baodan/api/insurance/ppt/routes.py

"""PPT 生成模块 Blueprint 路由。"""
import os
import uuid
import json
import logging
from datetime import datetime
from flask import Blueprint, request, jsonify, send_file
from insurance.middleware.auth_middleware import jwt_required
from insurance.utils.response import success, error, ErrorCode

logger = logging.getLogger(__name__)

ppt_bp = Blueprint("ppt", __name__)


def _get_session(session_id: str, user_id: str):
    """获取会话记录。"""
    from insurance.db.compat import db
    from insurance.models.ppt_session import PptSession
    session = PptSession.query.filter_by(id=session_id).first()
    if not session:
        return None
    if session.user_id != user_id:
        return None
    return session


def _save_session(session):
    """保存会话记录。"""
    from insurance.db.compat import db
    db.session.add(session)
    db.session.commit()


# ─── 健康检查 ─────────────────────────────────────────────

@ppt_bp.route("/health", methods=["GET"])
def health():
    return success({"status": "ok"})


# ─── 渲染选项 ─────────────────────────────────────────────

@ppt_bp.route("/render-options", methods=["GET"])
def render_options():
    """获取可用的公司和模板风格列表。"""
    from insurance.models.ppt_config import PptCompany, PptProduct, PptScenario, PptTemplate
    companies = PptCompany.query.filter_by(status=1, deleted_at=None).order_by(
        PptCompany.sort_order.asc(), PptCompany.id.asc()
    ).all()
    products = PptProduct.query.filter_by(status=1, deleted_at=None).order_by(
        PptProduct.sort_order.asc(), PptProduct.id.asc()
    ).all()
    templates = PptTemplate.query.filter_by(status=1, deleted_at=None).order_by(
        PptTemplate.plan_type.asc(), PptTemplate.id.asc()
    ).all()
    scenarios = PptScenario.query.filter_by(status=1, deleted_at=None).order_by(
        PptScenario.sort_order.asc(), PptScenario.code.asc()
    ).all()
    scenario_map = {item.code: item.to_dict() for item in scenarios}
    template_items = []
    for template in templates:
        item = template.to_dict()
        scenario_config = scenario_map.get(template.scenario_tag) or {}
        item["scenarioBase"] = scenario_config.get("baseScenario")
        item["generationMode"] = scenario_config.get("generationMode")
        item["scenarioName"] = scenario_config.get("name")
        template_items.append(item)
    from insurance.ppt.masking import public_company_option, public_product_option
    return success({
        "companies": [public_company_option(c.to_dict()) for c in companies],
        "products": [public_product_option(p.to_dict()) for p in products],
        "templates": template_items,
        "scenarios": [item.to_dict() for item in scenarios],
    })


# ─── 上传 PDF ─────────────────────────────────────────────

@ppt_bp.route("/upload", methods=["POST"])
@jwt_required
def upload_pdfs():
    """上传 PDF 文件并创建会话。

    支持 1～10 份计划书，五组并行数组(files/types/companies/products/passwords)长度必须一致。
    采用"先校验后写入"的原子上传策略：任一文件校验失败时不创建会话。
    """
    from flask import jsonify as _jsonify
    MAX_UPLOAD_FILES = 10
    VALID_TYPES = {"savings", "ci", "iul"}

    user_id = str(getattr(request, "user_id", "guest"))
    files = request.files.getlist("files")
    types = request.form.getlist("types")
    companies = request.form.getlist("companies")
    products = request.form.getlist("products")
    passwords = request.form.getlist("passwords")

    # ── 基础参数校验 ──
    if not files:
        return error(ErrorCode.PARAM_ERROR, "请至少上传 1 份计划书")
    if len(files) > MAX_UPLOAD_FILES:
        return error(ErrorCode.PARAM_ERROR, f"最多支持上传 {MAX_UPLOAD_FILES} 份计划书")
    for name, arr in [("types", types), ("companies", companies),
                      ("products", products), ("passwords", passwords)]:
        if len(arr) != len(files):
            return error(ErrorCode.PARAM_ERROR, f"{name} 数组长度与 files 不一致")
    for i, t in enumerate(types):
        if t not in VALID_TYPES:
            return error(ErrorCode.PARAM_ERROR, f"第 {i + 1} 份文件的险种 '{t}' 不合法")

    # ── 第一阶段：校验全部文件，不写入磁盘 ──
    from insurance.utils.security import prepare_pdf_upload
    from insurance.models.ppt_config import PptCompany, PptProduct

    validated = []       # (f, pdf_bytes, file_record) 通过校验的文件
    file_errors = []     # 结构化错误列表

    for i, f in enumerate(files):
        fname = f.filename or f"文件{i + 1}"
        if not f.filename or not f.filename.lower().endswith(".pdf"):
            file_errors.append({"index": i, "fileName": fname, "field": "file", "message": "仅支持 PDF 格式"})
            continue

        plan_type = types[i]
        company_id = companies[i] if i < len(companies) else ""
        product_id = products[i] if i < len(products) else ""
        password = passwords[i] if i < len(passwords) else ""

        # 保司校验
        company = PptCompany.query.filter_by(
            id=company_id, status=1, deleted_at=None
        ).first() if company_id else None
        if company_id and not company:
            file_errors.append({"index": i, "fileName": fname, "field": "company", "message": "所选保司不存在或已停用"})
            continue

        # 产品校验
        product = PptProduct.query.filter_by(
            id=product_id, status=1, deleted_at=None
        ).first() if product_id else None
        if product_id and (
            not product
            or product.plan_type != plan_type
            or (company_id and product.company_id != company_id)
        ):
            file_errors.append({"index": i, "fileName": fname, "field": "product", "message": "所选产品与险种或保司不匹配"})
            continue
        if product and not company_id:
            company_id = product.company_id

