instance-repo 1.3.2.dev3__tar.gz → 1.3.2.dev5__tar.gz

{instance_repo-1.3.2.dev3 → instance_repo-1.3.2.dev5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: instance-repo
-Version: 1.3.2.dev3
+Version: 1.3.2.dev5
 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-1.3.2.dev3 → instance_repo-1.3.2.dev5}/instance_repo/__init__.py

@@ -18,7 +18,7 @@ from .errors import (
     TransportError,
 )
 
-__version__ = "1.3.2.dev3"
+__version__ = "1.3.2.dev5"
 
 __all__ = [
     "Repo", "load_instances",

{instance_repo-1.3.2.dev3 → instance_repo-1.3.2.dev5}/instance_repo/clients/datasets.py

@@ -161,7 +161,8 @@ class DatasetsClient:
         except TransportError as e:
             if (e.biz_code == 92001 and fallback_to_oss
                     and self._content is not None):
-                return self._discover_detail(name, storage_env=environment)
+                return self._discover_detail(name, storage_env=environment,
+                                             metadata_model=metadata_model)
             raise
 
     def update(self, name: str, *, display_name: str | None = None,
@@ -396,13 +397,14 @@ class DatasetsClient:
     # ── OSS fallback discovery (private) ─────────────────────────────────
 
     def _discover_detail(self, name: str, *,
-                         storage_env: str | None = None) -> DatasetDetail:
+                         storage_env: str | None = None,
+                         metadata_model: str | None = None) -> DatasetDetail:
         from ..store.oss_discovery import (
             parse_dataset_level, build_synthetic_dataset_detail,
             _warn_if_truncated)
         info = parse_dataset_level(
             self._content, name, oss_prefix=self._oss_prefix,
-            metadata_model=self._metadata_model)
+            metadata_model=metadata_model or self._metadata_model)
         _warn_if_truncated(info, name, 500)
         if not info.versions and not info.unversioned_splits:
             from ..errors import DatasetNotFound

{instance_repo-1.3.2.dev3 → instance_repo-1.3.2.dev5}/instance_repo/clients/versions.py

@@ -229,7 +229,7 @@ class VersionsClient:
         except TransportError as e:
             if (e.biz_code == 92001 and fallback_to_oss
                     and self._content is not None):
-                return self._discover_versions(dataset)
+                return self._discover_versions(dataset, metadata_model=metadata_model)
             raise
 
     def list_paged(self, dataset: str, *, status: str | None = None,
@@ -264,7 +264,8 @@ class VersionsClient:
             if (e.biz_code == 92001 and fallback_to_oss
                     and self._content is not None):
                 return self._discover_versions_paged(
-                    dataset, status, int(page), int(page_size))
+                    dataset, status, int(page), int(page_size),
+                    metadata_model=metadata_model)
             raise
         rows = data.get("versions", data) if isinstance(data, dict) else data
         rows = rows if isinstance(rows, list) else []
@@ -295,7 +296,8 @@ class VersionsClient:
         except TransportError as e:
             if (e.biz_code == 92001 and fallback_to_oss
                     and self._content is not None):
-                return self._discover_version(dataset, version)
+                return self._discover_version(dataset, version,
+                                              metadata_model=metadata_model)
             raise
 
     def status(self, dataset: str, version: str, *,
@@ -891,12 +893,13 @@ class VersionsClient:
 
