"""Object detection via YOLOv8, with a graceful no-ML fallback.

When the optional ML stack (torch + ultralytics) is available and enabled, this
detects furniture-like objects so their floor footprints can be treated as
blocked space. Otherwise it returns an empty list and the floor is treated as
unobstructed — the rest of the pipeline still runs.

Grounding DINO is intentionally OUT for the MVP (YOLO covers detection).
"""
from __future__ import annotations

import functools
from dataclasses import dataclass, field
from pathlib import Path

from ..config import settings
from .ml_runtime import ml_available

# COCO labels that occupy floor space and matter for placement.
FURNITURE_LABELS = {
    "bed", "couch", "chair", "dining table", "tv", "potted plant",
    "refrigerator", "toilet", "sink", "bench", "vase", "book", "clock",
}


@dataclass
class Detection:
    label: str
    confidence: float
    box: list[float] = field(default_factory=list)  # [x1, y1, x2, y2] in pixels


@functools.lru_cache(maxsize=1)
def _load_model():
    from ultralytics import YOLO

    return YOLO(settings.YOLO_MODEL)


def detect_objects(image_path: Path, conf_threshold: float = 0.30) -> list[Detection]:
    """Detect objects in an image. Returns [] when ML is unavailable/failing."""
    if not ml_available():
        return []
    try:
        model = _load_model()
        result = model(str(image_path), verbose=False)[0]
        names = result.names
        out: list[Detection] = []
        for b in result.boxes:
            conf = float(b.conf)
            if conf < conf_threshold:
                continue
            label = str(names[int(b.cls)])
            xyxy = [float(v) for v in b.xyxy[0].tolist()]
            out.append(Detection(label=label, confidence=conf, box=xyxy))
        return out
    except Exception:
        # Missing weights / runtime error -> degrade gracefully.
        return []