From 414a89216ce8b038316a8cf1957bca8a9582f6bc Mon Sep 17 00:00:00 2001
From: sora <Wanght@meta-stone.com>
Date: Mon, 24 Aug 2026 03:37:15 +0000
Subject: [PATCH] Unify cache root override: loader.get/set_cache_root as
 single source, fixes --cache-dir not relocating .raw blobs

---
 evalharness/data/dataset.py | 21 ++++++++++++---------
 evalharness/data/loader.py  | 13 +++++++++++--
 2 files changed, 23 insertions(+), 11 deletions(-)

diff --git a/evalharness/data/dataset.py b/evalharness/data/dataset.py
index f4aaf79..057f74e 100644
--- a/evalharness/data/dataset.py
+++ b/evalharness/data/dataset.py
@@ -25,11 +25,20 @@ import time
 from pathlib import Path
 from typing import Any, Callable, Dict, Iterator, List, Optional, Union
 
-from .loader import load_raw_records
+from .loader import get_cache_root, load_raw_records
 from .sample import Sample
 from .spec import DatasetSpec
 
-CACHE_ROOT = Path(os.environ.get('EVALHARNESS_CACHE', '~/.cache/evalharness')).expanduser() / 'datasets'
+# Backward-compatible alias: the canonical root lives in loader.get_cache_root()
+# (a function, so runtime overrides via set_cache_root are always respected).
+CACHE_ROOT = get_cache_root()
+
+
+def set_cache_root(path) -> None:
+    """Override the cache root at runtime (used by the CLI --cache-dir flag)."""
+    from . import loader
+
+    loader.set_cache_root(path)
 
 
 def safe_filename(s: str, max_length: int = 255) -> str:
@@ -43,12 +52,6 @@ def gen_hash(s: str) -> str:
     return hashlib.md5(s.encode('utf-8')).hexdigest()
 
 
-def set_cache_root(path) -> None:
-    """Override the cache root at runtime (used by the CLI --cache-dir flag)."""
-    global CACHE_ROOT
-    CACHE_ROOT = Path(path).expanduser() / 'datasets'
-
-
 class Dataset:
     """Sequence-like lazy dataset. Nothing is downloaded until first use."""
 
@@ -71,7 +74,7 @@ class Dataset:
         if self.spec.version:
             parts += f'_{safe_filename(self.spec.version)}'
         subdir = f'{parts}-{gen_hash(key)[:6]}'
-        return CACHE_ROOT / safe_filename(self.spec.name) / subdir
+        return get_cache_root() / 'datasets' / safe_filename(self.spec.name) / subdir
 
     @property
     def is_materialized(self) -> bool:
diff --git a/evalharness/data/loader.py b/evalharness/data/loader.py
index 67c7296..e650bef 100644
--- a/evalharness/data/loader.py
+++ b/evalharness/data/loader.py
@@ -28,6 +28,16 @@ _RESERVED_PARAMS = {'hub', 'ms_files', 'filter_column'}
 _MS_API = 'https://www.modelscope.cn/api/v1/datasets'
 
 
+def get_cache_root() -> Path:
+    """Canonical evalharness cache root ($EVALHARNESS_CACHE or ~/.cache/evalharness)."""
+    return Path(os.environ.get('EVALHARNESS_CACHE', '~/.cache/evalharness')).expanduser()
+
+
+def set_cache_root(path) -> None:
+    """Override the cache root at runtime (also moves the .raw blob store)."""
+    os.environ['EVALHARNESS_CACHE'] = str(path)
+
+
 def load_raw_records(spec: DatasetSpec, raw_dir: Optional[Path] = None) -> List[Dict[str, Any]]:
     """Load raw records (list of dicts) from the spec's source.
 
@@ -220,8 +230,7 @@ def _load_from_modelscope(spec: DatasetSpec, raw_dir: Optional[Path] = None) ->
             f'modelscope {spec.source!r}. Available (first 10): {files[:10]}'
         )
     # shared blob store: download once per repo, hardlink into each cache entry
-    blob_dir = Path(os.environ.get('EVALHARNESS_CACHE', '~/.cache/evalharness')).expanduser()
-    blob_dir = blob_dir / '.raw' / hashlib.md5(spec.source.encode()).hexdigest()[:10]
+    blob_dir = get_cache_root() / '.raw' / hashlib.md5(spec.source.encode()).hexdigest()[:10]
     records: List[Dict[str, Any]] = []
     blobs: List[Path] = []
     for path in selected: