012-kaopeilian/backend/app/api/v1/practice.py

"""
陪练功能API
"""
from typing import Optional
import json
from datetime import datetime, timedelta
from fastapi import APIRouter, Depends, Query, HTTPException
from fastapi.responses import StreamingResponse
from sqlalchemy import select, func, or_
from sqlalchemy.ext.asyncio import AsyncSession
from cozepy import ChatEventType
from cozepy.exception import CozeError, CozeAPIError

from app.core.deps import get_db, get_current_user
from app.models.user import User
from app.models.practice import PracticeScene, PracticeSession, PracticeDialogue, PracticeReport
from app.schemas.practice import (
    PracticeSceneResponse,
    PracticeSceneCreate,
    PracticeSceneUpdate,
    StartPracticeRequest,
    InterruptPracticeRequest,
    ConversationsResponse,
    ExtractSceneRequest,
    ExtractSceneResponse,
    ExtractedSceneData,
    PracticeSessionCreate,
    PracticeSessionResponse,
    SaveDialogueRequest,
    PracticeDialogueResponse,
    PracticeReportResponse,
    PracticeAnalysisResult
)
from app.schemas.base import ResponseModel, PaginatedResponse
from app.services.coze_service import get_coze_service, CozeService
from app.services.ai.coze.client import get_auth_manager
from app.services.ai.practice_analysis_service import practice_analysis_service
import logging

logger = logging.getLogger(__name__)

router = APIRouter()


@router.get("/coze-token")
async def get_coze_token(
    current_user: User = Depends(get_current_user)
):
    """
    获取Coze OAuth Token用于前端直连WebSocket
    
    前端语音对话需要直连Coze WebSocket，但不能暴露私钥，
    因此通过此接口从后端获取临时Token
    """
    try:
        auth_manager = get_auth_manager()
        token = auth_manager.get_oauth_token()
        
        return ResponseModel(
            code=200,
            message="Token获取成功",
            data={"token": token}
        )
    except Exception as e:
        logger.error(f"获取Coze Token失败: {e}")
        raise HTTPException(status_code=500, detail=f"获取Token失败: {str(e)}")


@router.get("/scenes", response_model=ResponseModel[PaginatedResponse[PracticeSceneResponse]])
async def get_practice_scenes(
    page: int = Query(1, ge=1, description="页码"),
    size: int = Query(20, ge=1, le=100, description="每页数量"),
    type: Optional[str] = Query(None, description="场景类型筛选"),
    difficulty: Optional[str] = Query(None, description="难度筛选"),
    search: Optional[str] = Query(None, description="关键词搜索（名称、描述）"),
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    获取可用陪练场景列表
    
    仅返回status=active且未删除的场景
    支持分页、筛选和搜索
    """
    # 构建查询
    query = select(PracticeScene).where(
        PracticeScene.is_deleted == False,
        PracticeScene.status == "active"
    )
    
    # 类型筛选
    if type:
        query = query.where(PracticeScene.type == type)
    
    # 难度筛选
    if difficulty:
        query = query.where(PracticeScene.difficulty == difficulty)
    
    # 关键词搜索（搜索名称和描述）
    if search:
        search_pattern = f"%{search}%"
        query = query.where(
            or_(
                PracticeScene.name.like(search_pattern),
                PracticeScene.description.like(search_pattern)
            )
        )
    
    # 查询总数
    count_query = select(func.count()).select_from(query.subquery())
    total = await db.scalar(count_query)
    
    # 分页查询
    query = query.offset((page - 1) * size).limit(size).order_by(PracticeScene.created_at.desc())
    result = await db.scalars(query)
    scenes = list(result.all())
    
    logger.info(
        f"用户{current_user.id}查询陪练场景列表，"
        f"类型={type}, 难度={difficulty}, 搜索={search}, "
        f"返回{len(scenes)}条记录"
    )
    
    return ResponseModel(
        code=200,
        message="success",
        data=PaginatedResponse(
            items=scenes,
            total=total or 0,
            page=page,
            page_size=size,
            pages=(total + size - 1) // size if total else 0
        )
    )


@router.get("/scenes/{scene_id}", response_model=ResponseModel[PracticeSceneResponse])
async def get_practice_scene_detail(
    scene_id: int,
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    获取陪练场景详情
    
