refactor: 移除 PaddleOCR 命令行调用，只使用 Python API

- 启用 Python API（PADDLEOCR_API_AVAILABLE = True）
- 移除所有命令行回退逻辑
- 简化错误处理，直接返回失败
- 保留超时保护机制（60秒）

pdf_converter_v2/utils/paddleocr_fallback.py

@@ -49,13 +49,11 @@ except ImportError:
 
 try:
     from paddleocr import PaddleOCRVL
-    # 暂时禁用 Python API，因为在 VLLM 引擎运行时显存不足（需要约 11GB）
-    # 等 VLLM GPU 内存利用率降低或使用独立 GPU 后再启用
-    PADDLEOCR_API_AVAILABLE = False
-    logger.info("[PaddleOCR备用] Python API 已禁用（显存不足），使用命令行方式")
+    PADDLEOCR_API_AVAILABLE = True
+    logger.info("[PaddleOCR备用] Python API 已加载")
 except ImportError:
     PADDLEOCR_API_AVAILABLE = False
-    logger.warning("[PaddleOCR备用] paddleocr Python API 未安装，将使用命令行方式")
+    logger.warning("[PaddleOCR备用] paddleocr Python API 未安装")
 
 # PaddleOCR VL pipeline 单例（避免重复初始化）
 _PADDLEOCR_PIPELINE_CACHE: Dict[tuple, Any] = {}
@@ -956,104 +954,37 @@ def call_paddleocr_ocr(image_path: str, save_path: str) -> tuple[Optional[List[s
         save_path_base = os.path.join(save_path, image_basename)
         os.makedirs(save_path_base, exist_ok=True)
 
-        # 优先使用 Python API 方式（use_chart_recognition=False, use_layout_detection=False）
-        if PADDLEOCR_API_AVAILABLE:
-            logger.info(f"[PaddleOCR 文本识别] 使用 Python API 方式")
-            success, md_file = _call_paddleocr_api(
-                image_path, 
-                save_path, 
-                use_chart_recognition=False, 
-                use_layout_detection=False
-            )
-            
-            if success and md_file and os.path.exists(md_file):
-                try:
-                    with open(md_file, "r", encoding="utf-8") as f:
-                        md_content = f.read()
-                    if md_content.strip():
-                        texts = markdown_to_plain_text(md_content)
-                        if texts:
-                            json_file = os.path.join(save_path, f"{image_basename}_res.json")
-                            with open(json_file, "w", encoding="utf-8") as f:
-                                json.dump({"rec_texts": texts}, f, ensure_ascii=False, indent=0)
-                            logger.info(f"[PaddleOCR 文本识别] Python API 成功，得到 {len(texts)} 行文本")
-                            return texts, json_file
-                except Exception as e:
-                    logger.exception(f"[PaddleOCR 文本识别] Python API 结果处理失败: {e}")
+        # 使用 Python API 方式（use_chart_recognition=False, use_layout_detection=False）
+        if not PADDLEOCR_API_AVAILABLE:
+            logger.error(f"[PaddleOCR 文本识别] Python API 不可用")
+            return None, None
             
