Replace PaddleOCR SDK with HTTP async API due to pyyaml conflict

tests/tools/paddleocr/test_file_input.py

@@ -1,10 +1,38 @@
 import base64
 import os
 import sys
+from unittest.mock import MagicMock
 
 import pytest
 import yaml
 
+# Mock paddleocr module before any imports
+mock_paddleocr = MagicMock()
+
+# Mock public API classes
+mock_paddleocr.PaddleOCRClient = MagicMock
+mock_paddleocr.OCROptions = lambda **kw: MagicMock()
+mock_paddleocr.PPStructureV3Options = lambda **kw: MagicMock()
+mock_paddleocr.PaddleOCRVLOptions = lambda **kw: MagicMock()
+mock_paddleocr.AuthError = Exception
+mock_paddleocr.PaddleOCRAPIError = Exception
+
+# Mock internal modules for backward compatibility
+mock_paddleocr._api_client = MagicMock()
+mock_paddleocr._api_client.PaddleOCRClient = mock_paddleocr.PaddleOCRClient
+mock_paddleocr._api_client.models = MagicMock()
+mock_paddleocr._api_client.models.OCROptions = mock_paddleocr.OCROptions
+mock_paddleocr._api_client.models.PPStructureV3Options = mock_paddleocr.PPStructureV3Options
+mock_paddleocr._api_client.models.PaddleOCRVLOptions = mock_paddleocr.PaddleOCRVLOptions
+mock_paddleocr._api_client.errors = MagicMock()
+mock_paddleocr._api_client.errors.AuthError = mock_paddleocr.AuthError
+mock_paddleocr._api_client.errors.PaddleOCRAPIError = mock_paddleocr.PaddleOCRAPIError
+
+sys.modules["paddleocr"] = mock_paddleocr
+sys.modules["paddleocr._api_client"] = mock_paddleocr._api_client
+sys.modules["paddleocr._api_client.models"] = mock_paddleocr._api_client.models
+sys.modules["paddleocr._api_client.errors"] = mock_paddleocr._api_client.errors
+
 REPO_ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), "../../.."))
 PLUGIN_DIR = os.path.join(REPO_ROOT, "tools", "paddleocr")
 if PLUGIN_DIR not in sys.path:
@@ -45,21 +73,29 @@ def test_file_upload_is_base64_encoded():
         file_type=FileType.IMAGE,
     )
 
-    payload, normalized_file_type = normalize_file_input(file, "auto")
+    input_value, is_temp_file, file_type_code = normalize_file_input(file, "auto")
 
-    assert payload == base64.b64encode(b"image-bytes").decode("utf-8")
-    assert normalized_file_type == 1
+    # New implementation saves to temp file for SDK
+    assert os.path.exists(input_value)
+    assert is_temp_file is True
+    assert file_type_code == 1
+    # Clean up
+    os.unlink(input_value)
 
 
 def test_pdf_file_upload_infers_file_type():
     file = make_file(
         b"%PDF-1.7", filename="invoice.pdf", mime_type="application/pdf", extension=".pdf"
     )
 
-    payload, normalized_file_type = normalize_file_input(file, "auto")
+    input_value, is_temp_file, file_type_code = normalize_file_input(file, "auto")
 
-    assert payload == base64.b64encode(b"%PDF-1.7").decode("utf-8")
-    assert normalized_file_type == 0
+    # New implementation saves to temp file for SDK
+    assert os.path.exists(input_value)
+    assert is_temp_file is True
+    assert file_type_code == 0
+    # Clean up
+    os.unlink(input_value)
 
 
 def test_image_file_upload_infers_file_type_from_filename_when_mime_type_missing():
@@ -71,10 +107,14 @@ def test_image_file_upload_infers_file_type_from_filename_when_mime_type_missing
         file_type=FileType.IMAGE,
     )
 
-    payload, normalized_file_type = normalize_file_input(file, None)
+    input_value, is_temp_file, file_type_code = normalize_file_input(file, None)
 
