This is an automated email from the ASF dual-hosted git repository.
pitrou pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/arrow.git
The following commit(s) were added to refs/heads/main by this push:
new 61c96ca061 GH-49905: [Archery] Fix archery benchmark diff with pandas
3 (#49912)
61c96ca061 is described below
commit 61c96ca0612ae46ef05becfeb5f987197180cb2e
Author: Antoine Prouvost <[email protected]>
AuthorDate: Wed May 6 15:22:01 2026 +0200
GH-49905: [Archery] Fix archery benchmark diff with pandas 3 (#49912)
### Rationale for this change
Make archery benchmark diff work with pandas >= 3.0
### What changes are included in this PR?
Another way of patching the module
### Are these changes tested?
Yes, locally running `archery benchmark diff` with both pandas 2 and 3
### Are there any user-facing changes?
No
* GitHub Issue: #49905
Authored-by: AntoinePrv <[email protected]>
Signed-off-by: Antoine Pitrou <[email protected]>
---
dev/archery/archery/compat.py | 18 ++++++++++++++++--
1 file changed, 16 insertions(+), 2 deletions(-)
diff --git a/dev/archery/archery/compat.py b/dev/archery/archery/compat.py
index 33ff869668..a80e949383 100644
--- a/dev/archery/archery/compat.py
+++ b/dev/archery/archery/compat.py
@@ -17,6 +17,7 @@
import pathlib
import sys
+import types
def _is_path_like(path):
@@ -47,8 +48,21 @@ def _stringify_path(path):
def _import_pandas():
- # ARROW-13425: avoid importing PyArrow from Pandas
- sys.modules['pyarrow'] = None
+ # ARROW-13425: hide the in-tree PyArrow from pandas so it doesn't try
+ # to use a possibly-broken development build.
+ # ARROW-49905: The original workaround sets ``sys.modules['pyarrow'] =
None``,
+ # but pandas >= 3.0 has a different import mechanism resulting in
+ # ``AttributeError: 'NoneType' object has no attribute 'Array'``
+ # on the first DataFrame construction (in
``pandas._libs.lib.is_pyarrow_array``).
+ # Instead, we install a minimal stub where ``isinstance(obj, pa.Array)``
+ # check returns False for every real object.
+ if "pyarrow" not in sys.modules:
+ stub = types.ModuleType("pyarrow")
+ stub.Array = type("Array", (), {})
+ stub.ChunkedArray = type("ChunkedArray", (), {})
+ # pandas.compat.pyarrow parses ``pa.__version__`` at import time.
+ stub.__version__ = "0.0.0"
+ sys.modules["pyarrow"] = stub
import pandas as pd
return pd
