instance-repo 1.4.0.dev3__tar.gz → 1.4.0.dev6__tar.gz

{instance_repo-1.4.0.dev3 → instance_repo-1.4.0.dev6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: instance-repo
-Version: 1.4.0.dev3
+Version: 1.4.0.dev6
 Summary: InstanceRepo SDK — 评测 Instance 统一存储客户端（SDK-only；控制面由 apiserver 承载）
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown

{instance_repo-1.4.0.dev3 → instance_repo-1.4.0.dev6}/instance_repo/__init__.py

@@ -20,7 +20,7 @@ from .errors import (
     RegistrationFailed, TransportError,
 )
 
-__version__ = "1.4.0.dev3"
+__version__ = "1.4.0.dev6"
 
 __all__ = [
     "Repo", "load_instances",

{instance_repo-1.4.0.dev3 → instance_repo-1.4.0.dev6}/instance_repo/clients/instances.py

@@ -154,6 +154,41 @@ class InstancesClient:
         prefix = getattr(self._p, "oss_prefix", None) or P.DEFAULT_PREFIX
         return Layout(model, prefix=prefix)
 
+    def _instance_assets_object_root(self, dataset: str, instance_id: str,
+                                     version: str, split: str,
+                                     metadata_model: str | None = None, *,
+                                     environment: str | None = None) -> str:
+        """资产目录相对于 object API 根的路径；注册的 object_path 优先。"""
+        lay = self._layout(metadata_model)
+        if version and self._reg is not None:
+            # 非标准 Version storage_path 可能已经是 object API 根，不能
+            # 无条件加版本名。服务端 object_path 已相对于它的真实根计算。
+            env = environment
+            if env is None:
+                env = getattr(self._p, "dataset_environment", None) or None
+            meta = self._reg.get(
+                dataset, version, instance_id, split,
+                **_registry_overrides(env, lay.model))
+            if meta is not None:
+                for entry in meta.manifest:
+                    relative = (entry.object_path or "").rstrip("/")
+                    if (entry.content_type == "directory" and relative
+                            and relative.endswith("/" + instance_id)):
+                        _validate_custom_asset_rel(relative)
+                        return relative
+                for entry in meta.manifest:
+                    relative = entry.object_path or ""
+                    if (entry.content_type == "file"
+                            and relative.endswith("/" + instance_id + ".json")):
+                        _validate_custom_asset_rel(relative)
+                        metadata_dir = relative.rsplit("/", 1)[0]
+                        if metadata_dir == split or metadata_dir.endswith("/" + split):
+                            return f"{metadata_dir}-assets/{instance_id}"
+        dataset_root = lay.dataset_root(dataset)
+        assets_dir = lay.assets_dir(
+            dataset, instance_id, version=version, split=split)
+        return assets_dir[len(dataset_root):].rstrip("/")
+
     # ── 只读 ──
     def validate(self, local_path):
         """校验实例：自动检测格式（目录→harbor，dir-notask→custom，.json→swe）。"""
@@ -414,7 +449,8 @@ class InstancesClient:
         """列举实例资产目录下的文件/子目录（自动翻页）。
 
         调用控制面 ``GET /apis/v1/datasets/objects``，返回指定实例的
-        ``{split}-assets/{instance_id}/`` 路径下的目录和文件列表。
+        ``{version}/{split}-assets/{instance_id}/`` 路径下的目录和文件列表；
+        无版本时省略 ``{version}/``。prefix 相对于 Dataset 根目录。
         自动翻页合并所有结果，直到 ``truncated=False`` 或达到 ``max_pages``。
 