    返回指定ID的场景完整信息
    """
    # 查询场景
    result = await db.execute(
        select(PracticeScene).where(
            PracticeScene.id == scene_id,
            PracticeScene.is_deleted == False,
            PracticeScene.status == "active"
        )
    )
    scene = result.scalar_one_or_none()
    
    if not scene:
        logger.warning(f"用户{current_user.id}查询场景{scene_id}不存在或已禁用")
        raise HTTPException(status_code=404, detail="场景不存在或已禁用")
    
    logger.info(f"用户{current_user.id}查询场景{scene_id}详情")
    
    return ResponseModel(code=200, message="success", data=scene)


@router.post("/scenes", response_model=ResponseModel[PracticeSceneResponse])
async def create_practice_scene(
    scene_data: PracticeSceneCreate,
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    创建陪练场景
    
    仅管理员和经理可以创建场景
    """
    # 权限检查
    if current_user.role not in ["admin", "manager"]:
        raise HTTPException(status_code=403, detail="无权限创建陪练场景")
    
    # 创建场景
    scene = PracticeScene(
        **scene_data.model_dump(),
        created_by=current_user.id,
        updated_by=current_user.id
    )
    
    db.add(scene)
    await db.commit()
    await db.refresh(scene)
    
    logger.info(f"用户{current_user.id}创建陪练场景: {scene.name} (ID: {scene.id})")
    
    return ResponseModel(
        code=200,
        message="场景创建成功",
        data=scene
    )


@router.put("/scenes/{scene_id}", response_model=ResponseModel[PracticeSceneResponse])
async def update_practice_scene(
    scene_id: int,
    scene_data: PracticeSceneUpdate,
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    更新陪练场景
    
    仅管理员和经理可以更新场景
    """
    # 权限检查
    if current_user.role not in ["admin", "manager"]:
        raise HTTPException(status_code=403, detail="无权限更新陪练场景")
    
    # 查询场景
    result = await db.execute(
        select(PracticeScene).where(
            PracticeScene.id == scene_id,
            PracticeScene.is_deleted == False
        )
    )
    scene = result.scalar_one_or_none()
    
    if not scene:
        raise HTTPException(status_code=404, detail="场景不存在")
    
    # 更新字段
    update_data = scene_data.model_dump(exclude_unset=True)
    for field, value in update_data.items():
        setattr(scene, field, value)
    
    scene.updated_by = current_user.id
    
    await db.commit()
    await db.refresh(scene)
    
    logger.info(f"用户{current_user.id}更新陪练场景: {scene.name} (ID: {scene.id})")
    
    return ResponseModel(
        code=200,
        message="场景更新成功",
        data=scene
    )


@router.delete("/scenes/{scene_id}", response_model=ResponseModel)
async def delete_practice_scene(
    scene_id: int,
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    删除陪练场景（软删除）
    
    仅管理员和经理可以删除场景
    """
    # 权限检查
    if current_user.role not in ["admin", "manager"]:
        raise HTTPException(status_code=403, detail="无权限删除陪练场景")
    
    # 查询场景
    result = await db.execute(
        select(PracticeScene).where(
            PracticeScene.id == scene_id,
            PracticeScene.is_deleted == False
        )
    )
    scene = result.scalar_one_or_none()
    
    if not scene:
        raise HTTPException(status_code=404, detail="场景不存在")
    
    # 软删除
    scene.is_deleted = True
    scene.updated_by = current_user.id
    
    await db.commit()
    
    logger.info(f"用户{current_user.id}删除陪练场景: {scene.name} (ID: {scene.id})")
    
    return ResponseModel(
        code=200,
        message="场景删除成功",
        data={"scene_id": scene_id}
    )


@router.post("/start")
async def start_practice(
    request: StartPracticeRequest,
    current_user: User = Depends(get_current_user),
    coze_service: CozeService = Depends(get_coze_service)
):
    """
    开始陪练对话（SSE流式返回）
    
    ⚠️ 核心功能：
    - 首次消息（is_first=true）：构建完整场景提示词发送给Coze
    - 后续消息（is_first=false）：仅发送用户消息
    - 使用conversation_id保持对话上下文
    """
    logger.info(
        f"用户{current_user.id}开始陪练对话，"
        f"场景={request.scene_name}, "
        f"is_first={request.is_first}, "
        f"conversation_id={request.conversation_id}"
    )
    
