fix(galgame): 让 RapidOCR 模型配置实际生效 (#1194)

* 修复 Galgame 插件中 RapidOCR 的模型配置未生效问题。

  本次更改为 RapidOCR 运行时补充模型路径解析逻辑，按用户模型缓存优先、包内 models 目录兜底的顺序查找 det/cls/rec ONNX 模型，并在构造 RapidOCR 实例时传入解析后的模型路径，使 lang_type、ocr_version、
  model_type 配置能够实际影响模型加载。同时将默认 OCR 版本从 PP-OCRv5 调整为与当前内置模型一致的 PP-OCRv4，并更新手动 OCR pipeline 验证脚本以复用统一默认值。

  新增单元测试覆盖模型路径解析、用户缓存优先级、缺失模型回退以及从实际导入包定位 models 目录等场景，防止配置再次失效。

* fix(galgame): honor rapidocr_model_type when resolving ONNX filenames

- _resolve_rapidocr_model_paths 真正使用 model_type，server 变体走 PaddleOCR
  的 `_server_infer` 命名（cls 跨变体共享 mobile）
- _build_runtime_constructor_kwargs 把 model_search_dirs 列表换成具名
  package_models_dir 参数，去掉魔术索引和冗余的 per-path 判空
- 新增 server 变体回归测试，更新现有测试切到新参数名

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(galgame): correct rapidocr server filename infix + harden empty-module-file fallback

- 修正 server 变体文件名：infix 应在 `_det`/`_rec` 之后（PaddleOCR 实际命名是
  `ch_PP-OCRv4_det_server_infer.onnx`，参见 SWHL/RapidOCR HuggingFace repo），
  上一版误把 infix 放到了 lang/version 后面
- module_file 缺失时 package_models_dir 从 Path()（= CWD）改为 None，
  避免 _resolve_rapidocr_model_paths 误扫工作目录
- 同步修正 server 变体回归测试的文件名

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Hongzhi Wen <cartabio.coder1@gmail.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: 3 people

plugin/plugins/galgame_plugin/models.py

@@ -586,7 +586,7 @@ class GalgameRapidOcrConfig:
     rapidocr_engine_type: str = "onnxruntime"
     rapidocr_lang_type: str = "ch"
     rapidocr_model_type: str = "mobile"
-    rapidocr_ocr_version: str = "PP-OCRv5"
+    rapidocr_ocr_version: str = "PP-OCRv4"
 
 
 @dataclass(slots=True, init=False)

plugin/plugins/galgame_plugin/plugin.toml

@@ -118,7 +118,7 @@ install_timeout_seconds = 180
 engine_type = "onnxruntime"
 lang_type = "ch"
 model_type = "mobile"
-ocr_version = "PP-OCRv5"
+ocr_version = "PP-OCRv4"
 
 [plugin.i18n]
 default_locale = "zh-CN"

plugin/plugins/galgame_plugin/rapidocr_support.py

@@ -20,7 +20,7 @@
 DEFAULT_RAPIDOCR_ENGINE_TYPE = "onnxruntime"
 DEFAULT_RAPIDOCR_LANG_TYPE = "ch"
 DEFAULT_RAPIDOCR_MODEL_TYPE = "mobile"
-DEFAULT_RAPIDOCR_OCR_VERSION = "PP-OCRv5"
+DEFAULT_RAPIDOCR_OCR_VERSION = "PP-OCRv4"
 _INSTALL_STATE_NAME = "install_state.json"
 # Leave one core free for the OS / interactive use; floor at 2 so 1-2 core hosts still parallelise.
 _RAPIDOCR_INFERENCE_THREAD_LIMIT = max(2, (os.cpu_count() or 2) - 1)
@@ -99,6 +99,50 @@ def rapidocr_selected_model_name(
     )
 
