#!/usr/bin/env bash
# probe_x2.sh — read-only discovery for an AGIBOT X2 (or any unknown robot).
#
#   tools/probe_x2.sh <ip> <ssh-user>
#   ./fleet_install.sh probe x2 <ip> --user <ssh-user>    # same, via the installer
#
# WHY THIS EXISTS
# The X2 exposes no fixed topic contract, so sanad_api_x2.py reads every topic
# name and field path from .env instead of hard-coding them. This script finds
# the real values to put there.
#
# It ONLY READS: no install, no write, no package manager, no motion. Safe to run
# on a live robot. Everything it prints maps to a variable in agents/x2/.env.example.
set -uo pipefail

# No default login user is assumed for the X2 — its login account is a property
# of the robot, not something to guess. Pass it explicitly.
IP="${1:-}"; USER_="${2:-}"
[ -n "$IP" ] && [ -n "$USER_" ] \
  || { echo "usage: $0 <ip> <ssh-user>" >&2
       echo "  the X2's login user is not assumed — pass the account you SSH in as" >&2
       exit 1; }

SSH_OPTS=(-o BatchMode=yes -o ConnectTimeout=8 -o StrictHostKeyChecking=accept-new)
[ -n "${FLEET_SSH_KEY:-}" ] && SSH_OPTS+=(-i "$FLEET_SSH_KEY")
ssh -n "${SSH_OPTS[@]}" "$USER_@$IP" true 2>/dev/null \
  || { echo "ERROR: cannot SSH to $USER_@$IP (key auth?)" >&2; exit 1; }

echo "=============================================================="
echo " AGIBOT X2 probe — $USER_@$IP"
echo "=============================================================="

# NOTE: no -n here. The probe body is piped in on stdin, and -n would redirect
# stdin from /dev/null, so `bash -s` would read an empty script and print nothing.
ssh "${SSH_OPTS[@]}" "$USER_@$IP" 'bash -s' <<'REMOTE'
set -uo pipefail
sec(){ echo; echo "── $* ──────────────────────────────────────────"; }
have(){ command -v "$1" >/dev/null 2>&1; }

sec "1. PLATFORM              -> software.* / firmware.* in telemetry"
. /etc/os-release 2>/dev/null && echo "os:        ${PRETTY_NAME:-?}  (VERSION_ID=${VERSION_ID:-?})"
echo "kernel:    $(uname -r)"
echo "arch:      $(uname -m)                 # arm64 => build natively on the robot"
for f in /sys/firmware/devicetree/base/model /proc/device-tree/model; do
  [ -r "$f" ] && { echo "board:     $(tr -d '\0' < "$f")"; break; }
done
[ -r /etc/nv_tegra_release ] && echo "l4t:       $(head -1 /etc/nv_tegra_release)"
echo "python3:   $(python3 -V 2>&1)"

sec "2. DEPLOY PREREQS        -> can fleet_install.sh work here?"
if have docker; then
  echo "docker:    $(docker --version 2>&1)"
  docker ps >/dev/null 2>&1 \
    && echo "           OK  '$USER' can talk to dockerd (no sudo needed)" \
    || echo "           !!  '$USER' CANNOT use docker — add to the docker group"
else
  echo "docker:    !! NOT INSTALLED — required"
fi
systemctl --user show-environment >/dev/null 2>&1 \
  && echo "systemd:   OK  user bus available (user-level service, no sudo)" \
  || echo "systemd:   !!  no user bus — auto-start needs another mechanism"
echo "linger:    $(loginctl show-user "$USER" 2>/dev/null | grep -i linger || echo '?')"

sec "3. NETWORK               -> MAC_INTERFACE, REMOTE_HOST"
ip -o -4 addr show 2>/dev/null | awk '{printf "  %-10s %s\n", $2, $4}'
echo "  MACs:"
for n in /sys/class/net/*; do
  i="$(basename "$n")"; [ "$i" = lo ] && continue
  echo "    $(printf '%-10s' "$i") $(cat "$n/address" 2>/dev/null)"
done

sec "4. SERIAL NUMBER         -> SN (keys the robot on the fleet server)"
for f in /sys/class/dmi/id/product_serial /proc/device-tree/serial-number \
         /etc/machine-id /etc/robot_sn /etc/agibot/sn; do
  [ -r "$f" ] && echo "  $f = $(tr -d '\0' < "$f" 2>/dev/null | head -1)"
done
echo "  (prefer the serial printed on the robot / in AgiBot's tooling over machine-id)"