     # ── OSS fallback discovery (private) ─────────────────────────────────
 
-    def _discover_versions(self, dataset: str) -> list:
+    def _discover_versions(self, dataset: str,
+                           metadata_model: str | None = None) -> list:
         from ..store.oss_discovery import (
             parse_dataset_level, build_synthetic_version, _warn_if_truncated)
         info = parse_dataset_level(
             self._content, dataset, oss_prefix=self._oss_prefix,
-            metadata_model=self._discovery_mm())
+            metadata_model=metadata_model or self._discovery_mm())
         _warn_if_truncated(info, dataset, 500)
         # split_first 无版本布局（只有 split）也是发现成功，只是没有版本可列
         if not info.versions and not info.unversioned_splits:
@@ -905,20 +908,22 @@ class VersionsClient:
         return [build_synthetic_version(dataset, v) for v in info.versions]
 
     def _discover_versions_paged(self, dataset: str, status: str | None,
-                                 page: int, page_size: int):
+                                 page: int, page_size: int,
+                                 metadata_model: str | None = None):
         """OSS 回退的 list_paged：全量探测后按 status 过滤 + 内存分页。"""
         from ..store.oss_discovery import _in_memory_paged
-        items = self._discover_versions(dataset)
+        items = self._discover_versions(dataset, metadata_model=metadata_model)
         if (status or "").strip():
             items = [v for v in items if v.status == status.strip()]
         return _in_memory_paged(items, int(page), int(page_size))
 
-    def _discover_version(self, dataset: str, version: str):
+    def _discover_version(self, dataset: str, version: str,
+                          metadata_model: str | None = None):
         from ..store.oss_discovery import (
             parse_dataset_level, build_synthetic_version, _warn_if_truncated)
         info = parse_dataset_level(
             self._content, dataset, oss_prefix=self._oss_prefix,
-            metadata_model=self._discovery_mm())
+            metadata_model=metadata_model or self._discovery_mm())
         _warn_if_truncated(info, dataset, 500)
         for v in info.versions:
             if v.version == version:

{instance_repo-1.3.2.dev3 → instance_repo-1.3.2.dev5}/instance_repo/store/oss.py

@@ -100,6 +100,22 @@ def _unconfirmed_private_endpoint_error(endpoint: str) -> PrivateNetworkRequired
     )
 
 
+def _import_oss2():  # pragma: no cover - 真实 oss2
+    """惰性 import oss2，屏蔽其编译期 SyntaxWarning。
+
+    oss2 2.19.1 的 ``sign_url`` docstring 含非 raw 字符串里的 ``\\&``
+    （示例 URL 的 OSSAccessKeyId=...&Expires=...），首次 import 时 Python
+    编译器对其发出 ``SyntaxWarning: invalid escape sequence '\\&'``——纯第三方
+    文档噪音，与请求拼接无关。仅抑制 SyntaxWarning（编译期），不影响运行时
+    其他警告。
+    """
+    import warnings as _warnings
+    with _warnings.catch_warnings():
+        _warnings.simplefilter("ignore", SyntaxWarning)
+        import oss2
+    return oss2
+
+
 def _default_bucket_factory(sts: dict, bucket: str, *,
                             private_download: bool = False):  # pragma: no cover - 真实 oss2
     """Build the default oss2 bucket, hardening private-download transport.
@@ -108,7 +124,7 @@ def _default_bucket_factory(sts: dict, bucket: str, *,
     this transport guarantee. The built-in private-download path forces HTTPS, ignores ambient
     proxy variables, and rejects redirects before requests can follow them.
     """
-    import oss2
+    oss2 = _import_oss2()
 
     endpoint = sts["endpoint"]
     session = None
@@ -139,7 +155,7 @@ def _default_bucket_factory(sts: dict, bucket: str, *,
 
 
 def _default_resumable_upload(bucket, key: str, filename: str):  # pragma: no cover - 真实 oss2
-    import oss2
+    oss2 = _import_oss2()
     oss2.resumable_upload(bucket, key, filename)
 
 

{instance_repo-1.3.2.dev3 → instance_repo-1.3.2.dev5}/instance_repo/store/oss_discovery.py

@@ -23,6 +23,7 @@ from ..models import (
     DatasetInstance,
     DatasetSplit,
     DatasetVersion,
+    ManifestEntry,
     PagedResult,
     SplitManifest,
 )
@@ -564,6 +565,62 @@ def build_synthetic_instances(
     return instances
 
