"""yuanqi 本地 ASR 适配器 — ``POST /v1/audio/transcriptions`` (multipart)。

与 MiMo 走 ``/chat/completions`` + ``input_audio`` 不同，元炁本地推理引擎
（yuanqi）对外暴露的是 OpenAI Whisper 兼容的 multipart 端点::

    POST {base_url}/v1/audio/transcriptions
    Content-Type: multipart/form-data
    Authorization: Bearer {api_key}

    file: <audio bytes>
    language: zh  (可选)

响应::

    {"text": "转写后的文本"}

模型实际为 FunASR SenseVoiceSmall，转写文本已在 yuanqi 侧剥离了
``<|zh|><|NEUTRAL|><|Speech|><|withitn|>`` 等内部标签，此处无需二次清洗。
"""

from __future__ import annotations

import hashlib
import time

import httpx

from app.ai_provider.audio.base import (
    ASRProtocol,
    ASRRequest,
    ASRResponse,
    AudioAdapterBase,
    AudioServiceError,
)
from app.core.logging import logger

# multipart 使用的 MIME 映射（与 yuanqi _SUPPORTED_FORMATS 白名单对齐）
_MIME_MAP = {
    "wav": "audio/wav",
    "mp3": "audio/mpeg",
    "m4a": "audio/mp4",
    "flac": "audio/flac",
    "ogg": "audio/ogg",
    "opus": "audio/opus",
    "aac": "audio/aac",
    "webm": "audio/webm",
}

# yuanqi ASR 路径前缀（与 src/endpoints/asr.py 的 router 前缀一致）
_ASR_PATH = "/v1/audio/transcriptions"

# ASR 缓存：默认最多 128 条，TTL 1 小时（与 MiMo ASR 适配器一致）
_ASR_CACHE_MAX_SIZE = 128
_ASR_CACHE_TTL_SECONDS = 3600


class _ASRCacheEntry:
    __slots__ = ("text", "stored_at")

    def __init__(self, text: str) -> None:
        self.text = text
        self.stored_at = time.monotonic()


_asr_cache: dict[str, _ASRCacheEntry] = {}


def _asr_cache_key(model: str, audio_data: bytes, language: str, fmt: str) -> str:
    """生成 ASR 缓存键：SHA256(model + audio_bytes + language + format)。"""
    raw = f"{model}\0{language}\0{fmt}\0"
    raw_bytes = raw.encode() + audio_data
    return hashlib.sha256(raw_bytes).hexdigest()


def _asr_cache_get(key: str) -> str | None:
    entry = _asr_cache.get(key)
    if entry is None:
        return None
    if time.monotonic() - entry.stored_at > _ASR_CACHE_TTL_SECONDS:
        del _asr_cache[key]
        return None
    return entry.text


def _asr_cache_set(key: str, text: str) -> None:
    if len(_asr_cache) >= _ASR_CACHE_MAX_SIZE:
        sorted_keys = sorted(_asr_cache.keys(), key=lambda k: _asr_cache[k].stored_at)
        for old_key in sorted_keys[: _ASR_CACHE_MAX_SIZE // 2]:
            del _asr_cache[old_key]
    _asr_cache[key] = _ASRCacheEntry(text)


class YuanqiASRWhisper(AudioAdapterBase, ASRProtocol):
    """yuanqi 本地 ASR 适配器，走 OpenAI Whisper 兼容的 multipart 端点。

    将音频字节以 ``multipart/form-data`` 的 ``file`` 字段上传到 yuanqi 的
    ``/v1/audio/transcriptions``，从 JSON 响应中提取转写文本。
    """

    _adapter_type = "asr"
    _default_timeout = 300

    def __init__(
        self,
        *,
        base_url: str,
        api_key: str,
        model: str = "SenseVoiceSmall",
        request_timeout_seconds: int = 300,
    ) -> None:
        self._base_url = base_url.rstrip("/")
        self._api_key = api_key
        self._model = model
        self._timeout = request_timeout_seconds
        self._client: httpx.AsyncClient | None = None

    # ── 客户端 ───────────────────────────────────────────────────────────

    @property
    def _http(self) -> httpx.AsyncClient:
        if self._client is None or self._client.is_closed:
            self._client = httpx.AsyncClient(timeout=httpx.Timeout(self._timeout))
        return self._client

    async def close(self) -> None:
        if self._client is not None and not self._client.is_closed:
            await self._client.aclose()

    # ── ASRProtocol ──────────────────────────────────────────────────────

    async def transcribe(self, request: ASRRequest) -> ASRResponse:
        """通过 ``POST /v1/audio/transcriptions`` 转写音频。"""
        # ── 缓存查找 ──────────────────────────────────────────────────
        cache_key = _asr_cache_key(
            self._model, request.audio_data, request.language or "auto", request.format
        )
        cached = _asr_cache_get(cache_key)
        if cached is not None:
            logger.info("audio_asr.cache_hit", model=self._model, text_length=len(cached))
            return ASRResponse(text=cached)

        # ── 构造 multipart 请求 ───────────────────────────────────────
        mime = _MIME_MAP.get(request.format, "audio/wav")
        filename = f"audio.{request.format}"
        files = {"file": (filename, request.audio_data, mime)}
        data: dict[str, str] = {}
        if request.language:
            data["language"] = request.language

        # 本地 yuanqi 引擎通常无需鉴权；仅当配置了 api_key 才带 Authorization 头
        headers: dict[str, str] = {}
        if self._api_key:
            headers["Authorization"] = f"Bearer {self._api_key}"

        url = f"{self._base_url}{_ASR_PATH}"

        logger.info(
            "audio_asr.request",
            model=self._model,
            format=request.format,
            language=request.language,
            url=url,
            size_bytes=len(request.audio_data),
        )

        # ── API 调用 ────────────────────────────────────────────────
        try:
            response = await self._http.post(url, files=files, data=data, headers=headers)
        except httpx.TimeoutException as exc:
            raise AudioServiceError(
                status_code=504,
                code="audio_asr_timeout",
                message="ASR request timed out calling yuanqi transcriptions.",
            ) from exc
        except httpx.RequestError as exc:
            raise AudioServiceError(
                status_code=503,
                code="audio_asr_connection_error",
                message=f"ASR connection error: {exc}",
            ) from exc

        if response.status_code != 200:
            body = response.text[:500]
            raise AudioServiceError(
                status_code=response.status_code,
                code="audio_asr_api_error",
                message=f"ASR API returned {response.status_code}: {body}",
            )

        # 解析响应，提取转写文本
        text = self._extract_text(response.json())
        if not text:
            raise AudioServiceError(
                status_code=500,
                code="audio_asr_empty_response",
                message="ASR API returned empty transcription.",
            )

        # ── 写入缓存 ────────────────────────────────────────────────
        _asr_cache_set(cache_key, text)

        logger.info(
            "audio_asr.success",
            model=self._model,
            text_length=len(text),
            cache_size=len(_asr_cache),
        )
        return ASRResponse(text=text)

    # ── 辅助方法 ──────────────────────────────────────────────────────────

    @staticmethod
    def _extract_text(body: dict) -> str:
        """从 Whisper 兼容响应中提取转写文本。"""
        try:
            text = body.get("text")
            if text is None:
                raise AudioServiceError(
                    status_code=500,
                    code="audio_asr_no_text",
                    message="ASR response has no 'text' field.",
                )
            return text or ""
        except AudioServiceError:
            raise
        except Exception as exc:
            raise AudioServiceError(
                status_code=500,
                code="audio_asr_response_parse_error",
                message=f"Failed to parse ASR response: {exc}",
            ) from exc