-    assert payload == base64.b64encode(b"image-bytes").decode("utf-8")
-    assert normalized_file_type == 1
+    # New implementation saves to temp file for SDK
+    assert os.path.exists(input_value)
+    assert is_temp_file is True
+    assert file_type_code == 1
+    # Clean up
+    os.unlink(input_value)
 
 
 def test_explicit_file_type_overrides_inference():
@@ -86,17 +126,22 @@ def test_explicit_file_type_overrides_inference():
         file_type=FileType.IMAGE,
     )
 
-    payload, normalized_file_type = normalize_file_input(file, "pdf")
+    input_value, is_temp_file, file_type_code = normalize_file_input(file, "pdf")
 
-    assert payload == base64.b64encode(b"image-bytes").decode("utf-8")
-    assert normalized_file_type == 0
+    # New implementation saves to temp file for SDK
+    assert os.path.exists(input_value)
+    assert is_temp_file is True
+    assert file_type_code == 0
+    # Clean up
+    os.unlink(input_value)
 
 
 def test_legacy_file_string_is_passed_through():
-    payload, normalized_file_type = normalize_file_input("https://example.com/scan.pdf", "auto")
+    input_value, is_temp_file, file_type_code = normalize_file_input("https://example.com/scan.pdf", "auto")
 
-    assert payload == "https://example.com/scan.pdf"
-    assert normalized_file_type is None
+    assert input_value == "https://example.com/scan.pdf"
+    assert is_temp_file is False
+    assert file_type_code is None
 
 
 def test_missing_file_input_raises_clear_error():
@@ -106,21 +151,44 @@ def test_missing_file_input_raises_clear_error():
 
 def invoke_tool_with_mocked_api(monkeypatch, tool_cls, credentials, parameters):
     captured = {}
-    module_name = tool_cls.__module__.split(".")[-1]
-
-    def fake_api_request(api_url, params, access_token):
-        captured["api_url"] = api_url
-        captured["params"] = params
-        captured["access_token"] = access_token
-        return {
-            "errorCode": 0,
-            "result": {
-                "ocrResults": [{"prunedResult": {"rec_texts": ["hello", "world"]}}],
-                "layoutParsingResults": [{"markdown": {"text": "# Parsed", "images": {}}}],
-            },
-        }
-
-    monkeypatch.setattr(f"tools.{module_name}.make_paddleocr_api_request", fake_api_request)
+
+    def fake_sdk_call(**kwargs):
+        captured["kwargs"] = kwargs
+        # Return mock result - use simple dict instead of SDK classes
+        if tool_cls == TextRecognitionTool:
+            return type("OCRResult", (), {"job_id": "test-job", "pages": [
+                type("OCRPage", (), {"pruned_result": {"rec_texts": ["hello", "world"]}, "ocr_image_url": None})()
+            ]})()
+        else:
+            return type("DocParsingResult", (), {"job_id": "test-job", "pages": [
+                type("DocParsingPage", (), {"markdown_text": "# Parsed", "markdown_images": {}, "output_images": {}})()
+            ]})()
+
+    # Mock the entire SDK module and client
+    fake_client = MagicMock()
+    fake_client.ocr = fake_sdk_call
+    fake_client.parse_document = fake_sdk_call
+
+    # Mock utils module functions
+    import tools.utils as utils_module
+    monkeypatch.setattr(utils_module, "get_sdk_client", lambda *args: fake_client)
+    monkeypatch.setattr(utils_module, "base64_to_temp_file", lambda *args: "temp_file.png")
+    monkeypatch.setattr(utils_module, "cleanup_temp_file", lambda *args: None)
+
+    # Mock in the specific tool module (they import these directly from utils)
+    if tool_cls == TextRecognitionTool:
+        import tools.text_recognition as tr_module
+        monkeypatch.setattr(tr_module, "get_sdk_client", lambda *args: fake_client)
+        monkeypatch.setattr(tr_module, "cleanup_temp_file", lambda *args: None)
+    elif tool_cls == DocumentParsingTool:
+        import tools.document_parsing as dp_module
+        monkeypatch.setattr(dp_module, "get_sdk_client", lambda *args: fake_client)
+        monkeypatch.setattr(dp_module, "cleanup_temp_file", lambda *args: None)
+    else:
+        import tools.document_parsing_vl as dpv_module
+        monkeypatch.setattr(dpv_module, "get_sdk_client", lambda *args: fake_client)
+        monkeypatch.setattr(dpv_module, "cleanup_temp_file", lambda *args: None)
+
     tool = tool_cls.from_credentials(credentials)
     list(tool._invoke(parameters))
     return captured
@@ -145,11 +213,11 @@ def test_text_recognition_sends_normalized_file_to_api(monkeypatch):
         {"file": file, "fileType": "auto", "visualize": False},
     )
 
