REF: unwrap PandasArray earlier (#43728)

Author: jbrockmendel

pandas/core/construction.py

@@ -49,7 +49,10 @@
     is_object_dtype,
     is_timedelta64_ns_dtype,
 )
-from pandas.core.dtypes.dtypes import DatetimeTZDtype
+from pandas.core.dtypes.dtypes import (
+    DatetimeTZDtype,
+    PandasDtype,
+)
 from pandas.core.dtypes.generic import (
     ABCExtensionArray,
     ABCIndex,
@@ -494,6 +497,10 @@ def sanitize_array(
     if isinstance(data, ma.MaskedArray):
         data = sanitize_masked_array(data)
 
+    if isinstance(dtype, PandasDtype):
+        # Avoid ending up with a PandasArray
+        dtype = dtype.numpy_dtype
+
     # extract ndarray or ExtensionArray, ensure we have no PandasArray
     data = extract_array(data, extract_numpy=True)
 

pandas/core/dtypes/cast.py

@@ -77,6 +77,7 @@
     DatetimeTZDtype,
     ExtensionDtype,
     IntervalDtype,
+    PandasDtype,
     PeriodDtype,
 )
 from pandas.core.dtypes.generic import (
@@ -1305,6 +1306,9 @@ def astype_array_safe(
         raise TypeError(msg)
 
     dtype = pandas_dtype(dtype)
+    if isinstance(dtype, PandasDtype):
+        # Ensure we don't end up with a PandasArray
+        dtype = dtype.numpy_dtype
 
     try:
         new_values = astype_array(values, dtype, copy=copy)

pandas/core/internals/blocks.py

@@ -1128,7 +1128,6 @@ def interpolate(
             **kwargs,
         )
 
-        interp_values = maybe_coerce_values(interp_values)
         nbs = [self.make_block_same_class(interp_values)]
         return self._maybe_downcast(nbs, downcast)
 
@@ -1903,10 +1902,7 @@ def maybe_coerce_values(values: ArrayLike) -> ArrayLike:
     -------
     values : np.ndarray or ExtensionArray
     """
-
-    # Note: the only test that needs extract_array here is one where we
-    #  pass PandasDtype to Series.astype, then need to extract PandasArray here.
-    values = extract_array(values, extract_numpy=True)
+    # Caller is responsible for ensuring PandasArray is already extracted.
 
     if isinstance(values, np.ndarray):
         values = ensure_wrapped_if_datetimelike(values)

pandas/core/internals/managers.py

@@ -71,7 +71,6 @@
     ensure_block_shape,
     extend_blocks,
     get_block_type,
-    maybe_coerce_values,
     new_block,
 )
 from pandas.core.internals.ops import (
@@ -989,7 +988,6 @@ def iget(self, i: int) -> SingleBlockManager:
 
         # shortcut for select a single-dim from a 2-dim BM
         bp = BlockPlacement(slice(0, len(values)))
-        values = maybe_coerce_values(values)
         nb = type(block)(values, placement=bp, ndim=1)
         return SingleBlockManager(nb, self.axes[1])
 

pandas/tests/extension/test_numpy.py

@@ -336,6 +336,19 @@ def test_merge(self, data, na_value):
         # Fails creating expected (key column becomes a PandasDtype because)
         super().test_merge(data, na_value)
 
+    @pytest.mark.parametrize(
+        "in_frame",
+        [
+            True,
+            pytest.param(
+                False,
+                marks=pytest.mark.xfail(reason="PandasArray inconsistently extracted"),
+            ),
+        ],
+    )
+    def test_concat(self, data, in_frame):
+        super().test_concat(data, in_frame)
+
 
 class TestSetitem(BaseNumPyTests, base.BaseSetitemTests):
     @skip_nested