import json
import logging
import re
from typing import List, Optional
from pydantic import BaseModel

from openai import OpenAI

from config import config

logger = logging.getLogger(__name__)

client = OpenAI(
    api_key=config.llm.api_key or None,
    base_url=config.llm.base_url if config.llm.base_url else None,
)


class Entity(BaseModel):
    name: str
    entity_type: str
    description: str = ""


class Relation(BaseModel):
    subject: str
    subject_type: str
    predicate: str
    object: str
    object_type: str
    description: str = ""


class ActionItem(BaseModel):
    task: str
    assignee: str = ""
    deadline: str = ""
    status: str = "待办"
    priority: str = "中"


class Decision(BaseModel):
    content: str
    proposer: str = ""
    status: str = "已决"


class MeetingMetric(BaseModel):
    metric_name: str
    value: str
    target: str = ""
    owner: str = ""
    trend: str = ""


class MeetingExtraction(BaseModel):
    title: str
    date: str = ""
    participants: List[str] = []
    agenda: List[str] = []
    entities: List[Entity] = []
    relations: List[Relation] = []
    action_items: List[ActionItem] = []
    decisions: List[Decision] = []
    metrics: List[MeetingMetric] = []
    summary: str = ""


EXTRACTION_SYSTEM_PROMPT = """
你是一个专业的会议纪要信息抽取专家。你的任务是从中文会议记录中抽取结构化信息，并严格按照要求的JSON格式返回。

## 抽取内容

### 1. 实体
- 人物：参会人员、提及的人员
- 组织/部门：公司、部门、团队
- 项目/任务：正在进行的项目、任务
- 指标/KPI：关键绩效指标（如转化率、退单率等）
- 概念/制度：管理概念、制度要求
- 地点：会议地点、项目地点

### 2. 关系 (主体-关系谓词-客体)
抽取事实性关系，例如：
- {"subject": "建维部", "subject_type": "组织", "predicate": "负责", "object": "网络运维", "object_type": "任务", "description": ""}
- {"subject": "弱光指标", "subject_type": "指标", "predicate": "目标值", "object": "0.5以下", "object_type": "数值", "description": ""}

### 3. 行动项
谁负责什么任务，截止时间，优先级

### 4. 决策
做出的决定和结论

### 5. 指标数据
具体的数字指标：当前值、目标值、负责人、趋势(向好/持平/恶化)

## 规则
- 只提取事实性信息
- 过滤比喻、假设、主观评价
- 数字指标要精确提取
- entities、relations、action_items、decisions、metrics 如果没有则返回空数组
"""


def _call_llm(system: str, user: str) -> str:
    response = client.chat.completions.create(
        model=config.llm.model,
        messages=[
            {"role": "system", "content": system},
            {"role": "user", "content": user},
        ],
        max_tokens=config.llm.max_tokens,
        temperature=config.llm.temperature,
    )
    content = response.choices[0].message.content
    if content is None:
        raise ValueError("LLM returned empty response")
    return content


def extract_meeting_info(text: str) -> MeetingExtraction:
    user_prompt = f"""
从以下会议记录中抽取结构化信息。

JSON字段说明：
- title: 会议标题
- date: 会议日期
- participants: 参会人列表
- agenda: 议程列表
- entities: 实体列表，每个实体包含 name(名称), entity_type(类型), description(描述)
- relations: 关系列表，每个关系包含 subject(主体), subject_type(主体类型), predicate(关系谓词), object(客体), object_type(客体类型), description(描述)
- action_items: 行动项列表，每条包含 task(任务), assignee(负责人), deadline(截止时间), status(状态), priority(优先级)
- decisions: 决策列表，每条包含 content(决策内容), proposer(提出人), status(状态)
- metrics: 指标列表，每条包含 metric_name(指标名), value(当前值), target(目标值), owner(负责人), trend(趋势)
- summary: 会议摘要

请直接返回JSON对象。不要包含任何额外说明文字。

会议记录：
{text}
"""
    content = _call_llm(EXTRACTION_SYSTEM_PROMPT, user_prompt)
    data = _try_parse_json(content)
    return MeetingExtraction(**data)


def _try_parse_json(content: str) -> dict:
    try:
        return json.loads(content)
    except json.JSONDecodeError:
        logger.warning("JSON解析失败，尝试修复...")
        match = re.search(r'\{.*\}', content, re.DOTALL)
        if match:
            try:
                return json.loads(match.group())
            except json.JSONDecodeError as e:
                logger.error(f"修复后的JSON仍无法解析: {e}")
        raise