-    assert captured["api_url"] == "https://example.com/text-recognition"
-    assert captured["access_token"] == "token"
-    assert captured["params"]["file"] == base64.b64encode(b"image-bytes").decode("utf-8")
-    assert captured["params"]["fileType"] == 1
-    assert captured["params"]["visualize"] is False
+    # SDK receives file_path (temp file), not base64 directly
+    assert "file_path" in captured["kwargs"]
+    assert captured["kwargs"]["file_path"] == "temp_file.png"
+    assert captured["kwargs"]["options"] is not None
+    assert hasattr(captured["kwargs"]["options"], "visualize")
 
 
 def test_document_parsing_sends_normalized_file_to_api(monkeypatch):
@@ -167,10 +235,9 @@ def test_document_parsing_sends_normalized_file_to_api(monkeypatch):
         {"file": file, "fileType": "auto", "markdownIgnoreLabels": "header, footer"},
     )
 
-    assert captured["api_url"] == "https://example.com/document-parsing"
-    assert captured["params"]["file"] == base64.b64encode(b"%PDF-1.7").decode("utf-8")
-    assert captured["params"]["fileType"] == 0
-    assert captured["params"]["markdownIgnoreLabels"] == ["header", "footer"]
+    assert "file_path" in captured["kwargs"]
+    assert captured["kwargs"]["file_path"] == "temp_file.png"
+    assert captured["kwargs"]["options"] is not None
 
 
 def test_document_parsing_vl_sends_normalized_file_to_api(monkeypatch):
@@ -192,10 +259,8 @@ def test_document_parsing_vl_sends_normalized_file_to_api(monkeypatch):
         {"file": file, "fileType": "auto", "promptLabel": "undefined"},
     )
 
-    assert captured["api_url"] == "https://example.com/document-parsing-vl"
-    assert captured["params"]["file"] == base64.b64encode(b"image-bytes").decode("utf-8")
-    assert captured["params"]["fileType"] == 1
-    assert "promptLabel" not in captured["params"]
+    assert "file_path" in captured["kwargs"]
+    assert captured["kwargs"]["file_path"] == "temp_file.png"
 
 
 def load_tool_yaml(tool_name: str) -> dict:

tools/paddleocr/manifest.yaml

@@ -1,4 +1,4 @@
-version: 0.2.6
+version: 0.2.7
 type: plugin
 author: langgenius
 name: paddleocr

tools/paddleocr/provider/paddleocr.py

@@ -6,6 +6,7 @@
 from tools.document_parsing import DocumentParsingTool
 from tools.document_parsing_vl import DocumentParsingVlTool
 from tools.text_recognition import TextRecognitionTool
+from tools.utils import call_paddleocr_api, get_sdk_client
 
 
 class PaddleocrProvider(ToolProvider):
@@ -15,36 +16,26 @@ def _validate_credentials(self, credentials: dict[str, Any]) -> None:
                 "AI Studio access token must be provided"
             )
 
-        api_url_keys = (
-            "text_recognition_api_url",
-            "document_parsing_api_url",
-            "document_parsing_vl_api_url",
-        )
-        tool_classes = (
-            TextRecognitionTool,
-            DocumentParsingTool,
-            DocumentParsingVlTool,
-        )
+        # Get base_url (optional, uses SDK default if not provided)
+        base_url = credentials.get("base_url")
+
+        # Test with OCR (works for all models)
         test_file = "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png"
 