sec "5. ROS 2                 -> X2_SOURCE=ros2, X2_TOPIC_*, ROS_DOMAIN_ID"
if [ -d /opt/ros ]; then
  echo "distros:   $(ls /opt/ros 2>/dev/null | tr '\n' ' ')"
  D="${ROS_DISTRO:-$(ls /opt/ros 2>/dev/null | head -1)}"
  echo "using:     $D    ROS_DOMAIN_ID=${ROS_DOMAIN_ID:-0}"
  if [ -f "/opt/ros/$D/setup.bash" ]; then
    # ROS setup.bash reads unbound vars (AMENT_TRACE_SETUP_FILES, COLCON_TRACE,
    # …). Under `set -u` that is FATAL and kills this whole probe silently, so
    # relax nounset across the source and restore it after.
    set +u
    # shellcheck disable=SC1090
    . "/opt/ros/$D/setup.bash" 2>/dev/null || true
    echo "--- ros2 topic list (5s) ---"
    timeout 5 ros2 topic list 2>/dev/null | sed 's/^/  /' || echo "  (none / timed out)"
    echo "--- types for likely telemetry topics ---"
    for t in $(timeout 5 ros2 topic list 2>/dev/null | grep -Ei 'batt|power|bms|joint|motor|odom|imu|state|fsm|mode'); do
      printf "  %-42s %s\n" "$t" "$(timeout 3 ros2 topic info "$t" 2>/dev/null | awk -F': ' '/Type/{print $2; exit}')"
    done
    set -u
  fi
else
  echo "  no /opt/ros — the ros2 backend is unavailable unless AgiBot ships one"
fi

sec "6. AIMRT / VENDOR SDK    -> X2_SOURCE=aimrt (or the ros2 bridge)"
python3 -c 'import aimrt_py, sys; print("  aimrt_py importable:", aimrt_py.__file__)' 2>/dev/null \
  || echo "  aimrt_py not importable for this python3"
ls -d /opt/aimrt /opt/agibot /usr/local/aimrt ~/aimrt* ~/agibot* 2>/dev/null | sed 's/^/  path: /'
python3 -c 'import agibot, sys; print("  agibot module:", agibot.__file__)' 2>/dev/null || true

sec "7. LISTENING PORTS       -> X2_STATE_URL (http backend), REMOTE_PORTS"
(ss -lntp 2>/dev/null || netstat -lntp 2>/dev/null) | sed 's/^/  /' | head -30
echo "  probe a candidate:  curl -s http://127.0.0.1:<port>/  | head -c 400"

sec "8. RUNNING WORKLOAD      -> PROJECT_LOG_CONTAINER, what publishes state"
have docker && docker ps --format '  {{.Names}}  {{.Status}}  ({{.Image}})' 2>/dev/null
echo "  top processes:"
ps -eo comm,pcpu --sort=-pcpu 2>/dev/null | head -12 | sed 's/^/    /'

sec "9. MAPS                  -> MAPS_DIR / EXTRA_MAP_DIRS"
find "$HOME" -maxdepth 5 \( -name '*.pgm' -o -name 'map*.yaml' -o -name '*.db' -o -name '*.posegraph' \) \
     2>/dev/null | head -20 | sed 's/^/  /'
echo "  (the agent wants the DIRECTORY holding a matching .pgm + .yaml pair, or a .db)"

sec "10. DISK                 -> storage.* in telemetry"
df -h / 2>/dev/null | sed 's/^/  /'
REMOTE

cat <<'NEXT'

==============================================================
 NEXT STEPS
==============================================================
 1. Pick a backend from what appeared above:
      section 5 found battery/joint/odom topics  -> X2_SOURCE=ros2
      section 7 found a vendor JSON API          -> X2_SOURCE=http
      neither                                    -> X2_SOURCE=none
                                                    (identity + maps + logs
                                                     still work; battery null)

 2. Install with the real serial from section 4:

      ./fleet_install.sh install x2 <ip> --sn <serial> \
          --token <device-token> --server-url https://eco.yslootahrobotics.com

    ros2 backend? build against a ROS base matching the robot's distro:
      X2_BASE_IMAGE=ros:humble-ros-base ./fleet_install.sh install x2 <ip> ...

 3. Put the discovered names in the robot's .env and restart:

      ssh <user>@<ip> 'nano ~/sanad_api_x2/.env'          # X2_TOPIC_* / X2_FIELD_*
      ssh <user>@<ip> 'systemctl --user restart sanad-api-x2'

 4. Confirm what it is actually sending:

      ./fleet_install.sh data x2 <ip>

 5. VERIFY THE TWO UNIT CONVENTIONS on real hardware:
      - battery % looks right          -> else set X2_SOC_SCALE=fraction|percent
      - "charging" true while charging -> else set X2_CURRENT_SIGN=-1
NEXT