 
+def _resolve_rapidocr_model_paths(
+    *,
+    model_cache_dir: Path,
+    package_models_dir: Path | None,
+    lang_type: str,
+    ocr_version: str,
+    model_type: str,
+) -> tuple[str | None, str | None, str | None]:
+    lang = str(lang_type or DEFAULT_RAPIDOCR_LANG_TYPE).strip() or DEFAULT_RAPIDOCR_LANG_TYPE
+    version = str(ocr_version or DEFAULT_RAPIDOCR_OCR_VERSION).strip() or DEFAULT_RAPIDOCR_OCR_VERSION
+    # RapidOCR / PaddleOCR file naming (see SWHL/RapidOCR on HuggingFace):
+    #   mobile: ch_PP-OCRv4_det_infer.onnx
+    #   server: ch_PP-OCRv4_det_server_infer.onnx  (i.e. "_server" sits between "_det" and "_infer")
+    # Anything other than "server" falls back to mobile silently — invalid values
+    # would otherwise miss every candidate file and the runtime default mobile model
+    # would still load via RapidOCR's bundled config.yaml.
+    mt = (str(model_type or DEFAULT_RAPIDOCR_MODEL_TYPE).strip() or DEFAULT_RAPIDOCR_MODEL_TYPE).lower()
+    server_infix = "_server" if mt == "server" else ""
+    det_name = f"{lang}_{version}_det{server_infix}_infer.onnx"
+    rec_name = f"{lang}_{version}_rec{server_infix}_infer.onnx"
+    # cls is shared across mobile/server variants in PaddleOCR's released model zoo.
+    cls_name = "ch_ppocr_mobile_v2.0_cls_infer.onnx"
+
+    det_path: str | None = None
+    cls_path: str | None = None
+    rec_path: str | None = None
+    for search_dir in (model_cache_dir, package_models_dir):
+        if not search_dir or not search_dir.is_dir():
+            continue
+        if det_path is None:
+            candidate = search_dir / det_name
+            if candidate.is_file():
+                det_path = str(candidate)
+        if cls_path is None:
+            candidate = search_dir / cls_name
+            if candidate.is_file():
+                cls_path = str(candidate)
+        if rec_path is None:
+            candidate = search_dir / rec_name
+            if candidate.is_file():
+                rec_path = str(candidate)
+    return det_path, cls_path, rec_path
+
+
 @contextmanager
 def _rapidocr_import_context(
     *,
@@ -177,11 +221,35 @@ def _build_runtime_constructor_kwargs(
     model_type: str,
     ocr_version: str,
     model_cache_dir: Path,
+    package_models_dir: Path | None = None,
 ) -> dict[str, Any]:
     try:
         parameters = inspect.signature(runtime_class).parameters
     except (TypeError, ValueError):
         return {}
+
+    has_var_kwargs = any(
+        parameter.kind == inspect.Parameter.VAR_KEYWORD
+        for parameter in parameters.values()
+    )
+    if has_var_kwargs:
+        det_path, cls_path, rec_path = _resolve_rapidocr_model_paths(
+            model_cache_dir=model_cache_dir,
+            package_models_dir=package_models_dir,
+            lang_type=lang_type,
+            ocr_version=ocr_version,
+            model_type=model_type,
+        )
+        kwargs: dict[str, Any] = {}
+        if det_path and rec_path:
+            kwargs["det_model_path"] = det_path
+            kwargs["rec_model_path"] = rec_path
+            if cls_path:
+                kwargs["cls_model_path"] = cls_path
+        if engine_type:
+            kwargs["engine_type"] = engine_type
+        return kwargs
+
     kwargs: dict[str, Any] = {}
     direct_values = {
         "engine_type": engine_type,
@@ -273,6 +341,13 @@ def load_rapidocr_runtime(
         runtime_class = getattr(module, "RapidOCR", None)
         if runtime_class is None:
             raise RuntimeError("RapidOCR runtime class not found")
+        module_file = getattr(module, "__file__", "") or ""
+        # Sentinel must be None (not Path()) — Path() resolves to CWD and would
+        # let _resolve_rapidocr_model_paths inadvertently scan the working
+        # directory if `__file__` were ever missing.
+        package_models_dir: Path | None = (
+            Path(module_file).resolve().parent / "models" if module_file else None
+        )
         with _onnxruntime_intra_op_thread_cap(_RAPIDOCR_INFERENCE_THREAD_LIMIT):
             runtime = runtime_class(
                 **_build_runtime_constructor_kwargs(
@@ -282,6 +357,7 @@ def load_rapidocr_runtime(
                     model_type=model_type,
                     ocr_version=ocr_version,
                     model_cache_dir=model_cache_dir,
+                    package_models_dir=package_models_dir,
                 )
             )
     metadata = {

