From 87ead4fa6aee4f421422908c3a5aa070264d89ed Mon Sep 17 00:00:00 2001 From: panda <1415243231@qq.com> Date: Wed, 20 May 2026 12:04:02 +0800 Subject: [PATCH] =?UTF-8?q?feat:=20=E5=AF=B9=E8=AF=9D=E5=8C=BA=E5=9F=9F?= =?UTF-8?q?=E6=96=87=E4=BB=B6=E4=B8=8A=E4=BC=A0(=E7=B2=98=E8=B4=B4/?= =?UTF-8?q?=E6=8B=96=E6=8B=BD)=20+=20XLSX=E6=94=AF=E6=8C=81=20+=20?= =?UTF-8?q?=E4=BC=9A=E8=AF=9D=E5=88=87=E6=8D=A2=E6=97=A0=E9=99=90=E5=BE=AA?= =?UTF-8?q?=E7=8E=AF=E4=BF=AE=E5=A4=8D?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - 对话区域: st.file_uploader + 全局 paste/drop 事件监听 + sessionStorage 桥接 - 文件预览芯片: 上传后显示在对话区域，可逐文件移除 - OCR 双层解析全面接入: file_parser(文字) + ocr_extractor(字段提取) - XLSX 解析: openpyxl 逐工作表/逐行读取 - 修复: create_session 强制写入 agent_state.session_id - 修复: load_session_node 不再从磁盘覆盖 session_id - 修复: 切换会话 _last_switched_to 哨兵防止无限 rerun Co-Authored-By: Claude Opus 4.7 --- CLAUDE.md | 20 +++- agent/nodes.py | 6 +- app.py | 234 +++++++++++++++++++++++++++++++++++++---- backend/file_parser.py | 32 ++++++ backend/session.py | 4 +- requirements.txt | 1 + 6 files changed, 268 insertions(+), 29 deletions(-) diff --git a/CLAUDE.md b/CLAUDE.md index bdfc27e..9c30238 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -70,7 +70,7 @@ agent/graph.py (LangGraph 状态机) | 文件 | 职责 | 修改频率 | |------|------|---------| -| `app.py` | Streamlit UI 入口，聊天界面 + 侧边栏 + 下载 + 文件上传 | **高** | +| `app.py` | Streamlit UI 入口，聊天界面 + 对话文件上传(粘贴/拖拽) + 侧边栏 + 下载 | **高** | | `agent/state.py` | AgentState 类型定义（~24 字段，含 pending_failure_context） | 低 | | `agent/nodes.py` | 14 个工作流节点 + 流式生成 + 错误记录 | **高** | | `agent/graph.py` | 状态图编译 + 路由函数（预览跳过验证） | 中 | @@ -80,7 +80,7 @@ agent/graph.py (LangGraph 状态机) | `backend/logger.py` | 集中日志模块：JSON 格式化 + trace_id + 独立 llm.log | 低 | | `backend/rag_adapter.py` | RAGSearcher 单例，语义搜索接口 | 中 | | `backend/error_kb.py` | ErrorKB — 错误指纹去重 + ChromaDB 持久化 + 语义检索 | 中 | -| `backend/file_parser.py` | 文件解析: PDF/DOCX/图片(EasyOCR→PaddleOCR回退)/文本 | 中 | +| `backend/file_parser.py` | 文件解析: PDF/DOCX/XLSX/图片(EasyOCR→PaddleOCR回退)/文本 | 中 | | `backend/layout_analyzer.py` | A4模板分析: 比例检测/EasyOCR→PaddleOCR元素提取/行分组/JRXML行匹配 | 中 | | `backend/ocr_extractor.py` | OCR单据字段提取: 两阶段流水线 + 4种策略(精确KV/模糊KV/正则/表格) + 17个默认中文字段 | 中 | | `backend/embeddings.py` | 嵌入模型工厂 (HuggingFace/OpenAI) | 低 | @@ -132,11 +132,20 @@ agent/graph.py (LangGraph 状态机) - 流程：correct_jrxml 保存 last_error_case → validate 通过时自动入库 ### 文件上传 -- 侧边栏多文件上传（可逐文件移除） -- 支持: PDF(pdfplumber+PIL) / DOCX(python-docx) / 图片(PIL+PaddleOCR可选) / 纯文本 +- **对话区域上传（v3）**: `st.file_uploader` 位于聊天输入框上方，支持图片/PDF/DOCX/XLSX/文本 +- **粘贴/拖拽（v3）**: 全局 paste/drop 事件监听 + `sessionStorage` + 轮询桥接组件，Ctrl+V 粘贴或拖拽文件到页面任意位置 +- **文件预览芯片（v3）**: 上传后显示在对话区域，可逐文件移除（自动清理临时文件） +- 侧边栏多文件上传（可逐文件移除，向后兼容保留） +- 支持: PDF(pdfplumber+PIL) / DOCX(python-docx) / XLSX(openpyxl, v3) / 图片(PIL+EasyOCR优先→PaddleOCR回退) / 纯文本 - 上传文本自动注入下一条消息前缀 - 根据 `can_use_vision()` 判断是否走原生多模态（当前 MiniMax 不支持） +### 对话区域文件粘贴/拖拽技术方案（v3） +- `st.html()` 注入全局 paste/drop/dragover 监听器 → 文件转 base64 → 写入 `sessionStorage` +- `components.html(height=0)` 桥接组件每 800ms 轮询 `sessionStorage` → `Streamlit.setComponentValue` 回传 Python +- Python 解码 base64 → 临时文件 → `parse_file` + `analyze_layout` 双层 OCR 解析 +- 上限：单文件 20MB，单次最多 10 个文件 + ### A4 模板识别 - `backend/layout_analyzer.py` — 三种处理路径： - **完整 A4**: 比例匹配 + OCR 元素 → 全量布局描述 @@ -173,6 +182,9 @@ agent/graph.py (LangGraph 状态机) - **OCR 引擎**: 优先使用 EasyOCR（Windows 兼容性更好，`pip install easyocr`），回退 PaddleOCR。