feat: 对话区域文件上传(粘贴/拖拽) + XLSX支持 + 会话切换无限循环修复

- 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 <noreply@anthropic.com>

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 `<band>` + 1 个 `<textField>` 或 `<staticText>`，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。

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", "")

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("""
+<script>
+(function() {
+    if (window.__jrxml_drop_paste) return;
+    window.__jrxml_drop_paste = true;
+    var MAX_SIZE = 20 * 1024 * 1024;
+    function handleFiles(files) {
+        var fd = []; var n = 0; var total = Math.min(files.length, 10);
+        for (var i = 0; i < total; i++) {
+            var f = files[i];
+            if (f.size > MAX_SIZE) { n++; continue; }
+            var reader = new FileReader();
+            reader.onload = (function(file) {
+                return function(e) {
+                    fd.push({name: file.name, size: file.size, data: e.target.result});
+                    n++;
+                    if (n === total && fd.length) {
+                        sessionStorage.setItem('_jrxml_paste', JSON.stringify({ts: Date.now(), files: fd}));
+                    }
+                };
+            })(f);
+            reader.readAsDataURL(f);
+        }
+    }
+    document.addEventListener('paste', function(e) {
+        var fs = e.clipboardData && e.clipboardData.files;
+        if (fs && fs.length) { e.preventDefault(); handleFiles(fs); }
+    });
+    document.addEventListener('dragover', function(e) {
+        e.preventDefault(); e.dataTransfer.dropEffect = 'copy';
+    });
+    document.addEventListener('drop', function(e) {
+        var fs = e.dataTransfer && e.dataTransfer.files;
+        if (fs && fs.length) { e.preventDefault(); handleFiles(fs); }
+    });
+})();
+</script>
+""")
+
+# ---- 粘贴桥接组件 ----
+paste_data = components.html("""
+<script>
+(function poll() {
+    var raw = sessionStorage.getItem('_jrxml_paste');
+    if (raw) {
+        try { sessionStorage.removeItem('_jrxml_paste'); Streamlit.setComponentValue(JSON.parse(raw)); return; }
+        catch(e) {}
+    }
+    setTimeout(poll, 800);
+})();
+</script>
+""", 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,
         },
     )
 

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:

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)

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