-            logger.warning("[PaddleOCR 文本识别] Python API 失败，回退到命令行方式")
-        
-        # 回退到命令行方式
-        cmd = [
-            _get_paddleocr_executable(), "doc_parser", "-i", image_path,
-            "--precision", "fp32",
-            "--use_doc_unwarping", "False",
-            "--use_doc_orientation_classify", "True",
-            "--use_chart_recognition", "False",
-            "--use_layout_detection", "False",
-            "--save_path", save_path_base
-        ] + _paddle_ocr_device_args()
-        if VL_REC_BACKEND:
-            cmd.extend(["--vl_rec_backend", VL_REC_BACKEND])
-        if VL_REC_SERVER_URL:
-            cmd.extend(["--vl_rec_server_url", VL_REC_SERVER_URL])
-
-        logger.info(f"[PaddleOCR 文本识别] 执行命令(doc_parser): {' '.join(cmd)}")
-
-        result = subprocess.run(
-            cmd,
-            capture_output=True,
-            text=True,
-            timeout=300,
-            check=False,
-            env=_get_paddleocr_subprocess_env(),
+        logger.info(f"[PaddleOCR 文本识别] 使用 Python API 方式")
+        success, md_file = _call_paddleocr_api(
+            image_path, 
+            save_path, 
+            use_chart_recognition=False, 
+            use_layout_detection=False
         )
-
-        if result.returncode != 0:
-            logger.error(f"[PaddleOCR 文本识别] doc_parser 执行失败，返回码: {result.returncode}")
-            if result.stderr and ("too many values to unpack" in result.stderr or "Exception from the 'cv' worker" in result.stderr):
-                logger.warning("[PaddleOCR 文本识别] doc_parser 报 cv worker 解包错误，详见 README_STARTUP.md。")
-            if result.stderr:
-                logger.error(f"[PaddleOCR 文本识别] 错误输出: {result.stderr}")
-            return None, None
-
-        texts = []
-        md_file = os.path.join(save_path_base, f"{image_basename}.md")
-        if os.path.exists(md_file):
+        
+        if success and md_file and os.path.exists(md_file):
             try:
                 with open(md_file, "r", encoding="utf-8") as f:
                     md_content = f.read()
                 if md_content.strip():
                     texts = markdown_to_plain_text(md_content)
+                    if texts:
+                        json_file = os.path.join(save_path, f"{image_basename}_res.json")
+                        with open(json_file, "w", encoding="utf-8") as f:
+                            json.dump({"rec_texts": texts}, f, ensure_ascii=False, indent=0)
+                        logger.info(f"[PaddleOCR 文本识别] Python API 成功，得到 {len(texts)} 行文本")
+                        return texts, json_file
             except Exception as e:
-                logger.exception(f"[PaddleOCR 文本识别] 读取 Markdown 失败: {e}")
-        if not texts and result.stdout.strip():
-            parsed = parse_paddleocr_output(result.stdout.strip())
-            for item in parsed.get("parsing_res_list", []):
-                if isinstance(item, dict) and item.get("block_content"):
-                    block = item["block_content"].strip()
-                    if "\n" in block:
-                        texts.extend([line.strip() for line in block.split("\n") if line.strip()])
-                    else:
-                        texts.append(block)
-        if not texts:
-            logger.warning("[PaddleOCR 文本识别] doc_parser 未得到文本")
-            return None, None
-
-        json_file = os.path.join(save_path, f"{image_basename}_res.json")
-        try:
-            with open(json_file, "w", encoding="utf-8") as f:
-                json.dump({"rec_texts": texts}, f, ensure_ascii=False, indent=0)
-        except Exception as e:
-            logger.exception(f"[PaddleOCR 文本识别] 写入 rec_texts JSON 失败: {e}")
-            return texts, None
-
-        logger.info(f"[PaddleOCR 文本识别] doc_parser 成功提取 {len(texts)} 个文本片段，JSON: {json_file}")
-        return texts, json_file
-
-    except subprocess.TimeoutExpired:
-        logger.error("[PaddleOCR 文本识别] 命令执行超时")
+                logger.exception(f"[PaddleOCR 文本识别] Python API 结果处理失败: {e}")
+        
+        logger.error("[PaddleOCR 文本识别] Python API 失败")
         return None, None
+
     except Exception as e:
         logger.exception(f"[PaddleOCR 文本识别] 调用失败: {e}")
         return None, None
@@ -1078,96 +1009,34 @@ def call_paddleocr_doc_parser_for_text(image_path: str, save_path: str) -> tuple
         save_path_base = os.path.join(save_path, image_basename)
         os.makedirs(save_path_base, exist_ok=True)
         
-        # 优先使用 Python API 方式（use_chart_recognition=True, use_layout_detection=True）
-        if PADDLEOCR_API_AVAILABLE:
-            logger.info(f"[PaddleOCR 图表识别] 使用 Python API 方式")
-            success, md_file = _call_paddleocr_api(
-                image_path, 
-                save_path, 
-                use_chart_recognition=True, 
-                use_layout_detection=True
-            )
-            
-            if success and md_file and os.path.exists(md_file):
-                try:
-                    with open(md_file, 'r', encoding='utf-8') as f:
-                        markdown_content = f.read()
-                    
-                    if markdown_content.strip():
-                        plain_text_lines = markdown_to_plain_text(markdown_content)
-                        logger.info(f"[PaddleOCR 图表识别] Python API 成功，提取 {len(plain_text_lines)} 行纯文本")
-                        return plain_text_lines, md_file
-                except Exception as e:
-                    logger.exception(f"[PaddleOCR 图表识别] Python API 结果处理失败: {e}")
+        # 使用 Python API 方式（use_chart_recognition=True, use_layout_detection=True）
+        if not PADDLEOCR_API_AVAILABLE:
+            logger.error(f"[PaddleOCR 图表识别] Python API 不可用")
+            return None, None
             