两者均未安装时仅返回图片元信息，建议至少安装 EasyOCR。 - **OCR 字段提取**: `process_input` 自动检测上传图片，调用 `OcrExtractor` 提取 17 个常见中文字段（发票代码/号码/金额/日期等），提取结果自动注入 LLM 上下文。 - **会话持久化**: `session_id` 现已包含在 `save_session_node` 的持久化字段中，避免切换会话时因 `session_id` 丢失导致的无限 rerun bug。 +- **v3 修复**: `create_session` 现在存盘前强制写入 `agent_state["session_id"] = sid`。`load_session_node` 不再从磁盘覆盖 `session_id`。切换会话增加 `_last_switched_to` 哨兵防止重复触发。 - **MAX_RETRY**: 默认 3 次。重试耗尽后 `pending_failure_context` 记录失败信息，下次用户输入时自动注入。 - **验证最小内容检查**: 验证服务额外检查至少 1 个 `` + 1 个 `` 或 ``，拦截空壳 JRXML。 +- **XLSX 支持 (v3)**: 需要 `openpyxl>=3.1.0`（已加入 requirements.txt）。表格按工作表逐行读取，单元格用 `|` 分隔。 +- **粘贴功能限制**: 文件以 base64 编码在 sessionStorage 中传递，单文件上限 20MB。大文件建议使用 file_uploader 按钮。 - **torchvision**: `transformers` 库的懒加载需要 `torchvision`，已作为依赖安装。 diff --git a/agent/nodes.py b/agent/nodes.py index 3160d69..efae894 100644 --- a/agent/nodes.py +++ b/agent/nodes.py @@ -376,11 +376,11 @@ def load_session_node(state: AgentState) -> Dict: data = load_session(session_id) if data and data.get("agent_state"): saved = data["agent_state"] - # 恢复核心字段（不覆盖当前请求的 user_input / stage） - for key in ("session_id", "conversation_history", "full_conversation_history", + # 恢复核心字段（不覆盖当前请求的 user_input / stage / session_id） + for key in ("conversation_history", "full_conversation_history", "current_jrxml", "final_jrxml", "compressed_history", "session_name", "created_at", "history_states"): - if key in saved and key not in ("user_input", "stage"): + if key in saved and key not in ("user_input", "stage", "session_id"): state[key] = saved[key] state["session_name"] = data.get("session_name", "") state["created_at"] = data.get("created_at", "") diff --git a/app.py b/app.py index 64cb7ef..68160fc 100644 --- a/app.py +++ b/app.py @@ -21,6 +21,7 @@ import time from pathlib import Path import streamlit as st +import streamlit.components.v1 as components from dotenv import load_dotenv load_dotenv() @@ -117,6 +118,10 @@ if "graph" not in st.session_state: st.session_state.graph = build_graph() if "pending_action" not in st.session_state: st.session_state.pending_action = None +if "chat_attached_files" not in st.session_state: + st.session_state.chat_attached_files = [] # [{name, text, type, path}] +if "_paste_processed_ts" not in st.session_state: + st.session_state._paste_processed_ts = 0 if "agent_state" not in st.session_state: if url_session_id: @@ -402,16 +407,21 @@ with st.sidebar: if selected and session_options.get(selected) != current_session_id: new_sid = session_options[selected] - data = load_session(new_sid) - if data and data.get("agent_state"): - _app_log.info( - "切换会话", - extra={"from_session": current_session_id, "to_session": new_sid}, - ) - data["agent_state"]["session_id"] = new_sid - st.session_state.agent_state = data["agent_state"] - st.session_state.messages = [] - st.rerun() + if st.session_state.get("_last_switched_to") == new_sid: + # 防止同一会话重复切换导致的无限 rerun 循环 + st.session_state._last_switched_to = "" + else: + data = load_session(new_sid) + if data and data.get("agent_state"): + _app_log.info( + "切换会话", + extra={"from_session": current_session_id, "to_session": new_sid}, + ) + data["agent_state"]["session_id"] = new_sid + st.session_state.agent_state = data["agent_state"] + st.session_state.messages = [] + st.session_state._last_switched_to = new_sid + st.rerun() col1, col2 = st.columns(2) with col1: @@ -481,7 +491,7 @@ with st.sidebar: uploaded = st.file_uploader( "选择文件", - type=["png", "jpg", "jpeg", "bmp", "webp", "pdf", "docx", "txt", "csv", "json", "xml"], + type=["png", "jpg", "jpeg", "bmp", "webp", "pdf", "docx", "xlsx", "txt", "csv", "json", "xml"], accept_multiple_files=True, key="file_uploader", label_visibility="collapsed", @@ -614,6 +624,95 @@ with st.sidebar: key=f"dl_v{i}", ) +# ---- 文件粘贴/拖拽全局处理器 ---- +st.html(""" + +""") + +# ---- 粘贴桥接组件 ---- +paste_data = components.html(""" + +""", height=0, default=0) + +if paste_data and paste_data != 0: + pts = paste_data.get("ts", 0) + if pts > st.session_state._paste_processed_ts: + st.session_state._paste_processed_ts = pts + import base64, tempfile + from backend.file_parser import parse_file + from backend.layout_analyzer import analyze_layout + for fi in paste_data.get("files", []): + if not any(f["name"] == fi["name"] for f in st.session_state.chat_attached_files): + header, b64 = fi["data"].split(",", 1) + raw = base64.b64decode(b64) + suffix = Path(fi["name"]).suffix.lower() + with tempfile.NamedTemporaryFile(suffix=suffix, delete=False) as tmp: + tmp.write(raw) + tmp_path = tmp.name + result = parse_file(tmp_path, suffix) + text = result["text"] + file_type = result["file_type"] + img_suffixes = (".png", ".jpg", ".jpeg", ".bmp", ".webp") + if suffix in img_suffixes and result.get("method") not in ("metadata_only", None): + try: + layout = analyze_layout(tmp_path) + tt = layout.get("template_type", "unknown") + if tt == "full_a4": + text = layout["description"] + file_type = "a4_template" + elif tt == "partial_rows": + file_type = "a4_partial" + except Exception: + pass + st.session_state.chat_attached_files.append({ + "name": fi["name"], "text": text, "type": file_type, "path": tmp_path + }) + st.rerun() + # ---- 标题 ---- st.title("📝 JRXML 报表生成器") st.caption("用自然语言描述您的报表需求，我将逐步生成可用的 JRXML 模板。") @@ -633,29 +732,122 @@ for msg in st.session_state.messages: else: st.markdown(msg["content"]) +# ---- 已附加文件预览 ---- +if st.session_state.chat_attached_files: + n_files = len(st.session_state.chat_attached_files) + chip_cols = st.columns(min(n_files, 4)) + files_to_remove = [] + for i, f in enumerate(st.session_state.chat_attached_files): + with chip_cols[i % len(chip_cols)]: + c1, c2 = st.columns([5, 1]) + with c1: + name = f["name"] + short_name = name[:16] + ("…" if len(name) > 16 else "") + emoji_map = {"a4_template": "📷", "image": "🖼", "pdf": "📄", "docx": "📝", "xlsx": "📊"} + emoji = emoji_map.get(f["type"], "📎") + st.caption(f"{emoji} {short_name}") + with c2: + if st.button("✕", key=f"rm_chip_{i}"): + files_to_remove.append(i) + if files_to_remove: + for i in sorted(files_to_remove, reverse=True): + try: + Path(st.session_state.chat_attached_files[i]["path"]).unlink(missing_ok=True) + except Exception: + pass + st.session_state.chat_attached_files.pop(i) + st.rerun() + +# ---- 对话区域文件上传 ---- +col_fu, col_hint = st.columns([5, 1]) +with col_fu: + chat_uploads = st.file_uploader( + "附加文件", + type=["png", "jpg", "jpeg", "bmp", "webp", "pdf", "docx", "xlsx", "txt", "csv", "json", "xml"], + accept_multiple_files=True, + key="chat_file_uploader", + label_visibility="visible", + ) +with col_hint: + st.caption("Ctrl+V 粘贴\n或拖拽到页面") + +if chat_uploads: + newly_added = False + import tempfile + from backend.file_parser import parse_file + from backend.layout_analyzer import analyze_layout + for uf in chat_uploads: + if not any(f["name"] == uf.name for f in st.session_state.chat_attached_files): + suffix = Path(uf.name).suffix.lower() + with tempfile.NamedTemporaryFile(suffix=suffix, delete=False) as tmp: + tmp.write(uf.getvalue()) + tmp_path = tmp.name + result = parse_file(tmp_path, suffix) + text = result["text"] + file_type = result["file_type"] + img_suffixes = (".png", ".jpg", ".jpeg", ".bmp", ".webp") + if suffix in img_suffixes and result.get("method") not in ("metadata_only", None): + try: + layout = analyze_layout(tmp_path) + tt = layout.get("template_type", "unknown") + if tt == "full_a4": + text = layout["description"] + file_type = "a4_template" + elif tt == "partial_rows": + file_type = "a4_partial" + except Exception: + pass + st.session_state.chat_attached_files.append({ + "name": uf.name, "text": text, "type": file_type, "path": tmp_path + }) + newly_added = True + if newly_added: + st.session_state.chat_file_uploader = [] + st.rerun() + # ---- 聊天输入 ---- if prompt := st.chat_input("描述您的报表需求..."): - # 拼接上传文件的文本 - uploaded_texts = [] - uploaded_files_info = [] + # 拼接对话区域附加文件的文本 + file_texts = [] + attached_info = [] + for f in st.session_state.chat_attached_files: + file_texts.append(f"[附加文件: {f['name']} ({f['type']})]\n{f['text']}") + attached_info.append({"name": f["name"], "type": f["type"], "length": len(f["text"])}) + + # 同时拼接侧边栏上传的文件（向后兼容） if st.session_state.get("uploaded_files"): for f in st.session_state.uploaded_files: - uploaded_texts.append(f"[上传文件: {f['name']}]\n{f['text']}") - uploaded_files_info.append({"name": f["name"], "type": f["type"], "length": len(f["text"])}) - if uploaded_texts: - full_prompt = "\n\n".join(uploaded_texts) + "\n\n---\n用户需求:\n" + prompt - st.session_state.uploaded_files = [] # 用后即清 + file_texts.append(f"[上传文件: {f['name']}]\n{f['text']}") + attached_info.append({"name": f["name"], "type": f["type"], "length": len(f["text"])}) + + if file_texts: + full_prompt = "\n\n".join(file_texts) + "\n\n---\n用户需求:\n" + prompt else: full_prompt = prompt + # 将第一个图片文件的路径传给 agent，供 OCR 字段精确提取 + for f in st.session_state.chat_attached_files: + if f["type"] in ("image", "a4_template", "a4_partial"): + st.session_state.agent_state["uploaded_file_path"] = f["path"] + break + + # 清理临时文件和状态 + st.session_state.uploaded_files = [] + for f in st.session_state.chat_attached_files: + try: + Path(f["path"]).unlink(missing_ok=True) + except Exception: + pass + st.session_state.chat_attached_files = [] + _app_log.info( "收到用户输入", extra={ "session_id": current_session_id, "prompt_preview": prompt[:200], "prompt_length": len(prompt), - "has_uploaded_files": bool(uploaded_files_info), - "uploaded_files": uploaded_files_info, + "has_uploaded_files": bool(attached_info), + "uploaded_files": attached_info, }, ) diff --git a/backend/file_parser.py b/backend/file_parser.py index 357d085..685e080 100644 --- a/backend/file_parser.py +++ b/backend/file_parser.py @@ -51,6 +51,7 @@ def parse_file(file_path: str, file_type: str = "") -> dict: ".webp": _parse_image, ".pdf": _parse_pdf, ".docx": _parse_docx, + ".xlsx": _parse_xlsx, } parser = parsers.get(suffix) @@ -195,6 +196,37 @@ def _parse_docx(path: Path) -> dict: "error": "DOCX 解析需要安装 python-docx"} +def _parse_xlsx(path: Path) -> dict: + """提取 Excel (.xlsx) 表格内容为文本。""" + try: + import openpyxl + wb = openpyxl.load_workbook(path, read_only=True, data_only=True) + sheets_text = [] + for sheet_name in wb.sheetnames: + ws = wb[sheet_name] + rows = [] + for row in ws.iter_rows(values_only=True): + cells = [str(c) if c is not None else "" for c in row] + if any(c.strip() for c in cells): + rows.append(" | ".join(cells)) + if rows: + sheets_text.append(f"--- 工作表: {sheet_name} ---\n" + "\n".join(rows)) + wb.close() + if sheets_text: + return { + "text": "\n\n".join(sheets_text), + "file_type": "xlsx", + "method": "openpyxl", + "error": None, + } + except ImportError: + pass + except Exception: + pass + return {"text": "", "file_type": "xlsx", "method": "none", + "error": "XLSX 解析需要安装 openpyxl"} + + def _parse_text(path: Path) -> dict: """读取纯文本文件。""" try: diff --git a/backend/session.py b/backend/session.py index d541815..15f8748 100644 --- a/backend/session.py +++ b/backend/session.py @@ -38,12 +38,14 @@ def create_session(name: str = "", agent_state: Optional[dict] = None) -> dict: _ensure_dir() sid = generate_session_id() now = _now_iso() + agent_state = agent_state or {} + agent_state["session_id"] = sid data = { "session_id": sid, "session_name": name or f"新建报表 {now[:10]}", "created_at": now, "updated_at": now, - "agent_state": agent_state or {}, + "agent_state": agent_state, } with open(_session_path(sid), "w", encoding="utf-8") as f: json.dump(data, f, ensure_ascii=False, indent=2) diff --git a/requirements.txt b/requirements.txt index 8ef556f..75e782b 100644 --- a/requirements.txt +++ b/requirements.txt @@ -25,6 +25,7 @@ tqdm>=4.65.0 python-dotenv>=1.0.0 httpx>=0.27.0 tiktoken>=0.7.0 +openpyxl>=3.1.0 # 测试 pytest>=8.0.0