        # PDF 安全校验
        is_valid, err_msg, pdf_bytes = prepare_pdf_upload(f, password)
        if not is_valid:
            file_errors.append({"index": i, "fileName": fname, "field": "password", "message": err_msg or "PDF 校验失败"})
            continue

        validated.append((f, pdf_bytes, {
            "name": fname,
            "type": plan_type,
            "companyId": company_id,
            "productId": product_id,
        }))

    # 任一文件失败则整体拒绝
    if file_errors:
        return _jsonify({
            "code": ErrorCode.FILE_FORMAT_ERROR,
            "message": "部分计划书校验失败",
            "data": {"fileErrors": file_errors},
        }), 400

    if not validated:
        return error(ErrorCode.FILE_FORMAT_ERROR, "无有效 PDF 文件")

    # ── 第二阶段：全部通过，统一写入磁盘并创建会话 ──
    from insurance.config import get_storage_root
    upload_dir = os.path.join(get_storage_root(), "uploads", "ppt", user_id)
    os.makedirs(upload_dir, exist_ok=True)

    file_records = []
    for f, pdf_bytes, meta in validated:
        filename = f"{uuid.uuid4().hex[:8]}_{meta['name']}"
        filepath = os.path.join(upload_dir, filename)
        with open(filepath, "wb") as output:
            output.write(pdf_bytes)
        file_records.append({
            "path": filepath,
            "name": meta["name"],
            "type": meta["type"],
            "companyId": meta["companyId"],
            "productId": meta["productId"],
        })

    session_id = uuid.uuid4().hex
    from insurance.models.ppt_session import PptSession
    session = PptSession(
        id=session_id,
        user_id=user_id,
        status="created",
        files_json=json.dumps(file_records, ensure_ascii=False),
    )
    _save_session(session)

    return success({
        "sessionId": session_id,
        "files": [f["name"] for f in file_records],
    })


# ─── 解析 PDF ─────────────────────────────────────────────

@ppt_bp.route("/parse/<session_id>", methods=["POST"])
@jwt_required
def parse_session(session_id):
    """触发 AI 解析 PDF（异步任务）。"""
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    files = json.loads(session.files_json) if session.files_json else []
    if not files:
        return error(ErrorCode.PARAM_ERROR, "没有可解析的 PDF 文件")

    if session.status == "parsing":
        _reconcile_parse_task(session)
        if session.status == "parsing":
            return success(_build_parse_status(session), "解析任务正在进行")

    # 创建异步任务
    from insurance.generation import task_service
    session.status = "parsing"
    session.parse_progress = 0
    session.parse_message = "解析任务已提交"
    session.parse_error = None
    session.parse_started_at = datetime.now()
    session.parse_finished_at = None
    session.extractions_json = json.dumps([], ensure_ascii=False)
    _save_session(session)

    result = task_service.create_task(
        user_id=user_id,
        artifact_type="ppt",
        operation="parse",
        workspace_id=session_id,
        title=session.title or f"PPT {session_id[:8]}",
        input_snapshot={"files": files},
        input_revision=getattr(session, "draft_revision", None) or 1,
    )

    if result.get("code") != 0:
        message = result.get("message", "解析任务提交失败")
        session.status = "error"
        session.parse_progress = 100
        session.parse_message = "处理失败"
        session.parse_error = message
        session.parse_finished_at = datetime.now()
        _save_session(session)
        return error(ErrorCode.SERVER_ERROR, message, status=503)

    session.latest_task_id = result["data"]["id"]
    _save_session(session)
    message = "解析任务已启动"

    return success({
        "sessionId": session_id,
        "status": session.status,
        "progress": session.parse_progress or 0,
        "message": message,
        "taskId": result.get("data", {}).get("id") if result.get("code") == 0 else None,
    })


@ppt_bp.route("/parse/<session_id>/status", methods=["GET"])
@jwt_required
def parse_status(session_id):
    """获取 AI 解析进度。"""
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")
    _reconcile_parse_task(session)
    return success(_build_parse_status(session))


def _reconcile_parse_task(session):
    """修正任务表与 PPT 会话不一致的终态。"""
    if session.status != "parsing" or not session.latest_task_id:
        return

    from insurance.models.generation_task import GenerationTask

    task = GenerationTask.query.get(session.latest_task_id)
    if not task or task.status not in ("failed", "cancelled"):
        return

    session.status = "error"
    session.parse_progress = 100
    session.parse_message = "处理失败"
    session.parse_error = task.error_message or (
        "任务已取消" if task.status == "cancelled" else "解析任务失败，请重新提交"
    )
    session.parse_finished_at = task.finished_at or datetime.now()
    _save_session(session)


def _build_parse_status(session):
    extractions = json.loads(session.extractions_json) if session.extractions_json else []
    return {
        "sessionId": session.id,
        "status": session.status,
        "progress": session.parse_progress or 0,
        "message": session.parse_message or "",
        "error": session.parse_error,
        "extractions": [{
            "pdfName": e.get("pdfName", ""),
            "planType": e.get("planType", ""),
            "status": e.get("status", ""),
            "productName": e.get("productName", ""),
            "yearCount": e.get("yearCount", 0),
            "error": e.get("error"),
        } for e in extractions],
    }


def _reassess_extraction(ext: dict):
    from insurance.ppt.extraction import assess_extraction_payload, infer_plan_type

    data = ext.get("data")
    if not data:
        return

    plan_type = infer_plan_type(data)
    status, extraction_error = assess_extraction_payload(data, plan_type)
    data["product_type"] = plan_type
    ext["planType"] = plan_type
    ext["status"] = status
    ext["productName"] = (data.get("product_name") or "").strip() or "unknown"
    ext["error"] = extraction_error or None
    rows = data.get("benefit_illustration") or data.get("benefitRows") or []
    ext["yearCount"] = len(rows) if isinstance(rows, list) else 0


