"""
Vision Sampler — frame capture, normalization, entropy, motion detection

Samples camera frames at adaptive rate, computes evidence metrics:
- Shannon entropy (information density)
- Perceptual hash (for frame dedup / novelty)
- Motion score (frame delta vs previous)
"""

import base64
import hashlib
import io
import time
import numpy as np
from PIL import Image
from .stream_state import FrameEvidence, sha256_bytes


def parse_data_url(data_url: str) -> bytes:
    if "," in data_url and data_url.startswith("data:"):
        data_url = data_url.split(",", 1)[1]
    return base64.b64decode(data_url)


def normalize_jpeg(image_bytes: bytes, max_side: int = 960) -> tuple:
    im = Image.open(io.BytesIO(image_bytes)).convert("RGB")
    scale = min(1.0, max_side / max(im.width, im.height))
    if scale < 1:
        im = im.resize((int(im.width * scale), int(im.height * scale)))
    out = io.BytesIO()
    im.save(out, format="JPEG", quality=82, optimize=True)
    return out.getvalue(), im


def image_entropy(im: Image.Image) -> float:
    g = im.convert("L").resize((128, 128))
    arr = np.asarray(g, dtype=np.uint8)
    hist = np.bincount(arr.flatten(), minlength=256).astype(np.float64)
    probs = hist / max(1, hist.sum())
    probs = probs[probs > 0]
    return float(-(probs * np.log2(probs)).sum())


def average_hash(im: Image.Image, size: int = 8) -> str:
    g = im.convert("L").resize((size, size))
    arr = np.asarray(g, dtype=np.float32)
    mean = arr.mean()
    bits = (arr > mean).astype(np.uint8).flatten()
    value = 0
    for bit in bits:
        value = (value << 1) | int(bit)
    return f"{value:016x}"


def phash_delta(a: str | None, b: str) -> float:
    if not a:
        return 1.0
    x = int(a, 16)
    y = int(b, 16)
    return bin(x ^ y).count("1") / 64.0


def process_frame(data_url: str, previous_phash: str | None = None) -> FrameEvidence:
    raw = parse_data_url(data_url)
    jpeg, im = normalize_jpeg(raw)
    ph = average_hash(im)
    entropy = image_entropy(im)
    motion = phash_delta(previous_phash, ph)

    return FrameEvidence(
        ts=time.time(),
        sha256=sha256_bytes(jpeg),
        phash=ph,
        entropy=entropy,
        motion_score=motion,
        width=im.width,
        height=im.height,
        jpeg_b64=base64.b64encode(jpeg).decode("ascii"),
    )


def should_increase_fps(motion_scores: list, threshold: float = 0.15) -> bool:
    if len(motion_scores) < 3:
        return False
    recent = motion_scores[-3:]
    return sum(recent) / len(recent) > threshold