-            logger.warning("[PaddleOCR 图表识别] Python API 失败，回退到命令行方式")
-        
-        # 回退到命令行方式
-        cmd = [
-            _get_paddleocr_executable(), "doc_parser", "-i", image_path,
-            "--precision", "fp32",
-            "--use_doc_unwarping", "False",
-            "--use_doc_orientation_classify", "True",
-            "--use_chart_recognition", "True",
-            "--save_path", save_path_base
-        ] + _paddle_ocr_device_args()
-        if VL_REC_BACKEND:
-            cmd.extend(["--vl_rec_backend", VL_REC_BACKEND])
-        if VL_REC_SERVER_URL:
-            cmd.extend(["--vl_rec_server_url", VL_REC_SERVER_URL])
-        
-        logger.info(f"[PaddleOCR 图表识别] 执行命令: {' '.join(cmd)}")
-        
-        # 执行命令（env 含 PADDLE_PDX_DISABLE_MODEL_SOURCE_CHECK、FLAGS_use_stride_kernel）
-        result = subprocess.run(
-            cmd,
-            capture_output=True,
-            text=True,
-            timeout=300,  # 5分钟超时
-            check=False,
-            env=_get_paddleocr_subprocess_env(),
+        logger.info(f"[PaddleOCR 图表识别] 使用 Python API 方式")
+        success, md_file = _call_paddleocr_api(
+            image_path, 
+            save_path, 
+            use_chart_recognition=True, 
+            use_layout_detection=True
         )
         
-        if result.returncode != 0:
-            logger.error(f"[PaddleOCR 图表识别] 命令执行失败，返回码: {result.returncode}")
-            if result.stderr and ("too many values to unpack" in result.stderr or "Exception from the 'cv' worker" in result.stderr):
-                logger.warning(
-                    "[PaddleOCR 图表识别] 报 cv worker 解包错误，多为 PaddleX 与 PP-DocLayoutV3 不兼容。"
-                    " 可尝试: pip install -U paddlex；或改用 文本识别 提取文字。详见 README_STARTUP.md。"
-                )
-            logger.error(f"[PaddleOCR 图表识别] 错误输出: {result.stderr}")
-            return None, None
-
-        md_file = os.path.join(save_path_base, f"{image_basename}.md")
-        if not os.path.exists(md_file):
-            md_files = sorted(Path(save_path_base).rglob("*.md"))
-            if md_files:
-                md_file = str(md_files[0])
-                logger.info(f"[PaddleOCR 图表识别] 在子目录中找到Markdown文件: {md_file}")
-        
-        if not os.path.exists(md_file):
-            logger.warning(f"[PaddleOCR 图表识别] Markdown文件不存在: {md_file}")
-            return None, None
-        
-        try:
-            with open(md_file, 'r', encoding='utf-8') as f:
-                markdown_content = f.read()
-            
-            if not markdown_content.strip():
-                logger.warning("[PaddleOCR 图表识别] Markdown文件内容为空")
-                return [], md_file
-            
-            plain_text_lines = markdown_to_plain_text(markdown_content)
-            logger.info(f"[PaddleOCR 图表识别] 成功提取 {len(plain_text_lines)} 行纯文本，Markdown文件: {md_file}")
-            return plain_text_lines, md_file
+        if success and md_file and os.path.exists(md_file):
+            try:
+                with open(md_file, 'r', encoding='utf-8') as f:
+                    markdown_content = f.read()
                 
-        except Exception as e:
-            logger.exception(f"[PaddleOCR 图表识别] 读取Markdown文件失败: {e}")
-            return None, md_file
-            
-    except subprocess.TimeoutExpired:
-        logger.error("[PaddleOCR 图表识别] 命令执行超时")
+                if markdown_content.strip():
+                    plain_text_lines = markdown_to_plain_text(markdown_content)
+                    logger.info(f"[PaddleOCR 图表识别] Python API 成功，提取 {len(plain_text_lines)} 行纯文本")
+                    return plain_text_lines, md_file
+            except Exception as e:
+                logger.exception(f"[PaddleOCR 图表识别] Python API 结果处理失败: {e}")
+        
+        logger.error("[PaddleOCR 图表识别] Python API 失败")
         return None, None
+            
     except Exception as e:
         logger.exception(f"[PaddleOCR 图表识别] 调用失败: {e}")
         return None, None