"""
LectureLens — Alert Engine
Reads thresholds.yaml once at startup and exposes generate_alerts().
"""

from __future__ import annotations

import math
from pathlib import Path
from typing import Any, Dict, List

import yaml

from app.schemas import Alert, AlertCategory, AlertSeverity, AudioMetrics, VideoMetrics


# ── Loader ────────────────────────────────────────────────────────────────────

_thresholds_cache: Dict[str, Any] | None = None


def load_thresholds(path: str = "thresholds.yaml") -> Dict[str, Any]:
    global _thresholds_cache
    if _thresholds_cache is None:
        config_path = Path(path)
        if not config_path.exists():
            raise FileNotFoundError(f"Thresholds config not found at: {config_path}")
        with config_path.open("r", encoding="utf-8") as f:
            _thresholds_cache = yaml.safe_load(f)
    return _thresholds_cache


def reload_thresholds(path: str = "thresholds.yaml") -> None:
    """Force a reload (useful for tests or live config updates)."""
    global _thresholds_cache
    _thresholds_cache = None
    load_thresholds(path)


# ── Internal helpers ──────────────────────────────────────────────────────────

def _severity_from_ratio(ratio: float) -> AlertSeverity:
    """
    ratio = how far beyond the threshold, relative to the threshold magnitude.
      ratio > 0.5  → critical
      else         → warning
    """
    if ratio > 0.5:
        return AlertSeverity.critical
    return AlertSeverity.warning


def _make_alert(
    severity: AlertSeverity,
    category: AlertCategory,
    kpi: str,
    message: str,
    suggested_fix: str,
    timestamp_range=None,
) -> Alert:
    return Alert(
        severity=severity,
        category=category,
        kpi=kpi,
        message=message,
        suggested_fix=suggested_fix,
        timestamp_range=timestamp_range,
    )


# ── Audio alerts ──────────────────────────────────────────────────────────────

def _audio_alerts(metrics: AudioMetrics, cfg: Dict[str, Any]) -> List[Alert]:
    alerts: List[Alert] = []
    ac = cfg.get("audio", {})

    # integrated_loudness_lufs
    if metrics.integrated_loudness_lufs is not None:
        lufs_cfg = ac.get("integrated_loudness_lufs", {})
        target = lufs_cfg.get("target", -14)
        tolerance = lufs_cfg.get("tolerance", 2)
        val = metrics.integrated_loudness_lufs
        diff = abs(val - target)
        if diff > tolerance:
            ratio = (diff - tolerance) / max(abs(target), 1)
            sev = _severity_from_ratio(ratio)
            alerts.append(_make_alert(
                sev, AlertCategory.audio, "integrated_loudness_lufs",
                f"Integrated loudness is {val:.1f} LUFS — {diff - tolerance:.1f} LU "
                f"{'below' if val < target else 'above'} the target of {target} LUFS.",
                "Adjust microphone gain or apply loudness normalisation before uploading."
            ))

    # true_peak_dbtp
    if metrics.true_peak_dbtp is not None:
        tp_max = ac.get("true_peak_dbtp", {}).get("max", -1.0)
        val = metrics.true_peak_dbtp
        if val > tp_max:
            alerts.append(_make_alert(
                AlertSeverity.critical, AlertCategory.audio, "true_peak_dbtp",
                f"True peak is {val:.1f} dBTP — exceeds the {tp_max} dBTP ceiling.",
                "Lower recording level or apply a true-peak limiter."
            ))

    # clipped_samples_count
    if metrics.clipped_samples_count > 0:
        max_clips = ac.get("clipped_samples_count", {}).get("max", 0)
        alerts.append(_make_alert(
            AlertSeverity.critical, AlertCategory.audio, "clipped_samples_count",
            f"{metrics.clipped_samples_count:,} clipped sample(s) detected — audio is distorted.",
            "Reduce microphone input level to prevent clipping."
        ))

    # snr_db
    if metrics.snr_db is not None:
        snr_min = ac.get("snr_db", {}).get("min", 20)
        val = metrics.snr_db
        if val < snr_min:
            diff = snr_min - val
            sev = AlertSeverity.critical if diff > 10 else AlertSeverity.warning
            alerts.append(_make_alert(
                sev, AlertCategory.audio, "snr_db",
                f"SNR is {val:.1f} dB — below the {snr_min} dB minimum.",
                "Use a directional microphone, reduce background noise, or move to a quieter room."
            ))

