"""
What does a shelf map buy over track ids?

Walks the demo camera once, feeds every frame through both the live session and a
ShelfMap, and reports what each counts. Needs a running server on :8077.

    STEPS=58 python code/bench_shelf_map.py
"""
import json, sys, time, urllib.request
sys.path.insert(0, "code")
import cv2, numpy as np
from websockets.sync.client import connect
from shelf_map import ShelfMap

BASE = "http://127.0.0.1:8077"
STEPS = int(__import__("os").environ.get("STEPS", "58"))   # one walk over image6 at pan 0.02

def fetch(step):
    url = f"{BASE}/v1/demo/frame?samples_param=image6&step={step}&width=1280&height=720&quality=70&pan=0.02"
    return urllib.request.urlopen(url, timeout=10).read()

shelf = ShelfMap()
track_ids, frames, detections = set(), 0, 0
started = time.perf_counter()
with connect("ws://127.0.0.1:8077/v1/stream") as ws:
    ws.recv(timeout=15)
    for step in range(STEPS):
        payload = fetch(step)
        frame = cv2.imdecode(np.frombuffer(payload, np.uint8), cv2.IMREAD_COLOR)
        ws.send(step.to_bytes(4, "big") + payload)
        msg = json.loads(ws.recv(timeout=25))
        if msg.get("type") != "result":
            continue
        frames += 1
        tracks = msg["tracks"]
        detections += len(tracks)
        for track in tracks:
            track_ids.add(track["track_id"])
        shelf.update(step, frame, tracks)
        time.sleep(0.05)
elapsed = time.perf_counter() - started

stats = shelf.stats()
print(f"{frames} frames in {elapsed:.1f} s, {detections} detections")
print(f"  track ids seen (what the live session counts): {len(track_ids)}")
print(f"  facings on the shelf map (what is actually there): {stats['facings']}")
print(f"  merges {stats['merges']}, registrations {stats['registrations']}, failures {stats['registration_failures']}")
print(f"  median registration response {stats['median_response']}")
print(f"  facings whose pooled identity agrees with itself >= 0.8: {stats['identity_consistent']}")

facings = [f for f in shelf.facings() if f.observations >= 3]
print(f"\nfacings seen at least 3 times: {len(facings)}")
for facing in sorted(facings, key=lambda f: (f.shelf_box[1], f.shelf_box[0]))[:10]:
    sku, agreement = facing.best_identity()
    tag_sku, tag_agreement = facing.best_tag_identity()
    print(f"  shelf x={facing.shelf_box[0]:>5} y={facing.shelf_box[1]:>5} obs={facing.observations:>2} "
          f"tracks={len(facing.seen_track_ids):>2} identity={sku} ({agreement:.2f}) tag={tag_sku} ({tag_agreement:.2f})")

# render the map so duplicates are visible rather than inferred
import numpy as np
facings = shelf.facings()
xs = [f.shelf_box[0] for f in facings] + [f.shelf_box[2] for f in facings]
ys = [f.shelf_box[1] for f in facings] + [f.shelf_box[3] for f in facings]
x0, x1 = min(xs), max(xs)
y0, y1 = min(ys), max(ys)
scale = 0.6
canvas = np.full((int((y1 - y0) * scale) + 20, int((x1 - x0) * scale) + 20, 3), 20, np.uint8)
for facing in facings:
    bx = [(facing.shelf_box[0] - x0) * scale + 10, (facing.shelf_box[1] - y0) * scale + 10,
          (facing.shelf_box[2] - x0) * scale + 10, (facing.shelf_box[3] - y0) * scale + 10]
    colour = (0, 200, 0) if facing.observations >= 5 else (0, 120, 255)
    cv2.rectangle(canvas, (int(bx[0]), int(bx[1])), (int(bx[2]), int(bx[3])), colour, 1)
cv2.imwrite("/tmp/dorayaki-verify/shelf-map.png", canvas)
print(f"\nmap canvas {canvas.shape[1]}x{canvas.shape[0]} px covering shelf x {x0}..{x1}, y {y0}..{y1}")
from collections import Counter
print("observations per facing:", Counter(min(f.observations, 20) for f in facings).most_common(6))
print("facings per y band (100 px):", Counter((f.shelf_box[1] - y0) // 100 for f in facings).most_common(8))