Now complies with pre-commit requirements and added an entry in v3.0.0.rst

Author: afonso-antunes

doc/source/whatsnew/v3.0.0.rst

@@ -712,6 +712,7 @@ MultiIndex
 - :func:`MultiIndex.get_level_values` accessing a :class:`DatetimeIndex` does not carry the frequency attribute along (:issue:`58327`, :issue:`57949`)
 - Bug in :class:`DataFrame` arithmetic operations in case of unaligned MultiIndex columns (:issue:`60498`)
 - Bug in :class:`DataFrame` arithmetic operations with :class:`Series` in case of unaligned MultiIndex (:issue:`61009`)
+- Bug in :class:`MultiIndex.concat` where extension dtypes such as ``timestamp[pyarrow]`` were silently coerced to ``object`` instead of preserving their original dtype (:issue:`58421`)
 -
 
 I/O

pandas/core/reshape/concat.py

@@ -22,6 +22,7 @@
 
 from pandas.core.dtypes.common import (
     is_bool,
+    is_extension_array_dtype,
     is_scalar,
 )
 from pandas.core.dtypes.concat import concat_compat
@@ -36,6 +37,7 @@
     factorize_from_iterables,
 )
 import pandas.core.common as com
+from pandas.core.construction import array as pd_array
 from pandas.core.indexes.api import (
     Index,
     MultiIndex,
@@ -47,10 +49,6 @@
 )
 from pandas.core.internals import concatenate_managers
 
-from pandas.core.dtypes.common import is_extension_array_dtype
-
-from pandas.core.construction import array
-
 if TYPE_CHECKING:
     from collections.abc import (
         Callable,
@@ -832,9 +830,9 @@ def _concat_indexes(indexes) -> Index:
     sample = indexes[0]
     try:
         # this helps preserve extension types like timestamp[pyarrow]
-        arr = array(values, dtype=sample.dtype)
+        arr = pd_array(values, dtype=sample.dtype)
     except Exception:
-        arr = array(values)  # fallback
+        arr = pd_array(values)  # fallback
 
     return Index(arr)
 
@@ -906,7 +904,7 @@ def _make_concat_multiindex(indexes, keys, levels=None, names=None) -> MultiInde
                 unzipped = list(zip(*concat_index))
                 for i, level_values in enumerate(unzipped):
                     # reconstruct each level using original dtype
-                    arr = array(level_values, dtype=original_dtypes[i])
+                    arr = pd_array(level_values, dtype=original_dtypes[i])
                     level_codes, _ = factorize_from_iterable(arr)
                     levels.append(ensure_index(arr))
                     codes_list.append(level_codes)

pandas/tests/frame/methods/test_concat_arrow_index.py

@@ -1,13 +1,14 @@
-import pandas as pd
 import pytest
 
+import pandas as pd
 
 schema = {
     "id": "int64[pyarrow]",
     "time": "timestamp[s][pyarrow]",
     "value": "float[pyarrow]",
 }
 
+
 @pytest.mark.parametrize("dtype", ["timestamp[s][pyarrow]"])
 def test_concat_preserves_pyarrow_timestamp(dtype):
     dfA = (
@@ -38,9 +39,13 @@ def test_concat_preserves_pyarrow_timestamp(dtype):
     )
 
     df = pd.concat([dfA, dfB], keys=[0, 1], names=["run"])
-    
-    # chech whether df.index is multiIndex
-    assert isinstance(df.index, pd.MultiIndex), f"Expected MultiIndex, but received {type(df.index)}"
-    
+
+    # check whether df.index is multiIndex
+    assert isinstance(df.index, pd.MultiIndex), (
+        f"Expected MultiIndex, but received {type(df.index)}"
+    )
+
     # Verifying special dtype timestamp[s][pyarrow] stays intact after concat
-    assert df.index.levels[2].dtype == dtype, f"Expected {dtype}, but received {df.index.levels[2].dtype}"
+    assert df.index.levels[2].dtype == dtype, (
+        f"Expected {dtype}, but received {df.index.levels[2].dtype}"
+    )