-        if not any(key in credentials for key in api_url_keys):
-            raise ToolProviderCredentialValidationError(
-                "You should provide at least one API URL"
+        try:
+            client_config = get_sdk_client(
+                access_token=credentials["aistudio_access_token"],
+                base_url=base_url,
             )
-
-        for api_url_key, tool_cls in zip(api_url_keys, tool_classes):
-            if api_url_key in credentials:
-                try:
-                    self._test_tool_validation(tool_cls, credentials, test_file)
-                except Exception as e:
-                    raise ToolProviderCredentialValidationError(
-                        f"Invalid credentials for {tool_cls.__name__}"
-                    ) from e
-
-    def _test_tool_validation(
-        self, tool_cls, credentials: dict[str, Any], test_file: str
-    ) -> None:
-        tool = tool_cls.from_credentials(credentials)
-
-        for _ in tool.invoke(tool_parameters={"file": test_file}):
-            break
+            call_paddleocr_api(
+                model="PP-OCRv5",
+                file_url=test_file,
+                file_path=None,
+                options={},
+                client_config=client_config,
+                is_document_parsing=False,
+            )
+        except Exception as e:
+            raise ToolProviderCredentialValidationError(
+                f"Validation failed: {e}"
+            ) from e

tools/paddleocr/provider/paddleocr.yaml

@@ -34,39 +34,15 @@ credentials_for_provider:
             en_US: Get your AI Studio access token
             zh_Hans: 获取星河社区访问令牌
         url: https://aistudio.baidu.com/index/accessToken
-    text_recognition_api_url:
+    base_url:
         type: text-input
+        required: false
         label:
-            en_US: Text Recognition API URL
-            zh_Hans: 文字识别 API URL
+            en_US: Base URL (Optional)
+            zh_Hans: Base URL（可选）
         placeholder:
-            en_US: Text Recognition API URL
-            zh_Hans: 文字识别 API URL
+            en_US: https://paddleocr.aistudio-app.com
+            zh_Hans: https://paddleocr.aistudio-app.com
         help:
-            en_US: Click the "API" button in the upper-left corner, select "Text recognition(PP-OCRv5)", and copy the `API_URL`.
-            zh_Hans: 点击左上角的“API”，选择“文字识别（PP-OCRv5）”并复制 `API_URL`
-        url: https://aistudio.baidu.com/paddleocr/task
-    document_parsing_api_url:
-        type: text-input
-        label:
-            en_US: Document Parsing API URL
-            zh_Hans: 文档解析 API URL
-        placeholder:
-            en_US: Document Parsing API URL
-            zh_Hans: 文档解析 API URL
-        help:
-            en_US: Click the "API" button in the upper-left corner, select "Document parsing(PP-StructureV3)", and copy the `API_URL`.
-            zh_Hans: 点击左上角的“API”，选择“文档解析（PP-StructureV3）”并复制 `API_URL`
-        url: https://aistudio.baidu.com/paddleocr/task
-    document_parsing_vl_api_url:
-        type: text-input
-        label:
-            en_US: Large Model Document Parsing API URL
-            zh_Hans: 大模型文档解析 API URL
-        placeholder:
-            en_US: Large Model Document Parsing API URL
-            zh_Hans: 大模型文档解析 API URL
-        help:
-            en_US: Click the "API" button in the upper-left corner, select "Large Model document parsing(PaddleOCR-VL)", and copy the `API_URL`.
-            zh_Hans: 点击左上角的“API”，选择“大模型文档解析（PaddleOCR-VL）”并复制 `API_URL`
-        url: https://aistudio.baidu.com/paddleocr/task
+            en_US: Leave empty to use the default PaddleOCR service. Only needed for self-hosted deployments.
+            zh_Hans: 留空则使用默认的 PaddleOCR 服务。仅自建服务时需要填写。

tools/paddleocr/pyproject.toml

@@ -1,15 +1,16 @@
 [project]
-name = "paddleocr"
+name = "paddleocr-dify"
 version = "0.1.0"
 description = "Add your description here"
 readme = "README.md"
 requires-python = ">=3.12"
 
+# Managed with uv; refresh the lockfile with `uv lock`.
 dependencies = [
     "dify_plugin>=0.9.0",
     "requests>=2.34.2",
 ]
 
 # uv run black . -C -l 100 && uv run ruff check --fix
 [dependency-groups]
-dev = []
+dev = []