    # 构建发送给Coze的消息
    if request.is_first:
        # 首次消息：构建完整场景提示词
        message = coze_service.build_scene_prompt(
            scene_name=request.scene_name,
            scene_background=request.scene_background,
            scene_ai_role=request.scene_ai_role,
            scene_objectives=request.scene_objectives,
            scene_keywords=request.scene_keywords,
            scene_description=request.scene_description,
            user_message=request.user_message
        )
        logger.debug(f"场景提示词已构建，长度={len(message)}字符")
    else:
        # 后续消息：仅发送用户输入
        message = request.user_message
        logger.debug(f"用户消息: {message}")
    
    def generate_stream():
        """SSE流式生成器"""
        try:
            # 创建Coze流式对话
            stream = coze_service.create_stream_chat(
                user_id=str(current_user.id),
                message=message,
                conversation_id=request.conversation_id
            )
            
            # 处理Coze事件流
            for event in stream:
                # 对话创建事件
                if event.event == ChatEventType.CONVERSATION_CHAT_CREATED:
                    # 优先使用请求中的conversation_id（续接对话）
                    # 如果没有，使用Coze返回的新对话ID（首次对话）
                    final_conversation_id = request.conversation_id or event.chat.conversation_id
                    event_data = {
                        "conversation_id": final_conversation_id,
                        "chat_id": event.chat.id
                    }
                    yield f"event: conversation.chat.created\ndata: {json.dumps(event_data)}\n\n"
                    logger.debug(f"对话已创建/续接: conversation_id={final_conversation_id}, 来源={'请求参数' if request.conversation_id else 'Coze创建'}")
                
                # 消息增量事件（实时打字效果）
                elif event.event == ChatEventType.CONVERSATION_MESSAGE_DELTA:
                    event_data = {"content": event.message.content}
                    yield f"event: message.delta\ndata: {json.dumps(event_data)}\n\n"
                
                # 消息完成事件
                elif event.event == ChatEventType.CONVERSATION_MESSAGE_COMPLETED:
                    event_data = {}  # 不需要返回完整内容，前端已通过delta累积
                    yield f"event: message.completed\ndata: {json.dumps(event_data)}\n\n"
                    logger.info(f"消息已完成")
                
                # 对话完成事件
                elif event.event == ChatEventType.CONVERSATION_CHAT_COMPLETED:
                    # 安全地获取token用量
                    token_count = 0
                    input_count = 0
                    output_count = 0
                    if hasattr(event.chat, 'usage') and event.chat.usage:
                        token_count = getattr(event.chat.usage, 'token_count', 0)
                        input_count = getattr(event.chat.usage, 'input_count', 0)
                        output_count = getattr(event.chat.usage, 'output_count', 0)
                    
                    event_data = {
                        "token_count": token_count,
                        "input_count": input_count,
                        "output_count": output_count
                    }
                    yield f"event: conversation.completed\ndata: {json.dumps(event_data)}\n\n"
                    logger.info(f"对话已完成，Token用量={event_data['token_count']}")
                    break
                
                # 对话失败事件
                elif event.event == ChatEventType.CONVERSATION_CHAT_FAILED:
                    error_msg = str(event.chat.last_error) if event.chat.last_error else "对话失败"
                    event_data = {"error": error_msg}
                    yield f"event: error\ndata: {json.dumps(event_data)}\n\n"
                    logger.error(f"对话失败: {error_msg}")
                    break
            
            # 发送结束标记
            yield f"event: done\ndata: [DONE]\n\n"
            logger.info(f"SSE流结束")
            
        except (CozeError, CozeAPIError) as e:
            logger.error(f"Coze API错误: {e}", exc_info=True)
            error_data = {"error": f"对话失败: {str(e)}"}
            yield f"event: error\ndata: {json.dumps(error_data)}\n\n"
        except Exception as e:
            logger.error(f"陪练对话异常: {e}", exc_info=True)
            error_data = {"error": f"系统错误: {str(e)}"}
            yield f"event: error\ndata: {json.dumps(error_data)}\n\n"
    
    # 返回SSE流式响应
    return StreamingResponse(
        generate_stream(),
        media_type="text/event-stream",
        headers={
            "Cache-Control": "no-cache",
            "Connection": "keep-alive",
            "X-Accel-Buffering": "no"  # 禁用Nginx缓冲
        }
    )


@router.post("/interrupt", response_model=ResponseModel)
async def interrupt_practice(
    request: InterruptPracticeRequest,
    current_user: User = Depends(get_current_user),
    coze_service: CozeService = Depends(get_coze_service)
):
    """
    中断陪练对话
    
