import asyncio
import re
import random
from typing import TypedDict, List, Dict, Any, Literal
from langgraph.graph import StateGraph, START, END
import os
import sys

current_dir = os.path.dirname(os.path.abspath(__file__))
parent_dir = os.path.dirname(current_dir)
if parent_dir not in sys.path:
    sys.path.append(parent_dir)

from modelsAPI.model_api import OpenaiAPI
from utils.function_tracker import track_function_calls
from utils.word_generator import generate_report_word
from config import SERVER_CONFIG
import json


from workflows.history_manager import init_history, build_history_str


def _extract_images_from_text(text: str) -> List[Dict[str, str]]:
    """从文本中提取图片URL和描述信息"""
    if not text:
        return []

    images = []
    url_pattern = r'https?://[^\s<>"{}|\\^`\[\]]+\.(?:png|jpg|jpeg|gif|bmp|webp)'
    urls = re.findall(url_pattern, text, re.IGNORECASE)

    for url in urls:
        summary = ""
        url_index = text.find(url)
        if url_index != -1:
            after_url = text[url_index + len(url):]
            result_match = re.search(
                r'(?:\*\*图片识别结果：\*\*|图片识别结果[：:]|识别结果[：:])\s*(.*?)(?=\n\n\n|\n\n[A-Za-z\u4e00-\u9fa5]|$)',
                after_url,
                re.DOTALL
            )
            if result_match:
                summary = result_match.group(1).strip()[:300]

        images.append({
            "url": url,
            "summary": summary or "用户上传的图片"
        })

    return images


class HistoryQAAnalysisState(TypedDict):
    """历史问答分析工作流状态"""

    extracted_text: str
    combined_query: str

    history_message: str
    history_messages: List[Dict[str, Any]]

    extracted_items: List[Dict[str, Any]]
    extracted_images: List[Dict[str, str]]
    report_title: str
    report_text: str

    response: str
    actions: List[Dict[str, Any]]
    route_flag: Literal["", "report"]
    suggestedReplies: List[Dict[str, Any]]

    error_message: str


async def extract_analysis_items(state: HistoryQAAnalysisState) -> Dict[str, Any]:
    from utils.function_tracker import get_all_callbacks

    combined_query = state.get("combined_query", "") or state.get("extracted_text", "") or ""
    history_messages = state.get("history_messages", [])
    history_message = state.get("history_message", "")

    pre_extracted_images = _extract_images_from_text(history_message)

    history_context = build_history_str(history_messages, recent_count=10, assistant_truncate=200)

    if not history_context and history_message:
        history_context = history_message

    loading_prompts = [
        {"title": "提取相关历史问答数据中...", "details": f"历史消息数: {len(history_messages)}条"},
        {"title": "🔍 正在分析历史对话内容...", "details": f"共处理 {len(history_messages)} 条消息"},
        {"title": "📝 解析历史问答信息中...", "details": f"已加载 {len(history_messages)} 条历史记录"},
        {"title": "📊 整理历史对话数据...", "details": f"当前历史消息: {len(history_messages)} 条"},
        {"title": "🔑 提取对话关键信息...", "details": f"历史对话总数: {len(history_messages)}条"},
    ]
    selected_prompt = random.choice(loading_prompts)
    parent_event = {
        "type": "function_execution",
        "title": selected_prompt["title"],
        "details": selected_prompt["details"],
    }
    for callback in get_all_callbacks():
        try:
            callback(parent_event)
        except Exception:
            pass

    prompt = f"""
你是一名情报分析与知识管理专家。
请从以下【用户输入】中抽取用于统计分析的结构化信息。

【用户输入】
{combined_query}

【历史对话上下文】
{history_context if history_context else "（无）"}


【抽取规则】
输出一个JSON对象，包含以下字段：
- report_title：报告标题（根据历史对话内容生成一个简洁的标题，格式为"XXX故障处理报告"，不超过20个字）
- items：抽取记录数组，每条记录包含：
  - topic：问题主题（如 设备、系统、管理、流程、保障、训练 等）
  - key_entity：关键对象或名词
  - conclusion：结论 / 判断 / 回答要点
  - action_suggestion：建议或行动（若无可为空）
  - date：时间（若无法判断可为空）
- images：图片信息数组（仅当有用户上传图片时提取，每条记录包含）：
  - url：图片URL（直接从【用户上传图片信息】中提取）
  - summary：图片内容摘要（根据图片识别结果提炼关键信息，如设备类型、故障现象、异常情况等，不超过100字）