 
+def _instance_dir(dataset: str, version: str, split: str | None,
+                  instance_id: str, suffix: str, oss_prefix: str) -> str:
+    """拼接实例的 -assets/-report 目录；version 为空表示无版本 split 布局。"""
+    l1, l2 = P.split_dataset(dataset)
+    base = f"{oss_prefix}/{l1}/{l2}" + (f"/{version}" if version else "")
+    if split and split != "default":
+        return f"{base}/{split}-{suffix}/{instance_id}/"
+    return f"{base}-{suffix}/{instance_id}/"
+
+
+def _format_from_key(key: str) -> str:
+    if key.endswith(".jsonl"):
+        return "jsonl"
+    if key.endswith(".json"):
+        return "json"
+    if key.endswith(".tgz") or key.endswith(".tar.gz"):
+        return "tgz"
+    return ""
+
+
+def _instance_manifest(
+    content_store,
+    dataset: str,
+    version: str,
+    instance_id: str,
+    split: str | None,
+    metadata_key: str,
+    *,
+    oss_prefix: str,
+) -> list[ManifestEntry]:
+    """用 LIST 枚举实例的文件结构（metadata + {split}-assets + {split}-report）。
+
+    发现模式（list-only STS）无 GetObject，读不到 metadata JSON 内容，但目录
+    结构可通过 LIST 完整呈现——恰好满足「查看实例的目录文件结构」需求。
+    """
+    keys = [metadata_key]
+    for suffix in ("assets", "report"):
+        try:
+            keys.extend(k for k in content_store.list_objects(
+                _instance_dir(dataset, version, split, instance_id,
+                              suffix, oss_prefix),
+                list_only=True) if not k.endswith("/"))
+        except Exception:  # noqa: BLE001
+            pass  # best-effort：无 assets/report 目录时忽略
+    bucket = getattr(content_store, "bucket_name", "") or ""
+    return [
+        ManifestEntry(
+            file_path=f"oss://{bucket}/{k}" if bucket else k,
+            content_type="file",
+            content_format=_format_from_key(k),
+            object_path=k,
+        )
+        for k in keys
+    ]
+
+
 def build_synthetic_instance(
     content_store,
     dataset: str,
@@ -573,11 +630,17 @@ def build_synthetic_instance(
     *,
     oss_prefix: str = "swe/datasets",
     read_content: bool = False,
+    include_files: bool = True,
 ) -> DatasetInstance | None:
     """Build a single :class:`DatasetInstance` by exact key lookup in OSS.
 
     Returns ``None`` if the metadata JSON does not exist.  ``read_content``
     语义同 :func:`build_synthetic_instances`（发现模式默认跳过 JSON 回填）。
+
+    ``include_files``（默认 True）用 LIST 枚举实例文件结构（metadata + assets
+    + report 目录）回填 ``manifest``——文件名为真实值，digest/大小留空
+    （list-only STS 无 GetObject）。仅 instance get 单条路径开启，避免列表
+    路径 N 倍 LIST 开销。
     """
     if version:
         key = P.metadata_key(
@@ -609,6 +672,13 @@ def build_synthetic_instance(
         split=split or "default",
         registered=False,
     )
+    if include_files:
+        try:
+            inst.manifest = _instance_manifest(
+                content_store, dataset, version, instance_id, split, key,
+                oss_prefix=oss_prefix)
+        except Exception:  # noqa: BLE001
+            pass  # best-effort：manifest 枚举失败不阻断实例返回
     if read_content:
         data = _read_json_object(content_store, key)
         if data:

{instance_repo-1.3.2.dev3 → instance_repo-1.3.2.dev5}/instance_repo.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: instance-repo
-Version: 1.3.2.dev3
+Version: 1.3.2.dev5
 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-1.3.2.dev3 → instance_repo-1.3.2.dev5}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "instance-repo"
-version = "1.3.2.dev3"
+version = "1.3.2.dev5"
 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"