    调用Coze API中断当前进行中的对话
    """
    logger.info(
        f"用户{current_user.id}中断对话，"
        f"conversation_id={request.conversation_id}, "
        f"chat_id={request.chat_id}"
    )
    
    try:
        result = coze_service.cancel_chat(
            conversation_id=request.conversation_id,
            chat_id=request.chat_id
        )
        
        return ResponseModel(
            code=200,
            message="对话已中断",
            data={
                "conversation_id": request.conversation_id,
                "chat_id": request.chat_id
            }
        )
    except (CozeError, CozeAPIError) as e:
        logger.error(f"中断对话失败: {e}")
        raise HTTPException(status_code=500, detail=f"中断对话失败: {str(e)}")
    except Exception as e:
        logger.error(f"中断对话异常: {e}", exc_info=True)
        raise HTTPException(status_code=500, detail=f"系统错误: {str(e)}")


@router.post("/conversation/create", response_model=ResponseModel)
async def create_practice_conversation(
    current_user: User = Depends(get_current_user),
    coze_service: CozeService = Depends(get_coze_service)
):
    """
    创建新的陪练对话
    
    ⚠️ 关键：必须先创建conversation，然后才能续接对话
    返回conversation_id供后续对话使用
    """
    try:
        # 调用Coze API创建对话
        conversation = coze_service.client.conversations.create()
        
        conversation_id = conversation.id
        
        logger.info(f"用户{current_user.id}创建陪练对话，conversation_id={conversation_id}")
        
        return ResponseModel(
            code=200,
            message="对话创建成功",
            data={"conversation_id": conversation_id}
        )
    except (CozeError, CozeAPIError) as e:
        logger.error(f"创建对话失败: {e}")
        raise HTTPException(status_code=500, detail=f"创建对话失败: {str(e)}")
    except Exception as e:
        logger.error(f"创建对话异常: {e}", exc_info=True)
        raise HTTPException(status_code=500, detail=f"系统错误: {str(e)}")


@router.get("/conversations", response_model=ResponseModel[ConversationsResponse])
async def get_conversations(
    page: int = Query(1, ge=1, description="页码"),
    size: int = Query(20, ge=1, le=100, description="每页数量"),
    current_user: User = Depends(get_current_user)
):
    """
    获取对话列表
    
    查询用户在Coze平台上的对话历史
    
    注意：语音陪练使用前端直连Coze WebSocket，不经过后端中转
    """
    # TODO: 实现对话列表查询
    # 将在阶段四实现
    logger.info(f"用户{current_user.id}查询对话列表")
    raise HTTPException(status_code=501, detail="对话列表功能正在开发中")


@router.post("/extract-scene", response_model=ResponseModel[ExtractSceneResponse])
async def extract_scene(
    request: ExtractSceneRequest,
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    从课程提取陪练场景
    
    使用 Python 原生 AI 服务实现，直接调用 AI API 生成场景。
    
    流程：
    1. 验证课程是否存在
    2. 获取课程知识点
    3. 调用 AI 生成陪练场景
    4. 解析并返回场景数据
    """
    from app.models.course import Course
    from app.services.ai import practice_scene_service
    
    # 验证课程存在
    course = await db.get(Course, request.course_id)
    if not course:
        logger.warning(f"课程不存在: course_id={request.course_id}")
        raise HTTPException(status_code=404, detail="课程不存在")
    
    logger.info(f"用户{current_user.id}开始提取课程{request.course_id}的陪练场景")
    
    # 调用 Python 原生服务
    result = await practice_scene_service.prepare_practice_knowledge(
        db=db,
        course_id=request.course_id
    )
    
    if not result.success:
        # 根据错误类型返回适当的 HTTP 状态码
        if "没有可用的知识点" in result.error or "没有知识点" in result.error:
            raise HTTPException(
                status_code=400,
                detail="该课程尚未添加知识点，无法生成陪练场景。请先在课程管理中上传资料并分析知识点。"
            )
        raise HTTPException(status_code=500, detail=f"场景提取失败: {result.error}")
    