【输出要求】
1. 仅输出 JSON 对象
2. 不要附加任何解释
3. 如果没有图片信息，images字段返回空数组
"""

    try:
        result = await OpenaiAPI.open_api_chat_without_thinking(
            query=prompt,
            model=None,
            json_output=True
        )
        parsed_result = json.loads(result)
        if isinstance(parsed_result, dict):
            extracted_items = parsed_result.get("items", [])
            report_title = parsed_result.get("report_title", "历史问答分析报告")
            model_images = parsed_result.get("images", [])


        else:
            extracted_items = parsed_result if isinstance(parsed_result, list) else []
            report_title = "历史问答分析报告"
            model_images = []
    except Exception as e:
        extracted_items = []
        report_title = "历史问答分析报告"
        model_images = []

    final_images = pre_extracted_images if pre_extracted_images else model_images

    return {"extracted_items": extracted_items, "report_title": report_title, "extracted_images": final_images}


# @track_function_calls
async def generate_analysis_report_text(state: HistoryQAAnalysisState) -> Dict[str, Any]:
    from utils.function_tracker import get_all_callbacks
    import time

    extracted_items = state.get("extracted_items", [])
    report_title = state.get("report_title", "历史问答分析报告")
    combined_query = state.get("combined_query", "") or state.get("extracted_text", "") or ""
    history_messages = state.get("history_messages", [])
    extracted_images = state.get("extracted_images", [])
    history_message = state.get("history_message", "")

    def summarize_extracted_items(items) -> str:
        """将抽取结果转为中文描述，兼容 list / dict"""

        if not items:
            return "未提取到有效分析数据"

        # 如果是单个 dict，转为 list
        if isinstance(items, dict):
            items = [items]

        # 如果不是 list，直接返回
        if not isinstance(items, list):
            return "分析数据格式异常"

        topics = set()
        entities = set()

        for item in items:
            if not isinstance(item, dict):
                continue

            topic = item.get("topic")
            entity = item.get("key_entity")

            if topic:
                topics.add(str(topic))
            if entity:
                entities.add(str(entity))

        topic_text = "、".join(list(topics)[:5]) if topics else "未识别"
        entity_text = "、".join(list(entities)[:5]) if entities else "未识别"

        return f"识别主题 {len(items)} 条，涉及主题：{topic_text}；关键对象：{entity_text}"

    details_text = summarize_extracted_items(extracted_items)
    loading_prompts = [
        {"title": "📄 生成分析报告文本", "details": details_text},
        {"title": "✍️ 正在撰写分析报告", "details": details_text},
        {"title": "📋 生成历史问答分析报告", "details": details_text},
        {"title": "📝 构建分析报告结构", "details": details_text},
        {"title": "🎯 汇总历史问答数据生成报告", "details": details_text},
    ]
    selected_prompt = random.choice(loading_prompts)
    start_event = {
        "type": "function_execution",
        "title": selected_prompt["title"],
        "details": selected_prompt["details"]
    }
    for callback in get_all_callbacks():
        try:
            callback(start_event)
        except Exception:
            pass

    history_context = build_history_str(history_messages, recent_count=10, assistant_truncate=300)

    if not history_context and history_message:
        history_context = history_message

    if not history_context and not combined_query:
        return {
            "report_text": "",
            "response": "【无法生成分析报告】\n\n缺少历史对话信息和用户输入，请先进行相关问答。",
            "actions": [],
            "route_flag": "report",
            "suggestedReplies": [],
            "error_message": "缺少历史对话信息和用户输入"
        }

    random_letters = ''.join(random.choices('ABCDEFGHIJKLMNOPQRSTUVWXYZ', k=2))
    report_number = time.strftime(f"{random_letters}-%Y-%m-%d-%H%M")
    compile_time = time.strftime("%Y-%m-%d %H:%M")

    image_section = ""
    if extracted_images:
        image_parts = []
        for idx, img in enumerate(extracted_images, 1):
            image_parts.append(f"图片{idx}：{img['url']}")
            summary = img.get('summary') or img.get('description', '')
            if summary:
                image_parts.append(f"识别结果：{summary}")
            image_parts.append("")
        image_section = "\n".join(image_parts)

    prompt = f"""
<role>
你是一名设备运维与故障分析报告编写专家。
你的任务是根据用户需求、历史对话和抽取信息生成一份正式的故障处理报告。
</role>

