"""
远程 Ollama LLM 润色服务
通过局域网 HTTP 请求 Gemma4 模型，对交易复盘转写稿进行纪律审判式润色。
"""

from __future__ import annotations

import logging
import time
from typing import Tuple

import requests

from config import (
    HEALTH_CHECK_CACHE_SECONDS,
    HEALTH_CHECK_CONNECT_TIMEOUT,
    HEALTH_CHECK_READ_TIMEOUT,
    MODEL_NAME,
    OLLAMA_TIMEOUT,
    OLLAMA_URL,
    SYSTEM_PROMPT,
)

logger = logging.getLogger(__name__)

# 健康检查短时缓存，避免平板/手机反复打开页面时重复等待
_health_cache: dict = {"ts": 0.0, "ok": False, "msg": ""}


def _build_payload(raw_text: str) -> dict:
    """构造 Ollama /api/chat 非流式请求体。"""
    return {
        "model": MODEL_NAME,
        "messages": [
            {"role": "system", "content": SYSTEM_PROMPT},
            {
                "role": "user",
                "content": (
                    "以下是我的交易复盘录音转写原文，请严格按系统要求润色：\n\n"
                    f"{raw_text}"
                ),
            },
        ],
        "stream": False,
        "options": {
            "temperature": 0.7,
            "num_predict": 4096,
        },
    }


def _extract_content(response_json: dict) -> str:
    """从 Ollama 响应 JSON 中提取 assistant 文本。"""
    # /api/chat 标准格式
    message = response_json.get("message")
    if isinstance(message, dict):
        content = message.get("content", "").strip()
        if content:
            return content

    # 兼容 /api/generate 格式（部分旧版或代理）
    if "response" in response_json:
        content = str(response_json["response"]).strip()
        if content:
            return content

    raise ValueError(f"无法从 Ollama 响应中解析文本内容: {response_json}")


def polish_text(raw_text: str) -> Tuple[bool, str]:
    """
    调用远程 Ollama 对原始转写文本进行润色。

    Args:
        raw_text: Whisper 转写得到的原始口语文本

    Returns:
        (success, polished_text_or_error_message)
    """
    if not raw_text or not raw_text.strip():
        return False, "润色输入为空，请先完成语音识别。"

    payload = _build_payload(raw_text.strip())

    try:
        logger.info("正在请求 Ollama: %s, model=%s", OLLAMA_URL, MODEL_NAME)
        response = requests.post(
            OLLAMA_URL,
            json=payload,
            timeout=OLLAMA_TIMEOUT,
        )
        response.raise_for_status()

        data = response.json()
        polished = _extract_content(data)

        if not polished:
            return False, "Ollama 返回内容为空，请检查模型是否正常加载。"

        logger.info("润色完成，输出字数: %d", len(polished))
        return True, polished

    except requests.exceptions.ConnectTimeout:
        err = (
            f"连接 Ollama 超时（>{OLLAMA_TIMEOUT}s）。"
            f"请确认 {OLLAMA_URL} 可达且 Ollama 服务已启动。"
        )
        logger.error(err)
        return False, err

    except requests.exceptions.ReadTimeout:
        err = (
            f"Ollama 响应超时（>{OLLAMA_TIMEOUT}s）。"
            "模型可能正在加载或生成长度过长，请稍后重试。"
        )
        logger.error(err)
        return False, err

    except requests.exceptions.ConnectionError as exc:
        err = (
            f"无法连接到 Ollama 节点 ({OLLAMA_URL})。"
            "请检查局域网连通性、防火墙及 Ollama 是否监听 0.0.0.0:11434。\n"
            f"详情: {exc}"
        )
        logger.error(err)
        return False, err

    except requests.exceptions.HTTPError as exc:
        status = exc.response.status_code if exc.response is not None else "?"
        body = exc.response.text[:500] if exc.response is not None else ""
        err = (
            f"Ollama HTTP 错误 ({status})。"
            f"请确认模型 `{MODEL_NAME}` 已通过 ollama pull 下载。\n"
            f"响应片段: {body}"
        )
        logger.error(err)
        return False, err

    except ValueError as exc:
        logger.error("Ollama 响应解析失败: %s", exc)
        return False, str(exc)

    except requests.exceptions.RequestException as exc:
        err = f"Ollama 请求异常: {exc}"
        logger.exception(err)
        return False, err

    except Exception as exc:
        err = f"润色过程发生未知错误: {exc}"
        logger.exception(err)
        return False, err


def check_ollama_health(force: bool = False) -> Tuple[bool, str]:
    """
    快速检测 Ollama 节点是否在线（不触发完整推理）。
    默认 2+3 秒超时，结果缓存 30 秒，避免平板首屏长时间白屏。

    Returns:
        (online, message)
    """
    global _health_cache

    now = time.time()
    if (
        not force
        and _health_cache["msg"]
        and (now - _health_cache["ts"]) < HEALTH_CHECK_CACHE_SECONDS
    ):
        return _health_cache["ok"], _health_cache["msg"]

    base_url = OLLAMA_URL.rsplit("/api/", 1)[0]
    timeout = (HEALTH_CHECK_CONNECT_TIMEOUT, HEALTH_CHECK_READ_TIMEOUT)

    try:
        resp = requests.get(f"{base_url}/api/tags", timeout=timeout)
        resp.raise_for_status()
        tags = resp.json().get("models", [])
        model_names = [m.get("name", "") for m in tags]
        if any(MODEL_NAME.split(":")[0] in name for name in model_names):
            msg = f"Ollama 在线，已检测到模型: {MODEL_NAME}"
            ok = True
        else:
            ok = True
            msg = (
                f"Ollama 在线，但未找到模型 {MODEL_NAME}，"
                f"请执行: ollama pull {MODEL_NAME}"
            )
    except requests.exceptions.Timeout:
        ok, msg = False, (
            f"Ollama 检测超时（>{HEALTH_CHECK_READ_TIMEOUT}s）。"
            "页面已加载，可稍后点击「刷新状态」重试。"
        )
    except Exception as exc:
        ok, msg = False, f"Ollama 不可达: {exc}"

    _health_cache.update({"ts": now, "ok": ok, "msg": msg})
    return ok, msg