plugin/plugins/galgame_plugin/test_ocr_pipeline.py

@@ -32,6 +32,7 @@
     TesseractOcrBackend,
     _default_window_scanner,
 )
+from plugin.plugins.galgame_plugin.rapidocr_support import DEFAULT_RAPIDOCR_OCR_VERSION
 
 
 def _noop_logger():
@@ -159,7 +160,7 @@ async def main() -> None:
         rapidocr_engine_type="onnxruntime",
         rapidocr_lang_type="ch",
         rapidocr_model_type="mobile",
-        rapidocr_ocr_version="PP-OCRv5",
+        rapidocr_ocr_version=DEFAULT_RAPIDOCR_OCR_VERSION,
     )
     mgr = OcrReaderManager(logger=_noop_logger(), config=config)
     tick = await mgr.tick(bridge_sdk_available=False, memory_reader_runtime={})

plugin/tests/unit/plugins/test_galgame_rapidocr_support.py

@@ -0,0 +1,164 @@
+from __future__ import annotations
+
+from contextlib import nullcontext
+from pathlib import Path
+from types import SimpleNamespace
+
+import pytest
+
+from plugin.plugins.galgame_plugin import rapidocr_support
+
+
+pytestmark = pytest.mark.plugin_unit
+
+
+class _RapidOcrWithKwargs:
+    captured_kwargs: dict[str, object] | None = None
+
+    def __init__(self, config_path=None, **kwargs) -> None:
+        del config_path
+        type(self).captured_kwargs = dict(kwargs)
+
+
+def _touch(path: Path) -> Path:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text("", encoding="utf-8")
+    return path
+
+
+def test_rapidocr_kwargs_resolve_configured_model_paths(tmp_path: Path) -> None:
+    model_cache_dir = tmp_path / "RapidOCR" / "models"
+    package_models_dir = tmp_path / "package" / "models"
+    det_path = _touch(package_models_dir / "ch_PP-OCRv4_det_infer.onnx")
+    cls_path = _touch(package_models_dir / "ch_ppocr_mobile_v2.0_cls_infer.onnx")
+    rec_path = _touch(package_models_dir / "ch_PP-OCRv4_rec_infer.onnx")
+
+    kwargs = rapidocr_support._build_runtime_constructor_kwargs(
+        _RapidOcrWithKwargs,
+        engine_type="onnxruntime",
+        lang_type="ch",
+        model_type="mobile",
+        ocr_version="PP-OCRv4",
+        model_cache_dir=model_cache_dir,
+        package_models_dir=package_models_dir,
+    )
+
+    assert kwargs == {
+        "det_model_path": str(det_path),
+        "cls_model_path": str(cls_path),
+        "rec_model_path": str(rec_path),
+        "engine_type": "onnxruntime",
+    }
+
+
+def test_rapidocr_kwargs_prefers_user_model_cache(tmp_path: Path) -> None:
+    model_cache_dir = tmp_path / "RapidOCR" / "models"
+    package_models_dir = tmp_path / "package" / "models"
+    user_det_path = _touch(model_cache_dir / "japan_PP-OCRv4_det_infer.onnx")
+    user_rec_path = _touch(model_cache_dir / "japan_PP-OCRv4_rec_infer.onnx")
+    package_cls_path = _touch(package_models_dir / "ch_ppocr_mobile_v2.0_cls_infer.onnx")
+    _touch(package_models_dir / "japan_PP-OCRv4_det_infer.onnx")
+    _touch(package_models_dir / "japan_PP-OCRv4_rec_infer.onnx")
+
+    kwargs = rapidocr_support._build_runtime_constructor_kwargs(
+        _RapidOcrWithKwargs,
+        engine_type="onnxruntime",
+        lang_type="japan",
+        model_type="mobile",
+        ocr_version="PP-OCRv4",
+        model_cache_dir=model_cache_dir,
+        package_models_dir=package_models_dir,
+    )
+
+    assert kwargs["det_model_path"] == str(user_det_path)
+    assert kwargs["rec_model_path"] == str(user_rec_path)
+    assert kwargs["cls_model_path"] == str(package_cls_path)
+
+
+def test_rapidocr_kwargs_resolves_server_variant_filenames(tmp_path: Path) -> None:
+    model_cache_dir = tmp_path / "RapidOCR" / "models"