    # loudness_range_lu
    if metrics.loudness_range_lu is not None:
        lra_cfg = ac.get("loudness_range_lu", {})
        lra_min = lra_cfg.get("min", 4)
        lra_max = lra_cfg.get("max", 15)
        val = metrics.loudness_range_lu
        if val < lra_min:
            alerts.append(_make_alert(
                AlertSeverity.info, AlertCategory.audio, "loudness_range_lu",
                f"Loudness range is very low ({val:.1f} LU) — audio may sound over-compressed.",
                "Avoid applying heavy dynamic compression to the recording."
            ))
        elif val > lra_max:
            alerts.append(_make_alert(
                AlertSeverity.warning, AlertCategory.audio, "loudness_range_lu",
                f"Loudness range is {val:.1f} LU — high variation indicates inconsistent mic distance.",
                "Keep a consistent distance from the microphone throughout the lecture."
            ))

    # silence_segments
    silence_max = ac.get("silence_max_duration_sec", 20)
    if metrics.longest_silence_seconds > silence_max:
        count = sum(1 for s in metrics.silence_segments if (s.end - s.start) > silence_max)
        alerts.append(_make_alert(
            AlertSeverity.warning, AlertCategory.audio, "silence_segments",
            f"Detected {count} long silence segment(s). Total silence: {metrics.total_silence_seconds:.1f}s, Longest: {metrics.longest_silence_seconds:.1f}s.",
            "Check for accidental muting or long recording gaps.",
        ))

    # dnsmos_ovrl
    if metrics.dnsmos_ovrl is not None:
        dnsmos_min = ac.get("dnsmos_ovrl", {}).get("min", 3.0)
        val = metrics.dnsmos_ovrl
        if val < dnsmos_min:
            diff = dnsmos_min - val
            sev = AlertSeverity.critical if diff > 1.0 else AlertSeverity.warning
            alerts.append(_make_alert(
                sev, AlertCategory.audio, "dnsmos_ovrl",
                f"DNSMOS overall speech quality score is {val:.2f}/5 — below the {dnsmos_min} threshold.",
                "Improve acoustic environment, use a better microphone, or apply noise suppression."
            ))

    return alerts


# ── Video alerts ──────────────────────────────────────────────────────────────

def _video_alerts(metrics: VideoMetrics, cfg: Dict[str, Any]) -> List[Alert]:
    alerts: List[Alert] = []
    vc = cfg.get("video", {})

    # brightness
    if metrics.avg_brightness is not None:
        br_cfg = vc.get("brightness", {})
        br_min = br_cfg.get("min", 80)
        br_max = br_cfg.get("max", 180)
        val = metrics.avg_brightness
        if val < br_min:
            alerts.append(_make_alert(
                AlertSeverity.warning, AlertCategory.video, "avg_brightness",
                f"Average brightness is {val:.1f}/255 — video is too dark.",
                "Increase room lighting or adjust camera exposure settings."
            ))
        elif val > br_max:
            alerts.append(_make_alert(
                AlertSeverity.warning, AlertCategory.video, "avg_brightness",
                f"Average brightness is {val:.1f}/255 — video is overexposed.",
                "Reduce direct lighting on the speaker or lower camera exposure."
            ))

    # sharpness
    if metrics.avg_sharpness_laplacian is not None:
        sharp_min = vc.get("sharpness_laplacian", {}).get("min", 100)
        val = metrics.avg_sharpness_laplacian
        if val < sharp_min:
            sev = AlertSeverity.critical if val < sharp_min * 0.5 else AlertSeverity.warning
            alerts.append(_make_alert(
                sev, AlertCategory.video, "avg_sharpness_laplacian",
                f"Image sharpness (Laplacian variance) is {val:.1f} — video appears blurry or out-of-focus.",
                "Clean the camera lens, ensure correct focus, and avoid camera movement."
            ))

    # dropped frames
    if metrics.dropped_frames_ratio is not None:
        df_max = vc.get("dropped_frames_ratio", {}).get("max", 0.01)
        val = metrics.dropped_frames_ratio
        if val > df_max:
            alerts.append(_make_alert(
                AlertSeverity.warning, AlertCategory.video, "dropped_frames_ratio",
                f"Dropped frame ratio is {val * 100:.2f}% — exceeds the {df_max * 100:.1f}% limit.",
                "Check recording hardware performance and storage write speed."
            ))

    # frozen segments
    freeze_max = vc.get("freeze_max_duration_sec", 5)
    if metrics.longest_frozen_seconds > freeze_max:
        count = sum(1 for s in metrics.frozen_segments if (s.end - s.start) > freeze_max)
        alerts.append(_make_alert(
            AlertSeverity.critical, AlertCategory.video, "frozen_segments",
            f"Detected {count} frozen video segment(s). Total frozen: {metrics.total_frozen_seconds:.1f}s, Longest: {metrics.longest_frozen_seconds:.1f}s.",
            "Check network stability and recording settings.",
        ))

