handling tz aware ts explicitly, preventing re-deriving of ArrowDtype. Moving test to appropriate file.

Author: Koookadooo

pandas/core/dtypes/cast.py

@@ -1113,7 +1113,7 @@ def convert_dtypes(
     else:
         inferred_dtype = input_array.dtype
 
-    if dtype_backend == "pyarrow":
+    if dtype_backend == "pyarrow" and not isinstance(inferred_dtype, ArrowDtype):
         from pandas.core.arrays.arrow.array import to_pyarrow_type
         from pandas.core.arrays.string_ import StringDtype
 

pandas/core/dtypes/dtypes.py

@@ -2275,18 +2275,31 @@ def name(self) -> str:  # type: ignore[override]
     @cache_readonly
     def numpy_dtype(self) -> np.dtype:
         """Return an instance of the related numpy dtype."""
-        # For string-like arrow dtypes, pa.string().to_pandas_dtype() = object
-        # so we handle them explicitly.
-        if pa.types.is_string(self.pyarrow_dtype) or pa.types.is_large_string(
-            self.pyarrow_dtype
-        ):
+        pa_type = self.pyarrow_dtype
+
+        # handle tz-aware timestamps
+        if pa.types.is_timestamp(pa_type):
+            if pa_type.tz is not None:
+                # preserve tz by NOT calling numpy_dtype for this dtype.
+                return np.dtype("datetime64[ns]")
+            else:
+                # For tz-naive timestamps, just return the corresponding unit
+                return np.dtype(f"datetime64[{pa_type.unit}]")
+
+        if pa.types.is_duration(pa_type):
+            return np.dtype(f"timedelta64[{pa_type.unit}]")
+
+        if pa.types.is_string(pa_type) or pa.types.is_large_string(pa_type):
             return np.dtype(str)
 
         try:
-            np_dtype = self.pyarrow_dtype.to_pandas_dtype()
+            np_dtype = pa_type.to_pandas_dtype()
+            if isinstance(np_dtype, DatetimeTZDtype):
+                # In theory we shouldn't get here for tz-aware arrow timestamps
+                # if we've handled them above. This is a fallback.
+                return np.dtype("datetime64[ns]")
             return np.dtype(np_dtype)
         except (NotImplementedError, TypeError):
-            # Fallback if something unexpected happens
             return np.dtype(object)
 
     @cache_readonly

pandas/tests/dtypes/test_dtypes.py

@@ -1103,25 +1103,6 @@ def test_update_dtype_errors(self, bad_dtype):
         with pytest.raises(ValueError, match=msg):
             dtype.update_dtype(bad_dtype)
 
-
-class TestArrowDtype:
-    @pytest.mark.parametrize(
-        "tz", ["UTC", "America/New_York", "Europe/London", "Asia/Tokyo"]
-    )
-    def test_pyarrow_timestamp_tz_preserved(self, tz):
-        pytest.importorskip("pyarrow")
-        s = Series(
-            pd.to_datetime(range(5), unit="h", utc=True).tz_convert(tz),
-            dtype=f"timestamp[ns, tz={tz}][pyarrow]",
-        )
-
-        result = s.convert_dtypes(dtype_backend="pyarrow")
-        assert result.dtype == s.dtype, f"Expected {s.dtype}, got {result.dtype}"
-
-        assert str(result.iloc[0].tzinfo) == str(s.iloc[0].tzinfo)
-        tm.assert_series_equal(result, s)
-
-
 @pytest.mark.parametrize(
     "dtype", [CategoricalDtype, IntervalDtype, DatetimeTZDtype, PeriodDtype]
 )

pandas/tests/extension/test_arrow.py

@@ -51,6 +51,7 @@
 
 import pandas as pd
 import pandas._testing as tm
+from pandas import Series
 from pandas.api.extensions import no_default
 from pandas.api.types import (
     is_bool_dtype,
@@ -3505,3 +3506,19 @@ def test_map_numeric_na_action():
     result = ser.map(lambda x: 42, na_action="ignore")
     expected = pd.Series([42.0, 42.0, np.nan], dtype="float64")
     tm.assert_series_equal(result, expected)
+
+
+@pytest.mark.parametrize(
+    "tz", ["UTC", "America/New_York", "Europe/London", "Asia/Tokyo"]
+)
+def test_pyarrow_timestamp_tz_preserved(tz):
+    s = Series(
+        pd.to_datetime(range(5), unit="h", utc=True).tz_convert(tz),
+        dtype=f"timestamp[ns, tz={tz}][pyarrow]"
+    )
+
+    result = s.convert_dtypes(dtype_backend="pyarrow")
+    assert result.dtype == s.dtype, f"Expected {s.dtype}, got {result.dtype}"
+
+    assert str(result.iloc[0].tzinfo) == str(s.iloc[0].tzinfo)
+    tm.assert_series_equal(result, s)