PERF: extract_array earlier in DataFrame construction

pandas/core/construction.py

@@ -420,9 +420,9 @@ def extract_array(
                 return obj._values
             return obj
 
-        obj = obj.array
+        obj = obj._values
 
-    if extract_numpy and isinstance(obj, ABCPandasArray):
+    elif extract_numpy and isinstance(obj, ABCPandasArray):
         obj = obj.to_numpy()
 
     return obj

pandas/core/internals/blocks.py

@@ -1920,11 +1920,11 @@ def get_block_type(values, dtype: DtypeObj | None = None):
 
 
 def new_block(values, placement, *, ndim: int, klass=None) -> Block:
+    # caller is responsible for ensuring values is NOT a PandasArray
 
     if not isinstance(placement, BlockPlacement):
         placement = BlockPlacement(placement)
 
-    values, _ = extract_pandas_array(values, None, ndim)
     check_ndim(values, placement, ndim)
 
     if klass is None:

pandas/core/internals/managers.py

@@ -1775,11 +1775,6 @@ def create_block_manager_from_blocks(
     return mgr
 
 
-# We define this here so we can override it in tests.extension.test_numpy
-def _extract_array(obj):
-    return extract_array(obj, extract_numpy=True)
-
-
 def create_block_manager_from_arrays(
     arrays,
     names: Index,
@@ -1791,7 +1786,7 @@ def create_block_manager_from_arrays(
     # assert isinstance(axes, list)
     # assert all(isinstance(x, Index) for x in axes)
 
-    arrays = [_extract_array(x) for x in arrays]
+    arrays = [extract_array(x, extract_numpy=True) for x in arrays]
 
     try:
         blocks = _form_blocks(arrays, names, axes, consolidate)

pandas/core/strings/accessor.py

@@ -34,6 +34,7 @@
 from pandas.core.dtypes.missing import isna
 
 from pandas.core.base import NoNewAttributesMixin
+from pandas.core.construction import extract_array
 
 if TYPE_CHECKING:
     from pandas import (
@@ -213,10 +214,7 @@ def _validate(data):
         # see _libs/lib.pyx for list of inferred types
         allowed_types = ["string", "empty", "bytes", "mixed", "mixed-integer"]
 
-        # TODO: avoid kludge for tests.extension.test_numpy
-        from pandas.core.internals.managers import _extract_array
-
-        data = _extract_array(data)
+        data = extract_array(data)
 
         values = getattr(data, "categories", data)  # categorical / normal
 

pandas/tests/extension/test_numpy.py

@@ -23,32 +23,17 @@
     ExtensionDtype,
     PandasDtype,
 )
-from pandas.core.dtypes.generic import ABCPandasArray
 
 import pandas as pd
 import pandas._testing as tm
 from pandas.core.arrays.numpy_ import PandasArray
-from pandas.core.internals import (
-    blocks,
-    managers,
-)
+from pandas.core.internals import blocks
 from pandas.tests.extension import base
 
 # TODO(ArrayManager) PandasArray
 pytestmark = td.skip_array_manager_not_yet_implemented
 
 
-def _extract_array_patched(obj):
-    if isinstance(obj, (pd.Index, pd.Series)):
-        obj = obj._values
-    if isinstance(obj, ABCPandasArray):
-        # TODO for reasons unclear, we get here in a couple of tests
-        #  with PandasArray._typ *not* patched
-        obj = obj.to_numpy()
-
-    return obj
-
-
 def _can_hold_element_patched(obj, element) -> bool:
     if isinstance(element, PandasArray):
         element = element.to_numpy()
@@ -98,7 +83,6 @@ def allow_in_pandas(monkeypatch):
     """
     with monkeypatch.context() as m:
         m.setattr(PandasArray, "_typ", "extension")
-        m.setattr(managers, "_extract_array", _extract_array_patched)
         m.setattr(blocks, "can_hold_element", _can_hold_element_patched)
         m.setattr(tm.asserters, "assert_attr_equal", _assert_attr_equal)
         yield

pandas/tests/internals/test_internals.py

@@ -1376,9 +1376,11 @@ def test_make_block_no_pandas_array(block_maker):
     # PandasArray, no dtype
     result = block_maker(arr, slice(len(arr)), ndim=arr.ndim)
     assert result.dtype.kind in ["i", "u"]
-    assert result.is_extension is False
 
     if block_maker is make_block:
+        # new_block requires caller to unwrap PandasArray
+        assert result.is_extension is False
+
         # PandasArray, PandasDtype
         result = block_maker(arr, slice(len(arr)), dtype=arr.dtype, ndim=arr.ndim)
         assert result.dtype.kind in ["i", "u"]