    # 将 PracticeScene 转换为 ExtractedSceneData
    scene = result.scene
    scene_data = ExtractedSceneData(
        name=scene.name,
        description=scene.description,
        type=scene.type,
        difficulty=scene.difficulty,
        background=scene.background,
        ai_role=scene.ai_role,
        objectives=scene.objectives,
        keywords=scene.keywords
    )
    
    logger.info(
        f"场景提取成功: {scene.name}, course_id={request.course_id}, "
        f"provider={result.ai_provider}, tokens={result.ai_tokens}"
    )
    
    return ResponseModel(
        code=200,
        message="场景提取成功",
        data=ExtractSceneResponse(
            scene=scene_data,
            workflow_run_id=f"{result.ai_provider}_{result.ai_latency_ms}ms",
            task_id=f"native_{request.course_id}"
        )
    )


# ==================== 陪练会话管理API ====================

@router.post("/sessions/create", response_model=ResponseModel[PracticeSessionResponse])
async def create_practice_session(
    request: PracticeSessionCreate,
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    创建陪练会话
    
    用户开始陪练时调用，创建session记录
    """
    try:
        # 生成session_id（格式：PS + 时间戳后6位）
        session_id = f"PS{str(int(datetime.now().timestamp() * 1000))[-6:]}"
        
        # 创建session记录
        session = PracticeSession(
            session_id=session_id,
            user_id=current_user.id,
            scene_id=request.scene_id,
            scene_name=request.scene_name,
            scene_type=request.scene_type,
            conversation_id=request.conversation_id,
            start_time=datetime.now(),
            status="in_progress"
        )
        
        db.add(session)
        await db.commit()
        await db.refresh(session)
        
        logger.info(f"创建陪练会话: session_id={session_id}, user_id={current_user.id}, scene={request.scene_name}")
        
        return ResponseModel(
            code=200,
            message="会话创建成功",
            data=session
        )
    
    except Exception as e:
        logger.error(f"创建会话失败: {e}", exc_info=True)
        raise HTTPException(status_code=500, detail=f"创建会话失败: {str(e)}")


@router.post("/dialogues/save", response_model=ResponseModel)
async def save_dialogue(
    request: SaveDialogueRequest,
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    保存对话记录
    
    每一条对话（用户或AI）都实时保存
    """
    try:
        # 创建对话记录
        dialogue = PracticeDialogue(
            session_id=request.session_id,
            speaker=request.speaker,
            content=request.content,
            timestamp=datetime.now(),
            sequence=request.sequence
        )
        
        db.add(dialogue)
        await db.commit()
        
        logger.debug(f"保存对话: session_id={request.session_id}, speaker={request.speaker}, seq={request.sequence}")
        
        return ResponseModel(
            code=200,
            message="对话保存成功",
            data={"session_id": request.session_id, "sequence": request.sequence}
        )
    
    except Exception as e:
        logger.error(f"保存对话失败: {e}", exc_info=True)
        raise HTTPException(status_code=500, detail=f"保存对话失败: {str(e)}")


@router.post("/sessions/{session_id}/end", response_model=ResponseModel[PracticeSessionResponse])
async def end_practice_session(
    session_id: str,
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    结束陪练会话
    
    用户结束陪练时调用，更新会话状态和时长
    """
    try:
        # 查询会话
        result = await db.execute(
            select(PracticeSession).where(
                PracticeSession.session_id == session_id,
                PracticeSession.user_id == current_user.id,
                PracticeSession.is_deleted == False
            )
        )
        session = result.scalar_one_or_none()
        
        if not session:
            raise HTTPException(status_code=404, detail="会话不存在")
        
        # 查询对话数量
        result = await db.execute(
            select(func.count(PracticeDialogue.id)).where(
                PracticeDialogue.session_id == session_id
            )
        )
        dialogue_count = result.scalar() or 0
        
        # 更新会话状态
        session.end_time = datetime.now()
        session.duration_seconds = int((session.end_time - session.start_time).total_seconds())
        session.turns = dialogue_count
        session.status = "completed"
        
        await db.commit()
        await db.refresh(session)
        
        logger.info(f"结束陪练会话: session_id={session_id}, 时长={session.duration_seconds}秒, 轮次={session.turns}")
        
        return ResponseModel(
            code=200,
            message="会话已结束",
            data=session
        )
    
    except HTTPException:
        raise
    except Exception as e:
        logger.error(f"结束会话失败: {e}", exc_info=True)
        raise HTTPException(status_code=500, detail=f"结束会话失败: {str(e)}")


@router.post("/sessions/{session_id}/analyze", response_model=ResponseModel)
async def analyze_practice_session(
    session_id: str,
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    生成陪练分析报告
    