# ─── 获取会话状态 ─────────────────────────────────────────

@ppt_bp.route("/session/<session_id>", methods=["GET"])
@jwt_required
def get_session(session_id):
    """获取完整会话状态。"""
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    return success(session.to_dict())


# ─── 对话 ─────────────────────────────────────────────────

@ppt_bp.route("/chat/<session_id>", methods=["POST"])
@jwt_required
def chat(session_id):
    """AI 保险顾问对话。"""
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    data = request.get_json(silent=True) or {}
    message = data.get("message", "").strip()
    if not message:
        return error(ErrorCode.PARAM_ERROR, "消息不能为空")

    import asyncio
    from insurance.ppt.llm_client import llm_client

    # 构建上下文
    extractions = json.loads(session.extractions_json) if session.extractions_json else []
    context_parts = []
    for ext in extractions:
        if ext.get("data"):
            context_parts.append(f"产品: {ext['productName']}, 类型: {ext['planType']}")
            data_inner = ext["data"]
            policy = data_inner.get("policy", {})
            context_parts.append(f"年缴保费: {policy.get('annual_premium', 'N/A')}")
            context_parts.append(f"缴费年期: {policy.get('premium_payment_period', 'N/A')}")

    system_prompt = (
        "你是一位资深的香港保险顾问，擅长为保险经纪人分析保险计划书。"
        "请基于以下保单数据，用温暖、专业、数据驱动的方式回答问题。\n\n"
        f"保单数据：\n{''.join(context_parts)}"
    )

    try:
        response = asyncio.run(llm_client.chat(message, system_prompt))
        reply = response.content
    except Exception as e:
        reply = f"抱歉，暂时无法回答。错误信息：{e}"

    # 更新对话历史
    history = json.loads(session.chat_history_json) if session.chat_history_json else []
    history.append({"role": "user", "content": message})
    history.append({"role": "assistant", "content": reply})
    # 保留最近 20 条
    history = history[-20:]
    session.chat_history_json = json.dumps(history, ensure_ascii=False)
    _save_session(session)

    return success({
        "sessionId": session_id,
        "message": reply,
        "history": history,
    })


# ─── 生成 PPT ─────────────────────────────────────────────

@ppt_bp.route("/generate/<session_id>", methods=["POST"])
@jwt_required
def generate_ppt(session_id):
    """生成 PPT（异步任务）。

    创建任务后立即返回 202，前端通过任务接口轮询状态。
    """
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    data = request.get_json(silent=True) or {}
    theme = data.get("theme") or data.get("style") or "broker"
    requested_company_id = data.get("companyId", "")
    template_id = data.get("templateId", "")
    files = json.loads(session.files_json) if session.files_json else []
    product_ids = [item.get("productId") for item in files if item.get("productId")]
    company_ids = list(dict.fromkeys(
        item.get("companyId") for item in files if item.get("companyId")
    ))
    if not company_ids and requested_company_id:
        company_ids = [requested_company_id]

    from insurance.models.ppt_config import PptCompany
    configured_companies = (
        PptCompany.query.filter(
            PptCompany.id.in_(company_ids),
            PptCompany.status == 1,
            PptCompany.deleted_at.is_(None),
        ).all()
        if company_ids else []
    )
    companies_by_id = {item.id: item for item in configured_companies}
    missing_company_ids = [item for item in company_ids if item not in companies_by_id]
    if missing_company_ids:
        return error(ErrorCode.PARAM_ERROR, "所选保司不存在或已停用")
    configured_companies = [companies_by_id[item] for item in company_ids]
    company_id = company_ids[0] if len(company_ids) == 1 else ""

    from insurance.models.ppt_config import PptProduct
    configured_products = (
        PptProduct.query.filter(
            PptProduct.id.in_(product_ids),
            PptProduct.status == 1,
            PptProduct.deleted_at.is_(None),
        ).all()
        if product_ids else []
    )
    from insurance.ppt.masking import build_brand_policy
    brand_policy = build_brand_policy(
        [company.to_dict() for company in configured_companies],
        [product.to_dict() for product in configured_products],
    )