    # black segments
    black_max = vc.get("black_max_duration_sec", 5)
    if metrics.longest_black_seconds > black_max:
        count = sum(1 for s in metrics.black_segments if (s.end - s.start) > black_max)
        alerts.append(_make_alert(
            AlertSeverity.warning, AlertCategory.video, "black_segments",
            f"Detected {count} black screen segment(s). Total black: {metrics.total_black_seconds:.1f}s, Longest: {metrics.longest_black_seconds:.1f}s.",
            "Check for accidental screen-sharing stops or camera disconnections.",
        ))

    # compression artifacts
    if metrics.compression_artifact_score is not None:
        ca_max = vc.get("compression_artifact_score", {}).get("max", 0.3)
        val = metrics.compression_artifact_score
        if val > ca_max:
            alerts.append(_make_alert(
                AlertSeverity.info, AlertCategory.video, "compression_artifact_score",
                f"Compression artefact score is {val:.2f} — noticeable blocking/banding.",
                "Use a higher video bitrate in Zoom recording settings."
            ))

    return alerts


# ── Public API ────────────────────────────────────────────────────────────────

def generate_alerts(
    metrics: AudioMetrics | VideoMetrics,
    media_type: str,
    thresholds_path: str = "thresholds.yaml",
) -> List[Alert]:
    """
    Compare metrics against thresholds and return a list of Alert objects.
    """
    cfg = load_thresholds(thresholds_path)
    if media_type == "audio":
        return _audio_alerts(metrics, cfg)  # type: ignore[arg-type]
    return _video_alerts(metrics, cfg)  # type: ignore[arg-type]


# ── Composite score helpers ───────────────────────────────────────────────────

def compute_audio_score(metrics: AudioMetrics) -> float:
    """
    Weighted composite score (0–1) for audio quality.
    Higher = better.
    """
    scores = []

    # DNSMOS: weight 0.35
    if metrics.dnsmos_ovrl is not None:
        scores.append((metrics.dnsmos_ovrl / 5.0, 0.35))

    # SNR: weight 0.25
    if metrics.snr_db is not None:
        snr_score = min(metrics.snr_db / 40.0, 1.0)  # 40 dB = perfect
        scores.append((snr_score, 0.25))

    # Loudness: weight 0.20 — penalty for distance from -14 LUFS
    if metrics.integrated_loudness_lufs is not None:
        dist = abs(metrics.integrated_loudness_lufs - (-14))
        loudness_score = max(0.0, 1.0 - dist / 14.0)
        scores.append((loudness_score, 0.20))

    # Clipping penalty: weight 0.10
    clip_score = 1.0 if metrics.clipped_samples_count == 0 else 0.0
    scores.append((clip_score, 0.10))

    # True peak: weight 0.10
    if metrics.true_peak_dbtp is not None:
        tp_score = 1.0 if metrics.true_peak_dbtp <= -1.0 else 0.0
        scores.append((tp_score, 0.10))

    if not scores:
        return 0.5  # default when no data

    total_weight = sum(w for _, w in scores)
    weighted_sum = sum(s * w for s, w in scores)
    return round(weighted_sum / total_weight, 3)


def compute_video_score(metrics: VideoMetrics) -> float:
    """
    Weighted composite score (0–1) for video quality.
    Higher = better.
    """
    scores = []

    # Sharpness: weight 0.45
    if metrics.avg_sharpness_laplacian is not None:
        sharp_score = min(metrics.avg_sharpness_laplacian / 300.0, 1.0)
        scores.append((sharp_score, 0.45))

    # Brightness: weight 0.25
    if metrics.avg_brightness is not None:
        br = metrics.avg_brightness
        # Gaussian-like penalty centred on 130
        br_score = max(0.0, 1.0 - abs(br - 130) / 80.0)
        scores.append((br_score, 0.25))

    # Dropped frames: weight 0.20
    if metrics.dropped_frames_ratio is not None:
        df_score = max(0.0, 1.0 - metrics.dropped_frames_ratio * 50)
        scores.append((df_score, 0.20))

    # Frozen segments penalty: weight 0.10
    freeze_score = 1.0 if not metrics.frozen_segments else 0.3
    scores.append((freeze_score, 0.10))

    if not scores:
        return 0.5

    total_weight = sum(w for _, w in scores)
    weighted_sum = sum(s * w for s, w in scores)
    return round(weighted_sum / total_weight, 3)