+    package_models_dir = tmp_path / "package" / "models"
+    server_det_path = _touch(model_cache_dir / "ch_PP-OCRv4_det_server_infer.onnx")
+    server_rec_path = _touch(model_cache_dir / "ch_PP-OCRv4_rec_server_infer.onnx")
+    cls_path = _touch(package_models_dir / "ch_ppocr_mobile_v2.0_cls_infer.onnx")
+    # Mobile variants exist alongside server ones to ensure model_type drives selection.
+    _touch(package_models_dir / "ch_PP-OCRv4_det_infer.onnx")
+    _touch(package_models_dir / "ch_PP-OCRv4_rec_infer.onnx")
+
+    kwargs = rapidocr_support._build_runtime_constructor_kwargs(
+        _RapidOcrWithKwargs,
+        engine_type="onnxruntime",
+        lang_type="ch",
+        model_type="server",
+        ocr_version="PP-OCRv4",
+        model_cache_dir=model_cache_dir,
+        package_models_dir=package_models_dir,
+    )
+
+    assert kwargs == {
+        "det_model_path": str(server_det_path),
+        "rec_model_path": str(server_rec_path),
+        "cls_model_path": str(cls_path),
+        "engine_type": "onnxruntime",
+    }
+
+
+def test_rapidocr_kwargs_omits_model_paths_when_configured_model_is_missing(tmp_path: Path) -> None:
+    model_cache_dir = tmp_path / "RapidOCR" / "models"
+    package_models_dir = tmp_path / "package" / "models"
+    _touch(package_models_dir / "ch_PP-OCRv4_det_infer.onnx")
+    _touch(package_models_dir / "ch_ppocr_mobile_v2.0_cls_infer.onnx")
+    _touch(package_models_dir / "ch_PP-OCRv4_rec_infer.onnx")
+
+    kwargs = rapidocr_support._build_runtime_constructor_kwargs(
+        _RapidOcrWithKwargs,
+        engine_type="onnxruntime",
+        lang_type="ch",
+        model_type="mobile",
+        ocr_version="PP-OCRv5",
+        model_cache_dir=model_cache_dir,
+        package_models_dir=package_models_dir,
+    )
+
+    assert kwargs == {"engine_type": "onnxruntime"}
+
+
+def test_load_rapidocr_runtime_uses_imported_package_models_dir(
+    tmp_path: Path,
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    install_target = tmp_path / "RapidOCR"
+    bundled_package_dir = tmp_path / "bundled" / "rapidocr_onnxruntime"
+    _touch(bundled_package_dir / "__init__.py")
+    det_path = _touch(bundled_package_dir / "models" / "ch_PP-OCRv4_det_infer.onnx")
+    cls_path = _touch(bundled_package_dir / "models" / "ch_ppocr_mobile_v2.0_cls_infer.onnx")
+    rec_path = _touch(bundled_package_dir / "models" / "ch_PP-OCRv4_rec_infer.onnx")
+    _RapidOcrWithKwargs.captured_kwargs = None
+
+    monkeypatch.setattr(
+        rapidocr_support.importlib,
+        "import_module",
+        lambda name: SimpleNamespace(
+            RapidOCR=_RapidOcrWithKwargs,
+            __file__=str(bundled_package_dir / "__init__.py"),
+        ),
+    )
+    monkeypatch.setattr(rapidocr_support, "_onnxruntime_intra_op_thread_cap", lambda _limit: nullcontext())
+
+    runtime, metadata = rapidocr_support.load_rapidocr_runtime(
+        install_target_dir_raw=str(install_target),
+        engine_type="onnxruntime",
+        lang_type="ch",
+        model_type="mobile",
+        ocr_version="PP-OCRv4",
+    )
+
+    assert isinstance(runtime, _RapidOcrWithKwargs)
+    assert _RapidOcrWithKwargs.captured_kwargs == {
+        "det_model_path": str(det_path),
+        "cls_model_path": str(cls_path),
+        "rec_model_path": str(rec_path),
+        "engine_type": "onnxruntime",
+    }
+    assert metadata["detected_path"] == str(bundled_package_dir.resolve())
+    assert metadata["selected_model"] == "PP-OCRv4/ch/mobile"