    template_data = None
    template_asset_snapshot = None
    if template_id:
        from insurance.models.ppt_config import PptTemplate
        from insurance.ppt.comparison import detect_generation_scenario
        template = PptTemplate.query.filter_by(
            id=template_id, status=1, deleted_at=None
        ).first()
        if not template:
            return error(ErrorCode.PARAM_ERROR, "所选 PPT 模板不存在或已停用")
        template_data = template.to_dict()
        if template.source_template_asset_id:
            from insurance.ppt.template_asset_service import (
                resolve_template_asset,
                template_asset_sha256,
            )
            try:
                asset_path = resolve_template_asset(template.source_template_asset_id)
                asset_sha256 = template.asset_sha256 or (
                    template_asset_sha256(asset_path) if asset_path and os.path.isfile(asset_path) else None
                )
            except (OSError, ValueError):
                asset_path = None
                asset_sha256 = None
            if not asset_path or not asset_sha256:
                return error(ErrorCode.PARAM_ERROR, "所选 PPT 模板文件不存在或校验失败")
            template_asset_snapshot = {
                "assetId": template.source_template_asset_id,
                "version": template.asset_version or 1,
                "sha256": asset_sha256,
                "rendererMode": template.clone_renderer or "clone-edit-v2",
            }
        file_kinds = [
            {"kind": item.get("type")}
            for item in files
            if item.get("type")
        ]
        file_scenario = detect_generation_scenario(file_kinds) if file_kinds else ""
        primary_plan_type = file_kinds[0]["kind"] if file_kinds else ""
        from insurance.ppt.scenarios import template_scenario_compatible
        if (
            template.scenario_tag
            and file_scenario
            and not template_scenario_compatible(template.scenario_tag, file_scenario)
        ):
            return error(ErrorCode.PARAM_ERROR, "所选 PPT 模板不适用于当前计划书组合")
        if (
            not template.scenario_tag
            and primary_plan_type
            and template.plan_type != primary_plan_type
        ):
            return error(ErrorCode.PARAM_ERROR, "所选 PPT 模板不适用于当前险种")
        applicable_companies = template_data.get("applicableCompanyIds") or []
        applicable_products = template_data.get("applicableProductIds") or []
        if applicable_companies and any(
            item not in applicable_companies for item in company_ids
        ):
            return error(ErrorCode.PARAM_ERROR, "所选 PPT 模板不适用于当前保司")
        if applicable_products and not set(product_ids).intersection(applicable_products):
            return error(ErrorCode.PARAM_ERROR, "所选 PPT 模板不适用于当前产品")
        theme = template.style_preset

    extractions = json.loads(session.extractions_json) if session.extractions_json else []
    if not extractions:
        return error(ErrorCode.PARAM_ERROR, "无解析数据")

    # 快速校验：确保有可生成的数据
    has_valid = any(
        e.get("status") in ("success", "partial") and e.get("data")
        for e in extractions
    )
    if not has_valid:
        return error(ErrorCode.PARAM_ERROR, "无有效提取数据")

    from insurance.ppt.comparison import detect_generation_scenario
    scenario = detect_generation_scenario([
        {"kind": e.get("planType") or (e.get("data") or {}).get("product_type")}
        for e in extractions
        if e.get("status") in ("success", "partial") and e.get("data")
    ])

    # 更新会话草稿选项
    import uuid
    session.draft_options_json = json.dumps({
        "theme": theme, "templateId": template_id,
        "templateName": (template_data or {}).get("name") or template_id,
        "stylePreset": (template_data or {}).get("stylePreset") or theme,
        "companyId": company_id, "companyIds": company_ids, "productIds": product_ids,
        "brandPolicy": brand_policy,
        "scenario": scenario,
    }, ensure_ascii=False)
    session.draft_revision = (session.draft_revision or 1) + 1
    _save_session(session)

    # 创建异步任务
    from insurance.generation import task_service
    result = task_service.create_task(
        user_id=user_id,
        artifact_type="ppt",
        operation="generate",
        workspace_id=session_id,
        title=session.title or f"PPT {session_id[:8]}",
        input_snapshot={
            "theme": theme,
            "templateId": template_id,
            "companyId": company_id,
            "companyIds": company_ids,
            "productIds": product_ids,
            "brandPolicy": brand_policy,
            "scenario": scenario,
            "templateAsset": template_asset_snapshot,
        },
        input_revision=session.draft_revision or 1,
        idempotency_key=f"ppt_gen_{session_id}_{session.draft_revision}",
    )

    if result.get("code") != 0:
        return error(ErrorCode.PARAM_ERROR, result.get("message", "创建任务失败"))

    task_data = result["data"]
    session.latest_task_id = task_data["id"]
    _save_session(session)

    from flask import jsonify, make_response
    resp = make_response(jsonify({
        "code": 0,
        "data": {
            "taskId": task_data["id"],
            "status": "queued",
            "sessionId": session_id,
            "scenario": scenario,
            "pollUrl": f"/insurance/workspace/tasks/{task_data['id']}",
        },
    }))
    resp.status_code = 202
    return resp


# ─── 下载 PPT ─────────────────────────────────────────────

@ppt_bp.route("/download/<session_id>", methods=["GET"])
@jwt_required
def download_ppt(session_id):
    """下载生成的 PPT。"""
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    if not session.ppt_path or not os.path.exists(session.ppt_path):
        return error(ErrorCode.NOT_FOUND, "PPT 文件不存在")

    # 记录下载历史
    try:
        _record_history(
            user_id=user_id,
            action_type="download",
            session_id=session_id,
            file_url=session.ppt_path,
        )
    except Exception:
        logger.warning("记录下载历史失败", exc_info=True)

    return send_file(
        session.ppt_path,
        as_attachment=True,
        download_name=f"{session_id}.pptx",
        mimetype="application/vnd.openxmlformats-officedocument.presentationml.presentation",
    )


# ─── 验证提取数据 ─────────────────────────────────────────

@ppt_bp.route("/validate/<session_id>", methods=["GET"])
@jwt_required
def validate_extraction(session_id):
    """验证提取数据完整性。"""
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    extractions = json.loads(session.extractions_json) if session.extractions_json else []
    all_issues = []

    from insurance.ppt.normalizer import normalize_savings_plan, normalize_ci_plan, normalize_iul_plan
    from insurance.ppt.validator import validate_formal_savings_plan, validate_formal_ci_plan, validate_formal_iul_plan

    for idx, ext in enumerate(extractions):
        pdf_name = ext.get("pdfName", f"文件{idx + 1}")
        product_name = ext.get("productName", "unknown")