    使用 Python 原生 AI 服务实现。
    """
    try:
        # 1. 查询会话信息
        result = await db.execute(
            select(PracticeSession).where(
                PracticeSession.session_id == session_id,
                PracticeSession.user_id == current_user.id,
                PracticeSession.is_deleted == False
            )
        )
        session = result.scalar_one_or_none()
        
        if not session:
            raise HTTPException(status_code=404, detail="会话不存在")
        
        # 2. 查询对话历史
        result = await db.execute(
            select(PracticeDialogue).where(
                PracticeDialogue.session_id == session_id
            ).order_by(PracticeDialogue.sequence)
        )
        dialogues = result.scalars().all()
        
        if not dialogues or len(dialogues) < 2:
            raise HTTPException(status_code=400, detail="对话数量太少，无法生成分析报告")
        
        # 3. 构建对话历史数据
        dialogue_history = [
            {
                "speaker": d.speaker,
                "content": d.content,
                "timestamp": d.timestamp.isoformat()
            }
            for d in dialogues
        ]
        
        logger.info(f"开始分析陪练会话: session_id={session_id}, 对话数={len(dialogue_history)}")
        
        # 调用 Python 原生陪练分析服务
        v2_result = await practice_analysis_service.analyze(dialogue_history, db=db)
        
        if not v2_result.success:
            raise HTTPException(status_code=500, detail=f"分析失败: {v2_result.error}")
        
        analysis_data = v2_result.to_dict()
        
        logger.info(
            f"陪练分析完成 - total_score: {v2_result.total_score}, "
            f"provider: {v2_result.ai_provider}, latency: {v2_result.ai_latency_ms}ms"
        )
        
        # 解析分析结果
        analysis_result = analysis_data.get("analysis", {})
        
        # 保存分析报告
        report = PracticeReport(
            session_id=session_id,
            total_score=analysis_result.get("total_score"),
            score_breakdown=analysis_result.get("score_breakdown"),
            ability_dimensions=analysis_result.get("ability_dimensions"),
            dialogue_review=analysis_result.get("dialogue_annotations"),
            suggestions=analysis_result.get("suggestions"),
            workflow_run_id=f"{v2_result.ai_provider}_{v2_result.ai_latency_ms}ms",
            task_id=None
        )
        
        db.add(report)
        await db.commit()
        
        logger.info(f"分析报告已保存: session_id={session_id}, total_score={report.total_score}")
        
        return ResponseModel(
            code=200,
            message="分析报告生成成功",
            data={
                "session_id": session_id,
                "total_score": report.total_score,
                "workflow_run_id": report.workflow_run_id
            }
        )
    
    except HTTPException:
        raise
    except Exception as e:
        logger.error(f"生成分析报告失败: {e}, session_id={session_id}", exc_info=True)
        raise HTTPException(status_code=500, detail=f"生成分析报告失败: {str(e)}")


@router.get("/reports/{session_id}", response_model=ResponseModel[PracticeReportResponse])
async def get_practice_report(
    session_id: str,
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    获取陪练分析报告详情
    
    合并数据库对话记录和AI标注，生成完整的对话复盘
    """
    try:
        # 1. 查询会话信息
        result = await db.execute(
            select(PracticeSession).where(
                PracticeSession.session_id == session_id,
                PracticeSession.user_id == current_user.id,
                PracticeSession.is_deleted == False
            )
        )
        session = result.scalar_one_or_none()
        
        if not session:
            raise HTTPException(status_code=404, detail="会话不存在")
        
        # 2. 查询分析报告
        result = await db.execute(
            select(PracticeReport).where(
                PracticeReport.session_id == session_id
            )
        )
        report = result.scalar_one_or_none()
        
        if not report:
            raise HTTPException(status_code=404, detail="分析报告不存在，请先生成报告")
        
        # 3. 查询完整对话记录（从数据库）
        result = await db.execute(
            select(PracticeDialogue).where(
                PracticeDialogue.session_id == session_id
            ).order_by(PracticeDialogue.sequence)
        )
        dialogues = result.scalars().all()
        
        # 4. 合并对话记录和AI标注
        # dialogue_review字段存储的是标注信息（包含sequence, tags, comment）
        ai_annotations = report.dialogue_review or []
        