<user_requirement>
{combined_query}
</user_requirement>

<extracted_items>
{extracted_items}
</extracted_items>

<history_context>
{history_context if history_context else "（无历史对话上下文）"}
</history_context>

{f'''<user_images>
{image_section}
</user_images>''' if extracted_images else ''}

<output_rules>

必须严格遵守以下规则：

1. 输出为 **正式技术报告文本**
2. **使用段落形式，如果有表格生成，生成表格必须arkdown表格，不要html形式**
3. **如果给定结果在历史对话信息中没有，可以进行删除**
4. **输出内容必须严格按照历史对话信息内容进行生成，不得编造不实信息和不相干信息**
5. 不得输出解释说明
6. 不要使用列表符号
7. 所有内容使用完整中文句子。
8. **每个段落首行必须缩进，使用两个全角空格（　　）实现首行缩进两个中文字符的效果**
{"9. **报告中必须包含用户上传的图片信息，在故障信息部分引用图片URL**" if extracted_images else ""}

</output_rules>

<report_template>

{report_title}

报告编号： {report_number}
编制时间： {compile_time}
事件类型： 设备故障诊断与处置

一、事件概述
　　简要说明故障发生背景、报警情况、系统影响以及运维团队启动排查流程的情况。

二、故障信息
　　说明设备名称、型号、所属系统、报警时间、故障现象以及参与处置人员。
{"　　**附件图片：**" + chr(10) + chr(10).join([f"　　![图片{idx}]({img['url']})" for idx, img in enumerate(extracted_images, 1)]) if extracted_images else ""}

三、故障排查过程
　　按照时间顺序描述排查步骤，包括：用户反馈 → 初步判断 → 数据分析 → 现场检查 → 故障确认。

四、维修措施
　　描述维修人员采取的具体维修步骤，包括设备停机、拆卸检查、更换部件、紧固连接、恢复运行和测试过程。

五、后续建议
　　提出预防类似故障的改进措施，例如：维护周期优化、知识库记录、巡检项目增加、技术培训等。

六、总结
　　总结本次故障原因、处理效果以及经验意义。

</report_template>

<final_rule>
只输出完整报告正文，不要任何解释说明。
</final_rule>
"""

    try:
        report_text = await OpenaiAPI.open_api_chat_without_thinking(
            query=prompt,
            model=None,
            messages=history_messages
        )
        if not report_text:
            report_text = "抱歉，当前无法生成分析报告，请稍后重试。"

        word_dir = os.path.join(parent_dir, "created_word")
        os.makedirs(word_dir, exist_ok=True)

        safe_topic = re.sub(r'[\\/:*?"<>|]', '_', report_title)
        file_name = f"{safe_topic}.docx"
        output_path = os.path.join(word_dir, file_name)

        word_result = generate_report_word(
            markdown=report_text,
            title=report_title,
            output=output_path
        )

        actions: List[Dict[str, Any]] = []

        if word_result and os.path.exists(word_result):
            base_url = SERVER_CONFIG.get("report")
            download_url = f"http://192.168.1.164:9088/api/download/{file_name}"

            actions.append({
                "type": "download",
                "label": f"{report_title}.docx",
                "url": download_url,
                "fileName": file_name,
                "content": report_text
            })
        else:
            error_msg = f"\n\n[注意] Word 文档生成失败"
            report_text += error_msg

        # actions.append({
        #     "type": "regenerate",
        #     "label": "重新生成",
        #     "icon": "report"
        # })

        suggested_replies = await generate_suggested_replies_qa_report(report_text)

        return {
            "report_text": report_text,
            "response": report_text,
            "actions": actions,
            "route_flag": "report",
            "suggestedReplies": [],
#            "suggestedReplies": suggested_replies,
            "__details__": "分析报告生成完成"
        }
    except Exception as e:
        # error_event = {
        #     "type": "function_execution",
        #     "title": "步骤二失败：生成分析报告",
        #     "details": f"错误: {str(e)}",
        #     "has_result": False
        # }
        # for callback in get_all_callbacks():
        #     try:
        #         callback(error_event)
        #     except Exception:
        #         pass

        return {
            "report_text": f"生成分析报告失败: {str(e)}",
            "response": f"生成分析报告失败: {str(e)}",
            "actions": [],
            "route_flag": "report",
            "suggestedReplies": [],
            "error_message": str(e),
            "__details__": f"生成分析报告失败: {str(e)}"
        }


async def generate_suggested_replies_qa_report(report: str) -> List[Dict[str, Any]]:
    """
    生成针对历史问答分析报告的建议回复问题
    使用大模型生成两个相关问题
    """
    prompt = f"""
你是一个专业的数据分析助手。根据以下历史问答分析报告，生成两个用户可能关心的后续问题。

