"""
audio_transcriber.py — Background Audio → SRT Transcription

run_transcription_bg(job_id, audio_bytes, filename, language)
  → Spawns a daemon thread that runs independently of the browser session.
  → Progress saved to job_store → UI polls job_store to show status.
  → Browser close / refresh does NOT stop transcription.

Model: large-v3 only (best accuracy for JA/KO/ZH/TH/HI/EN).
"""

import os
import logging
import tempfile
import threading
from typing import Callable

from job_store import update_job, save_result

logger = logging.getLogger(__name__)

# ── Fixed model: large-v3 ─────────────────────────────────────────────────
MODEL      = "large-v3"
MODEL_DIR  = os.path.join(tempfile.gettempdir(), "whisper_models")

SUPPORTED_LANGUAGES = {
    "auto": "Auto-detect",
    "en":   "English",
    "ja":   "Japanese (日本語)",
    "ko":   "Korean (한국어)",
    "zh":   "Chinese (中文)",
    "th":   "Thai (ภาษาไทย)",
    "hi":   "Hindi (हिन्दी)",
    "fr":   "French",
    "de":   "German",
    "es":   "Spanish",
    "it":   "Italian",
    "pt":   "Portuguese",
    "ru":   "Russian",
    "ar":   "Arabic",
    "vi":   "Vietnamese",
    "id":   "Indonesian",
}


# ─────────────────────────────────────────────────────────────────────────────
# Time formatting
# ─────────────────────────────────────────────────────────────────────────────

def _fmt(seconds: float) -> str:
    ms = int(round((seconds % 1) * 1000))
    s  = int(seconds)
    h, m, s = s // 3600, (s % 3600) // 60, s % 60
    return f"{h:02}:{m:02}:{s:02},{ms:03}"


# ─────────────────────────────────────────────────────────────────────────────
# Segments → SRT string
# ─────────────────────────────────────────────────────────────────────────────

def _segments_to_srt(segments: list) -> str:
    lines = []
    idx   = 1
    for seg in segments:
        text = seg.text.strip()
        if not text:
            continue
        lines += [str(idx), f"{_fmt(seg.start)} --> {_fmt(seg.end)}", text, ""]
        idx += 1
    return "\n".join(lines)


# ─────────────────────────────────────────────────────────────────────────────
# Core transcription (runs in background thread)
# ─────────────────────────────────────────────────────────────────────────────

def _transcribe_worker(
    job_id: str,
    audio_bytes: bytes,
    filename: str,
    language: str,
) -> None:
    """
    Background worker — runs in its own thread, independent of browser session.
    Updates job_store so UI can poll progress even after refresh.
    Thread is daemon=False so it finishes even if main thread exits
    (as long as Docker container is alive).
    """
    tmp_path = None
    try:
        from faster_whisper import WhisperModel

        def cb(frac: float, msg: str) -> None:
            update_job(job_id, progress=frac, status_msg=msg)

        cb(0.0, "Model load လုပ်နေသည် (large-v3, ပထမဆုံးအကြိမ် ကြာနိုင်သည်)…")

        # Write audio to temp file
        suffix   = os.path.splitext(filename)[1] or ".mp3"
        fd, tmp_path = tempfile.mkstemp(suffix=suffix)
        os.write(fd, audio_bytes)
        os.close(fd)

        # Load Whisper large-v3
        model = WhisperModel(
            MODEL,
            device="cpu",
            compute_type="int8",      # fastest CPU quantization
            download_root=MODEL_DIR,
        )
        cb(0.05, f"large-v3 load ပြီး — Transcribing…")

        lang_code = None if language == "auto" else language

        segments_gen, info = model.transcribe(
            tmp_path,
            language=lang_code,
            beam_size=5,
            temperature=0.0,
            condition_on_previous_text=False,  # reduce hallucination
            vad_filter=True,
            vad_parameters={
                "min_silence_duration_ms": 500,
                "threshold": 0.35,
            },
        )

        detected  = info.language
        duration  = max(info.duration or 1.0, 1.0)
        cb(0.08, f"ဘာသာစကား: {detected} · {duration:.0f}s — Segments ထုတ်နေသည်…")

        segments = []
        for seg in segments_gen:
            segments.append(seg)
            pct  = min(0.08 + (seg.end / duration) * 0.90, 0.98)
            elapsed = f"{seg.end:.0f}s/{duration:.0f}s"
            cb(pct, f"Transcribing… {elapsed}  [{len(segments)} segments]")

        if not segments:
            update_job(job_id, status="failed",
                       error="Audio မှ speech မတွေ့ပါ",
                       status_msg="⚠ Speech မတွေ့ပါ")
            return

        cb(0.99, f"SRT ဖန်တီးနေသည်… ({len(segments)} segments)")
        srt_text = _segments_to_srt(segments)
        srt_bytes = srt_text.encode("utf-8")

        save_result(job_id, srt_bytes)
        # save_result already sets status=completed

    except ImportError:
        err = ("faster-whisper မထည့်ရသေးပါ — "
               "requirements.txt မှာ faster-whisper>=1.0.0 ပါပြီး redeploy လုပ်ပါ")
        update_job(job_id, status="failed", error=err, status_msg="❌ Import Error")

    except Exception as exc:
        logger.exception("Transcription worker error")
        update_job(job_id, status="failed",
                   error=str(exc)[:400],
                   status_msg=f"❌ Error: {exc}")
    finally:
        if tmp_path and os.path.exists(tmp_path):
            try: os.unlink(tmp_path)
            except Exception: pass


# ─────────────────────────────────────────────────────────────────────────────
# Public API — launch background thread
# ─────────────────────────────────────────────────────────────────────────────

def run_transcription_bg(
    job_id: str,
    audio_bytes: bytes,
    filename: str,
    language: str = "auto",
) -> threading.Thread:
    """
    Spawn a background thread to transcribe audio.
    The thread runs independently of the Streamlit browser session.
    Browser close / page refresh does NOT stop the transcription.

    Returns the Thread object (caller can ignore it).
    """
    update_job(job_id, status="running", progress=0.0,
               status_msg="Background thread started…")

    t = threading.Thread(
        target=_transcribe_worker,
        args=(job_id, audio_bytes, filename, language),
        daemon=True,   # daemon=True: thread exits with container (correct behavior)
        name=f"whisper-{job_id}",
    )
    t.start()
    logger.info("Transcription thread started: job=%s file=%s", job_id, filename)
    return t