         Parameters
@@ -455,12 +491,6 @@ class InstancesClient:
             raise SchemaError(
                 "instance_id is required for instances.list_objects")
         split = (split or "default").strip()
-        # 计算实例资产根前缀：{split}-assets/{instance_id}/
-        instance_asset_prefix = f"{split}-assets/{instance_id.strip()}"
-        if path:
-            full_prefix = f"{instance_asset_prefix}/{path.strip().strip('/')}"
-        else:
-            full_prefix = instance_asset_prefix
         # environment / metadata_model：优先显式值，其次 profile 派生
         env = (environment or "").strip()
         if not env:
@@ -469,6 +499,14 @@ class InstancesClient:
         if requested_model is None:
             requested_model = getattr(self._p, "metadata_model", None)
         mm = business_metadata_model(requested_model)
+        # object API 的 prefix 相对于 Dataset 根，而非 Version 根。
+        instance_asset_prefix = self._instance_assets_object_root(
+            dataset.strip(), instance_id.strip(), version, split, mm,
+            environment=env or None)
+        if path:
+            full_prefix = f"{instance_asset_prefix}/{path.strip().strip('/')}"
+        else:
+            full_prefix = instance_asset_prefix
         # 自动翻页
         all_dirs: list = []
         all_files: list = []
@@ -1467,11 +1505,11 @@ class InstancesClient:
                     # 递归列举：从根开始，遇到子目录继续深入
                     asset_rels = []
                     stack: list[str | None] = [None]  # None = 实例资产根
-                    # list_objects 返回的 f.path 包含完整前缀
-                    # （如 test-assets/105-creep-forecast/solution/solve.py），
-                    # 需要去掉 {split}-assets/{instance_id}/ 前缀得到相对路径。
+                    # 返回路径相对于 Dataset 根；去掉包含可选版本段的
+                    # 实例资产前缀，才是相对于实例目录的下载路径。
                     split_str = (split or "default").strip()
-                    assets_root = f"{split_str}-assets/{instance_id.strip()}/"
+                    assets_root = self._instance_assets_object_root(
+                        dataset, instance_id, version, split_str) + "/"
                     while stack:
                         cur_path = stack.pop()
                         objs = self.list_objects(
@@ -1520,7 +1558,8 @@ class InstancesClient:
                     asset_rels: list[str] = []
                     stack: list[str | None] = [None]
                     split_str = (split or "default").strip()
-                    assets_root = f"{split_str}-assets/{instance_id.strip()}/"
+                    assets_root = self._instance_assets_object_root(
+                        dataset, instance_id, version, split_str) + "/"
                     while stack:
                         cur = stack.pop()
                         page = self.list_objects(

{instance_repo-1.4.0.dev3 → instance_repo-1.4.0.dev6}/instance_repo/errors.py

@@ -110,6 +110,13 @@ class DatasetNotFound(BizCodeError):
     biz_code = 94021
 
 
+class DatasetNamespaceConflict(BizCodeError):
+    """L1 命名空间冲突：根与子数据集不能共存（根先子后合法、子先根后拦截）。"""
+
+    code = "E_DATASET_NAMESPACE_CONFLICT"
+    biz_code = 92024
+
+
 class TransportError(InstanceRepoError):
     """apiserver HTTP 非预期响应。
 
@@ -169,6 +176,7 @@ CODE_TO_EXC: dict[str, type[InstanceRepoError]] = {
         CredExpired, PrivateNetworkRequired, WriteOutcomeUnknown,
         DatasetAccessDenied, DatasetClaimRequired, DatasetWriteDenied,
         DatasetIdentifierConflict, DatasetIdentifierRequired, DatasetNotFound,
+        DatasetNamespaceConflict,
     )
 }
 
@@ -178,6 +186,7 @@ BIZ_CODE_TO_EXC: dict[int, type[BizCodeError]] = {
     for e in (
         DatasetAccessDenied, DatasetClaimRequired, DatasetWriteDenied,
         DatasetIdentifierConflict, DatasetIdentifierRequired, DatasetNotFound,
+        DatasetNamespaceConflict,
     )
 }
 

{instance_repo-1.4.0.dev3 → instance_repo-1.4.0.dev6}/instance_repo.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: instance-repo
-Version: 1.4.0.dev3
+Version: 1.4.0.dev6
 Summary: InstanceRepo SDK — 评测 Instance 统一存储客户端（SDK-only；控制面由 apiserver 承载）
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown

{instance_repo-1.4.0.dev3 → instance_repo-1.4.0.dev6}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "instance-repo"
-version = "1.4.0.dev3"
+version = "1.4.0.dev6"
 description = "InstanceRepo SDK — 评测 Instance 统一存储客户端（SDK-only；控制面由 apiserver 承载）"
 requires-python = ">=3.10"
 readme = "README.md"