallback.
  cp "$STATE_DIR/package.json.bak" package.json 2>/dev/null || true
  cp "$STATE_DIR/package-lock.json.bak" package-lock.json 2>/dev/null || true
  npm i --silent >/dev/null 2>&1 || true

  if patch_only_fallback \
     && npm test >"$STATE_DIR/test.log" 2>&1 \
     && node scripts/smoke-test.mjs >"$STATE_DIR/smoke.log" 2>&1; then
    {
      echo; echo "=== post-fallback npm audit ==="; npm audit 2>&1 || true
      echo; echo "=== post-fallback osv-scanner ==="
      if [ -x "$OSV_BIN" ]; then "$OSV_BIN" scan source -L package-lock.json 2>&1 || true; fi
    } | tee -a "$REPORT"
    OSV_SUMMARY="$(grep -E '^Total [0-9]+ packages affected' "$REPORT" | tail -1)"
    [ -n "$OSV_SUMMARY" ] || OSV_SUMMARY="OSV scan: no summary"

    if [ -n "$(git status --porcelain package.json package-lock.json)" ]; then
      git add package.json package-lock.json
      git commit -m "chore(deps): scheduled dependency update + audit ($(LC_ALL=C date +%F))" >/dev/null 2>&1 || true
      if [ "$NO_PUSH" -eq 0 ]; then
        if timeout 180 git push origin main >"$STATE_DIR/push.log" 2>&1; then
          log "pushed patch-only dependency update"
          notify "Dependency update applied via the patch-only fallback, tests pass, pushed. ${OSV_SUMMARY}"
        else
          log "push failed/skipped (see $STATE_DIR/push.log)"
          notify "Patch-only dependency update committed locally but the push failed. ${OSV_SUMMARY}"
        fi
      else
        log "committed locally (--no-push)"
        notify "Patch-only dependency update applied and committed locally. ${OSV_SUMMARY}"
      fi
    else
      log "patch-only fallback produced no package changes"
      notify "Dependency scan complete: nothing safe to update. ${OSV_SUMMARY}"
    fi
    finish_timer
    rm -f "$FAILURE_FILE"
    log "done (patch-only fallback)"
    exit 0
  fi

  # Both the full and the patch-only update failed: restore and retry in ~3 days.
  mkdir -p "$FAIL_DIR/fallback"
  cp "$STATE_DIR/test.log" "$STATE_DIR/smoke.log" "$FAIL_DIR/fallback"/ 2>/dev/null || true
  cp package.json package-lock.json "$FAIL_DIR/fallback"/ 2>/dev/null || true
  cp "$STATE_DIR/package.json.bak" package.json 2>/dev/null || true
  cp "$STATE_DIR/package-lock.json.bak" package-lock.json 2>/dev/null || true
  npm i --silent >/dev/null 2>&1 || true
  RETRY_DAYS=3; [ "$INTERVAL_DAYS" -gt 4 ] || RETRY_DAYS=1
  echo $(( $(date +%s) - (INTERVAL_DAYS - RETRY_DAYS) * 86400 )) > "$LAST"
  notify "Dependency update REVERTED — full and patch fallback failed (see $FAIL_DIR). ${OSV_SUMMARY}"
  timeout 90 "$REPO/scripts/send-ops-alert.sh" "The dependency update failed and was rolled back. The app is unchanged. The next retry is in about 3 days." || true
  exit 1
fi