【分析报告】
{report[:500]}

请生成两个简洁、实用的问题，这些问题应该：
1. 不得出现需要归档和下载等操作相关的问题
2. 与分析报告相关，能够帮助用户更好地理解或使用分析结果
3. 问题要具体、可操作
4. 每个问题不超过20个字

请以JSON格式输出，格式如下：
[
    {{"title": "问题1的标题", "content": "问题1的完整内容"}},
    {{"title": "问题2的标题", "content": "问题2的完整内容"}}
]

只输出JSON，不要有其他文字说明。
"""
    try:
        answer = await OpenaiAPI.open_api_chat_without_thinking(
            query=prompt,
            model=None
        )
        json_match = re.search(r'\[.*\]', answer, re.DOTALL)
        if json_match:
            json_str = json_match.group(0)
            suggested_replies = json.loads(json_str)
            if isinstance(suggested_replies, list) and len(suggested_replies) >= 2:
                formatted_replies = []
                for reply in suggested_replies[:2]:
                    if isinstance(reply, dict) and "title" in reply and "content" in reply:
                        formatted_replies.append({
                            "title": str(reply["title"]),
                            "content": str(reply["content"])
                        })
                if len(formatted_replies) == 2:
                    return formatted_replies

        return [
            {"title": "如何优化问题分布？", "content": "如何优化问题分布？"},
            {"title": "有哪些改进建议？", "content": "有哪些改进建议？"}
        ]
    except Exception as e:
        return [
            {"title": "如何优化问题分布？", "content": "如何优化问题分布？"},
            {"title": "有哪些改进建议？", "content": "有哪些改进建议？"}
        ]


def create_history_qa_analysis_workflow():
    graph = StateGraph(HistoryQAAnalysisState)

    graph.add_node("extract_items", extract_analysis_items)
    graph.add_node("generate_text", generate_analysis_report_text)

    graph.add_edge(START, "extract_items")
    graph.add_edge("extract_items", "generate_text")
    graph.add_edge("generate_text", END)

    return graph.compile()


# @track_function_calls
async def run_qa_report_agent(
        extracted_text: str,
        combined_query: str,
        history_message: str = "",
        route_flag: str = "",
        extracted_images: List[Dict[str, str]] = None
) -> Dict[str, Any]:
    app = create_history_qa_analysis_workflow()

    initial_state: HistoryQAAnalysisState = {
        "extracted_text": extracted_text,
        "combined_query": combined_query,
        **init_history(history_message),
        "extracted_items": [],
        "extracted_images": extracted_images or [],
        "report_title": "",
        "report_text": "",
        "response": "",
        "actions": [],
        "route_flag": "",
        "suggestedReplies": [],
        "error_message": ""
    }

    return await app.ainvoke(initial_state)


if __name__ == "__main__":
    import asyncio


    async def test():
        history_message = json.dumps(
            [
                {"role": "user", "content": "我需要分析最近的历史问答数据"},
                {"role": "assistant", "content": "好的，我可以帮您生成历史问答分析报告，请提供相关数据。"},
                {"role": "user", "content": "请重点关注设备故障和维护相关的内容"}
            ],
            ensure_ascii=False,
        )

        result = await run_qa_report_agent(
            extracted_text="请重点关注设备故障和维护相关的内容",
            combined_query="用户问题：请重点关注设备故障和维护相关的内容",
            history_message=history_message
        )

        print("====== 历史问答分析报告文本 ======")
        print(result["response"])
        print("\n====== 推荐问题 ======")
        for reply in result.get("suggestedReplies", []):
            print(f"- {reply['title']}: {reply['content']}")
        print("\n====== 操作按钮 ======")
        for action in result.get("actions", []):
            print(f"- {action.get('label', 'Unknown')}: {action.get('type', 'Unknown')}")


    asyncio.run(test())