When getting the numpy datatype, return datetime64[D] for pa.Date32 and return datetime64[ms] for pa.Date64

doc/source/whatsnew/v3.0.0.rst

@@ -1036,6 +1036,7 @@ Conversion
 - Bug in :meth:`Series.convert_dtypes` and :meth:`DataFrame.convert_dtypes` removing timezone information for objects with :class:`ArrowDtype` (:issue:`60237`)
 - Bug in :meth:`Series.reindex` not maintaining ``float32`` type when a ``reindex`` introduces a missing value (:issue:`45857`)
 - Bug in :meth:`to_datetime` and :meth:`to_timedelta` with input ``None`` returning ``None`` instead of ``NaT``, inconsistent with other conversion methods (:issue:`23055`)
+- Bug in :meth:`ArrowDtype._get_common_dtype` and :meth:`ExtentionDtype._get_common_dtype` when using ``date32[pyarrow]`` and ``date64[pyarrow]`` types (:issue:`62343`)
 
 Strings
 ^^^^^^^

pandas/core/dtypes/common.py

@@ -1872,6 +1872,11 @@ def pandas_dtype(dtype) -> DtypeObj:
             result = result()
         return result
 
+    # try a pyarrow dtype
+    from pandas.core.dtypes.dtypes import ArrowDtype
+    if isinstance(dtype, ArrowDtype):
+        return ArrowDtype(dtype)
+
     # try a numpy dtype
     # raise a consistent TypeError if failed
     try:

pandas/core/dtypes/dtypes.py

@@ -2277,6 +2277,12 @@ def name(self) -> str:  # type: ignore[override]
     @cache_readonly
     def numpy_dtype(self) -> np.dtype:
         """Return an instance of the related numpy dtype"""
+        if pa.types.is_date32(self.pyarrow_dtype) or pa.types.is_date64(
+            self.pyarrow_dtype
+        ):
+            # date32 and date64 are pyarrow timestamps but do not have a
+            # corresponding numpy dtype.
+            return np.dtype(object)
         if pa.types.is_timestamp(self.pyarrow_dtype):
             # pa.timestamp(unit).to_pandas_dtype() returns ns units
             # regardless of the pyarrow timestamp units.
@@ -2453,6 +2459,18 @@ def _get_common_dtype(self, dtypes: list[DtypeObj]) -> DtypeObj | None:
 
         null_dtype = type(self)(pa.null())
 
+        # Cover cases where numpy does not have a corresponding dtype, but
+        # only one non-null dtype is received, or all dtypes are null.
+        single_dtype = {
+            dtype
+            for dtype in dtypes
+            if dtype != null_dtype
+        }
+        if len(single_dtype) == 0:
+            return null_dtype
+        if len(single_dtype) == 1:
+            return single_dtype.pop()
+
         new_dtype = find_common_type(
             [
                 dtype.numpy_dtype if isinstance(dtype, ArrowDtype) else dtype

pandas/tests/dtypes/cast/test_find_common_type.py

@@ -173,3 +173,19 @@ def test_interval_dtype_with_categorical(dtype):
 
     result = find_common_type([dtype, cat.dtype])
     assert result == dtype
+
+
+@pytest.mark.parametrize(
+    "dtypes,expected",
+    [
+        (
+            ["date32[pyarrow]", "null[pyarrow]"],
+            "date32[day][pyarrow]",
+        ),
+    ],
+)
+def test_pyarrow_dtypes(dtypes, expected):
+    """Test finding common types with pyarrow dtypes not in numpy."""
+    source_dtypes = [pandas_dtype(dtype) for dtype in dtypes]
+    result = find_common_type(source_dtypes)
+    assert result == pandas_dtype(expected)