        # 解析失败的文件必须产生阻断问题，不能静默跳过
        if ext.get("status") not in ("success", "partial") or not ext.get("data"):
            error_detail = ext.get("error", "未知错误")
            all_issues.append({
                "extractionIndex": idx,
                "pdfName": pdf_name,
                "field": "EXTRACTION_FAILED",
                "path": "",
                "section": "fields",
                "severity": "error",
                "message": f"{pdf_name}（{product_name}）解析失败：{error_detail}",
                "suggestedAction": "review",
                "state": "unresolved",
            })
            continue

        data = ext["data"]
        plan_type = (ext.get("planType") or data.get("product_type") or "savings").lower()
        pdf_path = ext.get("pdfPath")
        try:
            if plan_type == "ci":
                normalized = normalize_ci_plan(data, pdf_path)
                issues = validate_formal_ci_plan(normalized)
            elif plan_type == "iul":
                normalized = normalize_iul_plan(data, pdf_path)
                issues = validate_formal_iul_plan(normalized)
            else:
                normalized = normalize_savings_plan(data, pdf_path)
                issues = validate_formal_savings_plan(normalized)
            all_issues.extend([{
                "extractionIndex": idx,
                "pdfName": pdf_name,
                "field": i.code,
                "path": i.path or "",
                "section": i.section or "",
                "severity": i.level,
                "message": i.message,
                "suggestedAction": i.suggested_action or "review",
                "state": "unresolved",
            } for i in issues])
        except Exception as e:
            all_issues.append({
                "extractionIndex": idx,
                "pdfName": pdf_name,
                "field": "general",
                "severity": "error",
                "message": str(e),
            })

    # ── 跨文件兼容性校验（同险种比较场景） ──
    valid_extractions = [
        e for e in extractions
        if e.get("status") in ("success", "partial") and e.get("data")
    ]
    if len(valid_extractions) >= 2:
        try:
            from insurance.ppt.comparison import (
                detect_generation_scenario,
                generation_mode_for_scenario,
                build_comparison_contract,
                SCENARIO_GENERIC_COMPARE,
                SCENARIO_MULTI_SAVINGS,
            )
            normalized_for_check = []
            from insurance.ppt.normalizer import normalize_savings_plan as _ns, normalize_ci_plan as _nc, normalize_iul_plan as _ni
            for ext in valid_extractions:
                data = ext["data"]
                pt = (ext.get("planType") or data.get("product_type") or "savings").lower()
                try:
                    if pt == "ci":
                        n = _nc(data, ext.get("pdfPath"))
                    elif pt == "iul":
                        n = _ni(data, ext.get("pdfPath"))
                    else:
                        n = _ns(data, ext.get("pdfPath"))
                    n["kind"] = pt
                    normalized_for_check.append(n)
                except Exception:
                    pass

            if len(normalized_for_check) >= 2:
                scenario = detect_generation_scenario(normalized_for_check)
                mode = generation_mode_for_scenario(scenario)
                try:
                    contract = build_comparison_contract(normalized_for_check, mode=mode)
                    for w in contract.get("warnings", []):
                        all_issues.append({"field": "comparison", "severity": "warn", "message": w})
                except ValueError as ve:
                    all_issues.append({"field": "comparison", "severity": "error", "message": str(ve)})
        except Exception as e:
            logger.warning("跨文件兼容性校验异常: %s", e)

    error_count = sum(1 for i in all_issues if i["severity"] == "error")
    warn_count = sum(1 for i in all_issues if i["severity"] == "warn")
    info_count = sum(1 for i in all_issues if i["severity"] == "info")
    return success({
        "sessionId": session_id,
        "validated": error_count == 0,
        "canProceed": error_count == 0,
        "blockerCount": error_count,
        "warningCount": warn_count,
        "infoCount": info_count,
        "errorCount": error_count,
        "warnCount": warn_count,
        "issues": all_issues,
    })


# ─── 更新提取数据 ─────────────────────────────────────────

@ppt_bp.route("/session/<session_id>/extractions", methods=["PUT"])
@jwt_required
def update_extractions(session_id):
    """保存用户修改后的提取数据。"""
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    data = request.get_json(silent=True) or {}
    extractions = data.get("extractions")
    if not isinstance(extractions, list):
        return error(ErrorCode.PARAM_ERROR, "extractions 必须是数组")

    # 合并更新：只更新 data 字段，保留 pdfPath/status 等元信息
    existing = json.loads(session.extractions_json) if session.extractions_json else []
    existing_map = {e["pdfName"]: e for e in existing}

    for ext in extractions:
        pdf_name = ext.get("pdfName")
        if not pdf_name or pdf_name not in existing_map:
            continue
        # 更新数据字段
        if "data" in ext:
            existing_map[pdf_name]["data"] = ext["data"]
        if "productName" in ext:
            existing_map[pdf_name]["productName"] = ext["productName"]
        if "planType" in ext:
            existing_map[pdf_name]["planType"] = ext["planType"]
        _reassess_extraction(existing_map[pdf_name])

    updated = list(existing_map.values())
    session.extractions_json = json.dumps(updated, ensure_ascii=False)
    session.status = "parsed"  # 回到 parsed 状态，需要重新生成
    _save_session(session)

    return success({
        "sessionId": session_id,
        "status": "updated",
        "extractions": [{
            "pdfName": e["pdfName"],
            "planType": e["planType"],
            "status": e["status"],
            "productName": e["productName"],
            "yearCount": e["yearCount"],
        } for e in updated],
    })