        # 创建标注映射（sequence -> {tags, comment}）
        annotations_map = {}
        for annotation in ai_annotations:
            seq = annotation.get('sequence')
            if seq:
                annotations_map[seq] = {
                    'tags': annotation.get('tags', []),
                    'comment': annotation.get('comment', '')
                }
        
        # 构建完整对话复盘（数据库对话 + AI标注）
        dialogue_review = []
        for dialogue in dialogues:
            # 计算时间（从会话开始时间算起）
            time_offset = int((dialogue.timestamp - session.start_time).total_seconds())
            time_str = f"{time_offset // 60:02d}:{time_offset % 60:02d}"
            
            # 获取标注
            annotation = annotations_map.get(dialogue.sequence, {})
            
            dialogue_review.append({
                "speaker": "顾问" if dialogue.speaker == "user" else "客户",
                "time": time_str,
                "content": dialogue.content,
                "tags": annotation.get('tags', []),
                "comment": annotation.get('comment', '')
            })
        
        # 5. 构建响应数据
        # 5.1 处理score_breakdown字段（兼容字典和列表格式）
        score_breakdown_data = report.score_breakdown or []
        if isinstance(score_breakdown_data, str):
            try:
                score_breakdown_data = json.loads(score_breakdown_data)
            except json.JSONDecodeError:
                logger.warning(f"无法解析score_breakdown JSON: {score_breakdown_data}")
                score_breakdown_data = []
        
        # 如果是字典格式，转换为列表格式
        if isinstance(score_breakdown_data, dict):
            score_breakdown_data = [
                {"name": k, "score": int(v), "description": ""} 
                for k, v in score_breakdown_data.items()
            ]
        
        # 5.2 处理ability_dimensions字段（兼容字典和列表格式）
        ability_dimensions_data = report.ability_dimensions or []
        if isinstance(ability_dimensions_data, str):
            try:
                ability_dimensions_data = json.loads(ability_dimensions_data)
            except json.JSONDecodeError:
                logger.warning(f"无法解析ability_dimensions JSON: {ability_dimensions_data}")
                ability_dimensions_data = []
        
        # 如果是字典格式，转换为列表格式
        if isinstance(ability_dimensions_data, dict):
            ability_dimensions_data = [
                {"name": k, "score": int(v), "feedback": ""} 
                for k, v in ability_dimensions_data.items()
            ]
        
        # 5.3 处理suggestions字段
        suggestions_data = report.suggestions or []
        if isinstance(suggestions_data, str):
            try:
                suggestions_data = json.loads(suggestions_data)
            except json.JSONDecodeError:
                logger.warning(f"无法解析suggestions JSON: {suggestions_data}")
                suggestions_data = []
        
        analysis = PracticeAnalysisResult(
            total_score=report.total_score,
            score_breakdown=score_breakdown_data,
            ability_dimensions=ability_dimensions_data,
            dialogue_review=dialogue_review,  # 使用合并后的对话
            suggestions=suggestions_data
        )
        
        response_data = PracticeReportResponse(
            session_info=session,
            analysis=analysis
        )
        
        logger.info(f"获取分析报告: session_id={session_id}, total_score={report.total_score}, 对话数={len(dialogue_review)}")
        
        return ResponseModel(
            code=200,
            message="success",
            data=response_data
        )
    
    except HTTPException:
        raise
    except Exception as e:
        logger.error(f"获取报告失败: {e}, session_id={session_id}", exc_info=True)
        raise HTTPException(status_code=500, detail=f"获取报告失败: {str(e)}")


# ==================== 陪练记录查询API ====================

@router.get("/sessions/list", response_model=ResponseModel[PaginatedResponse])
async def get_practice_sessions_list(
    page: int = Query(1, ge=1, description="页码"),
    size: int = Query(20, ge=1, le=100, description="每页数量"),
    keyword: Optional[str] = Query(None, description="关键词搜索"),
    scene_type: Optional[str] = Query(None, description="场景类型"),
    start_date: Optional[str] = Query(None, description="开始日期"),
    end_date: Optional[str] = Query(None, description="结束日期"),
    min_score: Optional[int] = Query(None, ge=0, le=100, description="最低分数"),
    max_score: Optional[int] = Query(None, ge=0, le=100, description="最高分数"),
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    获取陪练记录列表
    
    支持关键词搜索、场景筛选、时间范围筛选、分数筛选
    """
    try:
        # 构建查询（关联practice_reports表获取分数）
        query = select(
            PracticeSession,
            PracticeReport.total_score
        ).outerjoin(
            PracticeReport,
            PracticeSession.session_id == PracticeReport.session_id
        ).where(
            PracticeSession.user_id == current_user.id,
            PracticeSession.is_deleted == False,
            PracticeSession.status == "completed"  # 只查询已完成的会话
        )
        
        # 关键词搜索
        if keyword:
            query = query.where(
                or_(
                    PracticeSession.scene_name.contains(keyword),
                    PracticeSession.session_id.contains(keyword)
                )
            )
        
        # 场景类型筛选
        if scene_type:
            query = query.where(PracticeSession.scene_type == scene_type)
        
        # 时间范围筛选
        if start_date:
            query = query.where(PracticeSession.start_time >= start_date)
        if end_date:
            query = query.where(PracticeSession.start_time <= end_date)
        
        # 分数筛选
        if min_score is not None:
            query = query.where(PracticeReport.total_score >= min_score)
        if max_score is not None:
            query = query.where(PracticeReport.total_score <= max_score)
        
        # 按开始时间倒序排列
        query = query.order_by(PracticeSession.start_time.desc())
        
        # 计算总数
        count_query = select(func.count()).select_from(query.subquery())
        total = await db.scalar(count_query) or 0
        
        # 分页查询
        results = await db.execute(
            query.offset((page - 1) * size).limit(size)
        )
        
        # 构建响应数据
        items = []
        for session, total_score in results:
            # 计算result等级
            result_level = "needs_improvement"
            if total_score:
                if total_score >= 90:
                    result_level = "excellent"
                elif total_score >= 80:
                    result_level = "good"
                elif total_score >= 70:
                    result_level = "average"
            
            items.append({
                "session_id": session.session_id,
                "scene_name": session.scene_name,
                "scene_type": session.scene_type,
                "start_time": session.start_time,
                "duration_seconds": session.duration_seconds,
                "turns": session.turns,
                "total_score": total_score,
                "result": result_level
            })
        
        logger.info(f"查询陪练记录: user_id={current_user.id}, 返回{len(items)}条记录")
        
        return ResponseModel(
            code=200,
            message="success",
            data=PaginatedResponse(
                items=items,
                total=total,
                page=page,
                page_size=size,
                pages=(total + size - 1) // size
            )
        )
    
    except Exception as e:
        logger.error(f"查询陪练记录失败: {e}", exc_info=True)
        raise HTTPException(status_code=500, detail=f"查询失败: {str(e)}")


@router.get("/stats", response_model=ResponseModel)
async def get_practice_stats(
    db: AsyncSession = Depends(get_db),
    current_user: User = Depends(get_current_user)
):
    """
    获取陪练统计数据
    
    返回：总次数、平均分、总时长、本月进步
    """
    try:
        # 查询总次数和总时长
        result = await db.execute(
            select(
                func.count(PracticeSession.id).label('total_count'),
                func.sum(PracticeSession.duration_seconds).label('total_duration')
            ).where(
                PracticeSession.user_id == current_user.id,
                PracticeSession.is_deleted == False,
                PracticeSession.status == "completed"
            )
        )
        stats = result.first()
        
        total_count = stats.total_count or 0
        total_duration = stats.total_duration or 0
        total_duration_hours = round(total_duration / 3600, 1)
        
        # 查询平均分
        result = await db.execute(
            select(func.avg(PracticeReport.total_score)).where(
                PracticeReport.session_id.in_(
                    select(PracticeSession.session_id).where(
                        PracticeSession.user_id == current_user.id,
                        PracticeSession.is_deleted == False
                    )
                )
            )
        )
        avg_score = result.scalar() or 0
        avg_score = round(float(avg_score), 1) if avg_score else 0
        
        # 计算本月进步（简化：与上月平均分对比）
        # TODO: 实现真实的月度对比逻辑
        month_improvement = 15  # 暂时使用固定值
        
        logger.info(f"查询陪练统计: user_id={current_user.id}, total={total_count}, avg={avg_score}")
        
        return ResponseModel(
            code=200,
            message="success",
            data={
                "total_count": total_count,
                "avg_score": avg_score,
                "total_duration_hours": total_duration_hours,
                "month_improvement": month_improvement
            }
        )
    
    except Exception as e:
        logger.error(f"查询统计数据失败: {e}", exc_info=True)
        raise HTTPException(status_code=500, detail=f"查询失败: {str(e)}")