fix: performance of _ls_tree

src/huggingface_hub/hf_file_system.py

@@ -1,4 +1,3 @@
-import copy
 import os
 import re
 import tempfile
@@ -397,7 +396,7 @@ def _ls_tree(
                 parent_path = self._parent(cache_path_info["name"])
                 self.dircache.setdefault(parent_path, []).append(cache_path_info)
                 out.append(cache_path_info)
-        return copy.deepcopy(out)  # copy to not let users modify the dircache
+        return out
 
     def glob(self, path, **kwargs):
         # Set expand_info=False by default to get a x10 speed boost
@@ -561,7 +560,7 @@ def info(self, path: str, refresh: bool = False, revision: Optional[str] = None,
                 if not expand_info:
                     out = {k: out[k] for k in ["name", "size", "type"]}
         assert out is not None
-        return copy.deepcopy(out)  # copy to not let users modify the dircache
+        return out
 
     def exists(self, path, **kwargs):
         """Is there a file at the given path"""

tests/test_hf_file_system.py

@@ -1,3 +1,4 @@
+import copy
 import datetime
 import io
 import os
@@ -391,7 +392,10 @@ def test_find_root_directory_no_revision_with_incomplete_cache(self):
             repo_type="dataset",
         )
 
-        files = self.hffs.find(self.hf_path, detail=True)
+        # Copy the result to make it robust to the cache modifications
+        # See discussion in https://github.com/huggingface/huggingface_hub/pull/2103
+        # for info on why this is not done in `HfFileSystem.find` by default
+        files = copy.deepcopy(self.hffs.find(self.hf_path, detail=True))
 
         # some directories not in cache
         self.hffs.dircache.pop(self.hf_path + "/data/sub_data")