# ─── 公司知识库匹配 ───────────────────────────────────────

@ppt_bp.route("/company-kb/match", methods=["POST"])
@jwt_required
def match_company():
    """匹配公司知识库。"""
    data = request.get_json(silent=True) or {}
    product_name = data.get("productName")
    company_hint = data.get("companyHint")
    forced_company_id = data.get("companyId")

    from insurance.models.ppt_config import PptCompany, PptProduct
    from insurance.ppt.knowledge import match_company_knowledge

    companies = [c.to_dict() for c in PptCompany.query.all()]
    products = [p.to_dict() for p in PptProduct.query.all()]

    result = match_company_knowledge(
        product_name=product_name,
        company_hint=company_hint,
        forced_company_id=forced_company_id,
        companies=companies,
        products=products,
    )

    return success(result)


# ─── 历史记录 ─────────────────────────────────────────────

def _record_history(user_id, action_type, session_id=None, company_id=None,
                    product_id=None, template_id=None, content_snapshot=None,
                    file_url=None):
    """写入历史记录（内部函数）。"""
    from insurance.db.compat import db
    from insurance.models.ppt_history import PptHistory
    from flask import request as req
    record = PptHistory(
        user_id=user_id,
        session_id=session_id,
        action_type=action_type,
        company_id=company_id,
        product_id=product_id,
        template_id=template_id,
        content_snapshot=json.dumps(content_snapshot, ensure_ascii=False) if content_snapshot else None,
        file_url=file_url,
        ip=req.remote_addr,
        user_agent=req.headers.get("User-Agent", "")[:500],
    )
    db.session.add(record)
    db.session.commit()


@ppt_bp.route("/history", methods=["GET"])
@jwt_required
def list_history():
    """当前用户的历史记录列表。"""
    from insurance.db.compat import db
    from insurance.models.ppt_history import PptHistory

    user_id = str(getattr(request, "user_id", "guest"))
    page = max(1, request.args.get("page", 1, type=int))
    page_size = min(100, max(1, request.args.get("page_size", 20, type=int)))
    company_id = request.args.get("company_id", "")
    action_type = request.args.get("action_type", "")

    query = db.session.query(PptHistory).filter(PptHistory.user_id == user_id)
    if company_id:
        query = query.filter(PptHistory.company_id == company_id)
    if action_type:
        query = query.filter(PptHistory.action_type == action_type)
    query = query.order_by(PptHistory.created_at.desc())
    total = query.count()
    items = query.offset((page - 1) * page_size).limit(page_size).all()

    return success({
        "total": total,
        "items": [h.to_dict() for h in items],
    })


@ppt_bp.route("/history/<int:history_id>", methods=["GET"])
@jwt_required
def get_history(history_id):
    """单条历史详情。"""
    from insurance.models.ppt_history import PptHistory
    user_id = str(getattr(request, "user_id", "guest"))
    record = PptHistory.query.get(history_id)
    if not record or record.user_id != user_id:
        return error(ErrorCode.NOT_FOUND, "记录不存在")
    return success(record.to_dict())


@ppt_bp.route("/history/<int:history_id>/re-download", methods=["GET", "POST"])
@jwt_required
def re_download(history_id):
    """重新下载历史文件。"""
    from insurance.models.ppt_history import PptHistory
    user_id = str(getattr(request, "user_id", "guest"))
    record = PptHistory.query.get(history_id)
    if not record or record.user_id != user_id:
        return error(ErrorCode.NOT_FOUND, "记录不存在")
    if not record.file_url or not os.path.exists(record.file_url):
        return error(ErrorCode.NOT_FOUND, "文件不存在")
    return send_file(record.file_url, as_attachment=True)


# ─── 预览接口 ─────────────────────────────────────────────

@ppt_bp.route("/preview/<session_id>", methods=["GET"])
@jwt_required
def get_preview(session_id):
    """获取幻灯片预览数据（结构化 JSON + 质量报告）。"""
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    versions = json.loads(session.versions_json) if session.versions_json else []
    requested_revision = request.args.get("revision", type=int)
    selected_version = next(
        (item for item in versions if item.get("revision") == requested_revision),
        None,
    ) if requested_revision is not None else None
    if requested_revision is not None and not selected_version:
        return error(ErrorCode.NOT_FOUND, "PPT 版本不存在")

    preview_status = session.preview_status or "none"
    slides_data = None
    slides_json_path = (
        selected_version.get("slidesJsonPath")
        if selected_version else session.slides_json_path
    )

    if slides_json_path:
        if os.path.exists(slides_json_path):
            try:
                with open(slides_json_path, "r", encoding="utf-8") as f:
                    slides_data = json.load(f)
            except Exception as e:
                logger.warning("读取 slides.json 失败: %s", e)
                preview_status = "failed"
        else:
            logger.warning("slides.json 已被清理: %s", slides_json_path)
            preview_status = "failed"

    quality_report = None
    if session.quality_report_json:
        try:
            quality_report = json.loads(session.quality_report_json)
        except Exception:
            pass

    return success({
        "sessionId": session_id,
        "previewStatus": preview_status,
        "slides": slides_data,
        "qualityReport": quality_report,
        "slideCount": session.slide_count or 0,
        "versions": versions,
        "generatedRevision": session.generated_revision or 0,
        "viewingRevision": (
            requested_revision
            if requested_revision is not None else session.generated_revision or 0
        ),
        "generationConfig": (
            json.loads(session.draft_options_json)
            if session.draft_options_json else {}
        ),
    })


@ppt_bp.route("/preview/<session_id>/versions/<int:revision>/restore", methods=["POST"])
@jwt_required
def restore_preview_version(session_id, revision):
    """把历史版本复制为新的当前版本，保留原历史文件。"""
    import shutil
    import uuid
    from insurance.config import get_storage_root
    from insurance.db.compat import db

    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")
    versions = json.loads(session.versions_json) if session.versions_json else []
    source = next((item for item in versions if item.get("revision") == revision), None)
    if not source:
        return error(ErrorCode.NOT_FOUND, "PPT 版本不存在")

    ppt_path = source.get("path")
    slides_path = source.get("slidesJsonPath")
    if not ppt_path or not os.path.exists(ppt_path):
        return error(ErrorCode.NOT_FOUND, "历史版本文件已被清理")

    output_root = os.path.abspath(os.path.join(get_storage_root(), "outputs", "ppt"))
    source_abs = os.path.abspath(ppt_path)
    if not source_abs.startswith(output_root):
        return error(ErrorCode.PARAM_ERROR, "历史版本路径无效")

    new_revision = max(
        [int(item.get("revision") or 0) for item in versions] + [session.generated_revision or 0]
    ) + 1
    output_dir = os.path.join(
        output_root, str(user_id), f"restore_{uuid.uuid4().hex[:12]}"
    )
    os.makedirs(output_dir, exist_ok=True)
    target_ppt = os.path.join(output_dir, "presentation.pptx")
    shutil.copy2(source_abs, target_ppt)
    target_slides = None
    if slides_path and os.path.exists(slides_path):
        slides_abs = os.path.abspath(slides_path)
        if not slides_abs.startswith(output_root):
            return error(ErrorCode.PARAM_ERROR, "历史预览路径无效")
        slides_dir = os.path.join(output_dir, "slides")
        os.makedirs(slides_dir, exist_ok=True)
        target_slides = os.path.join(slides_dir, "slides.json")
        shutil.copy2(slides_abs, target_slides)

    restored = {
        "revision": new_revision,
        "path": target_ppt,
        "slidesJsonPath": target_slides,
        "deckPath": source.get("deckPath"),
        "slideCount": source.get("slideCount") or 0,
        "sourceRevision": revision,
        "source": "restore",
        "createdAt": __import__("datetime").datetime.now().isoformat(),
    }
    versions.append(restored)
    session.versions_json = json.dumps(versions, ensure_ascii=False)
    session.ppt_path = target_ppt
    session.latest_output_path = target_ppt
    session.slides_json_path = target_slides
    session.slide_count = restored["slideCount"]
    session.preview_status = "ready" if target_slides else "failed"
    session.generated_revision = new_revision
    session.draft_revision = max(session.draft_revision or 1, new_revision)
    db.session.commit()
    _record_history(
        user_id=user_id,
        action_type="restore_version",
        session_id=session_id,
        content_snapshot={"sourceRevision": revision, "newRevision": new_revision},
        file_url=target_ppt,
    )
    return success({
        "sessionId": session_id,
        "revision": new_revision,
        "sourceRevision": revision,
    })


@ppt_bp.route("/preview/<session_id>/slide/<int:index>", methods=["PUT"])
@jwt_required
def update_slide(session_id, index):
    """更新指定页的编辑内容。"""
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    if not session.slides_json_path or not os.path.exists(session.slides_json_path):
        return error(ErrorCode.NOT_FOUND, "预览数据不存在")

    data = request.get_json(silent=True) or {}
    shapes = data.get("shapes")
    hidden = data.get("hidden")

    try:
        with open(session.slides_json_path, "r", encoding="utf-8") as f:
            slides_data = json.load(f)
    except Exception:
        return error(ErrorCode.SERVER_ERROR, "读取预览数据失败")

    slides = slides_data.get("slides", [])
    if index < 0 or index >= len(slides):
        return error(ErrorCode.PARAM_ERROR, f"页码超出范围 (0-{len(slides) - 1})")

    if shapes is not None:
        if not isinstance(shapes, list):
            return error(ErrorCode.PARAM_ERROR, "shapes 必须是数组")
        slides[index]["shapes"] = shapes
    if hidden is not None:
        slides[index]["hidden"] = bool(hidden)

    try:
        with open(session.slides_json_path, "w", encoding="utf-8") as f:
            json.dump(slides_data, f, ensure_ascii=False)
    except Exception:
        return error(ErrorCode.SERVER_ERROR, "保存预览数据失败")

    # 记录编辑历史
    try:
        _record_history(
            user_id=user_id,
            action_type="edit_slide",
            session_id=session_id,
            content_snapshot={"slideIndex": index, "shapeCount": len(shapes)},
        )
    except Exception:
        logger.warning("记录编辑历史失败", exc_info=True)

    return success({"sessionId": session_id, "slideIndex": index, "updated": True})


@ppt_bp.route("/preview/<session_id>/quality-confirm", methods=["PUT"])
@jwt_required
def update_quality_confirm(session_id):
    """更新人工质量确认项。"""
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    if not session.quality_report_json:
        return error(ErrorCode.NOT_FOUND, "质量报告不存在")

    data = request.get_json(silent=True) or {}
    key = data.get("key", "").strip()
    confirmed = bool(data.get("confirmed"))

    if not key:
        return error(ErrorCode.PARAM_ERROR, "key 必填")

    try:
        report = json.loads(session.quality_report_json)
    except Exception:
        return error(ErrorCode.SERVER_ERROR, "质量报告解析失败")

    # 更新人工确认项
    updated = False
    for item in report.get("manual", []):
        if item["key"] == key:
            item["confirmed"] = confirmed
            updated = True
            break

    if not updated:
        return error(ErrorCode.PARAM_ERROR, f"未知的确认项: {key}")

    # 更新待确认计数
    report["summary"]["manualPending"] = sum(
        1 for m in report.get("manual", []) if not m.get("confirmed")
    )

    session.quality_report_json = json.dumps(report, ensure_ascii=False)
    from insurance.db.compat import db
    db.session.commit()

    return success({"sessionId": session_id, "key": key, "confirmed": confirmed})


@ppt_bp.route("/preview/<session_id>/regenerate", methods=["POST"])
@jwt_required
def regenerate_version(session_id):
    """基于当前编辑内容重新生成新版本 PPT。

    不覆盖当前版本，生成后追加到版本历史。
    """
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    if not session.ppt_path or not os.path.exists(session.ppt_path):
        return error(ErrorCode.PARAM_ERROR, "当前无可用的 PPT 文件")

    # 检查是否有运行中的任务
    from insurance.generation import task_service
    if session.latest_task_id:
        from insurance.models.generation_task import GenerationTask
        active = GenerationTask.query.filter(
            GenerationTask.id == session.latest_task_id,
            GenerationTask.status.in_(["queued", "running"]),
        ).first()
        if active:
            return error(ErrorCode.PARAM_ERROR, "有正在执行的任务，请等待完成")

    # 读取当前编辑后的 slides.json
    edits_data = None
    if session.slides_json_path and os.path.exists(session.slides_json_path):
        try:
            with open(session.slides_json_path, "r", encoding="utf-8") as f:
                edits_data = json.load(f)
        except Exception:
            pass

    # 递增版本号
    session.draft_revision = (session.draft_revision or 1) + 1
    session.status = "generating"
    _save_session(session)

    # 创建异步任务，传递编辑数据
    result = task_service.create_task(
        user_id=user_id,
        artifact_type="ppt",
        operation="regenerate",
        workspace_id=session_id,
        title=session.title or f"PPT {session_id[:8]}",
        input_snapshot={
            "revision": session.draft_revision,
            "edits": edits_data,
        },
        input_revision=session.draft_revision or 1,
        idempotency_key=f"ppt_regen_{session_id}_{session.draft_revision}",
    )

    if result.get("code") != 0:
        return error(ErrorCode.PARAM_ERROR, result.get("message", "创建任务失败"))

    task_data = result["data"]
    session.latest_task_id = task_data["id"]
    _save_session(session)

    return success({
        "taskId": task_data["id"],
        "status": "queued",
        "sessionId": session_id,
        "revision": session.draft_revision,
    })


# ─── PDF 页面预览 ─────────────────────────────────────────

@ppt_bp.route("/pdf-preview/<session_id>/<int:extraction_index>/<int:page_number>", methods=["GET"])
@jwt_required
def pdf_page_preview(session_id, extraction_index, page_number):
    """返回 PDF 指定页的 PNG 预览图。

    用于核对页面展示 PDF 原文证据。
    """
    import tempfile
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    extractions = json.loads(session.extractions_json) if session.extractions_json else []
    if extraction_index < 0 or extraction_index >= len(extractions):
        return error(ErrorCode.PARAM_ERROR, "提取索引越界")

    ext = extractions[extraction_index]
    pdf_path = ext.get("pdfPath")
    if not pdf_path or not os.path.exists(pdf_path):
        return error(ErrorCode.NOT_FOUND, "PDF 文件不存在")

    try:
        try:
            import fitz
        except ImportError:
            import pymupdf as fitz

        doc = fitz.open(pdf_path)
        if page_number < 1 or page_number > len(doc):
            doc.close()
            return error(ErrorCode.PARAM_ERROR, f"页码越界（共 {len(doc)} 页）")

        page = doc[page_number - 1]  # 1-based to 0-based
        # 渲染为 PNG，2x 缩放（约 150 DPI，平衡质量和大小）
        pixmap = page.get_pixmap(matrix=fitz.Matrix(2.0, 2.0))
        doc.close()

        # 写入临时文件返回
        img_bytes = pixmap.tobytes("png")
        import io
        buf = io.BytesIO(img_bytes)
        buf.seek(0)
        return send_file(
            buf,
            mimetype="image/png",
            as_attachment=False,
            download_name=f"page-{page_number}.png",
        )
    except ImportError:
        return error(ErrorCode.INTERNAL_ERROR, "PDF 渲染库未安装（需要 PyMuPDF）")
    except Exception as e:
        logger.error(f"PDF 页面预览失败: {e}")
        return error(ErrorCode.INTERNAL_ERROR, f"PDF 渲染失败: {str(e)[:100]}")


@ppt_bp.route("/pdf-info/<session_id>/<int:extraction_index>", methods=["GET"])
@jwt_required
def pdf_info(session_id, extraction_index):
    """返回 PDF 基本信息（总页数等）。"""
    user_id = str(getattr(request, "user_id", "guest"))
    session = _get_session(session_id, user_id)
    if not session:
        return error(ErrorCode.NOT_FOUND, "会话不存在")

    extractions = json.loads(session.extractions_json) if session.extractions_json else []
    if extraction_index < 0 or extraction_index >= len(extractions):
        return error(ErrorCode.PARAM_ERROR, "提取索引越界")

    ext = extractions[extraction_index]
    pdf_path = ext.get("pdfPath")
    if not pdf_path or not os.path.exists(pdf_path):
        return error(ErrorCode.NOT_FOUND, "PDF 文件不存在")

    try:
        try:
            import fitz
        except ImportError:
            import pymupdf as fitz
        doc = fitz.open(pdf_path)
        page_count = len(doc)
        doc.close()
        return success({"pageCount": page_count, "pdfName": ext.get("pdfName", "")})
    except ImportError:
        return error(ErrorCode.INTERNAL_ERROR, "PDF 渲染库未安装")
    except Exception as e:
        return error(ErrorCode.INTERNAL_ERROR, str(e)[:100])