REF: ensure soft_convert_objects returns ArrayLike (#39961)

Author: jbrockmendel

pandas/_libs/lib.pyx

@@ -1986,7 +1986,7 @@ def maybe_convert_numeric(ndarray[object] values, set na_values,
 
     Parameters
     ----------
-    values : ndarray
+    values : ndarray[object]
         Array of object elements to convert.
     na_values : set
         Set of values that should be interpreted as NaN.
@@ -2007,7 +2007,8 @@ def maybe_convert_numeric(ndarray[object] values, set na_values,
 
     Returns
     -------
-    Array of converted object values to numerical ones.
+    np.ndarray
+        Array of converted object values to numerical ones.
     """
     if len(values) == 0:
         return np.array([], dtype='i8')
@@ -2159,7 +2160,7 @@ def maybe_convert_objects(ndarray[object] objects, bint try_float=False,
 
     Parameters
     ----------
-    values : ndarray
+    values : ndarray[object]
         Array of object elements to convert.
     try_float : bool, default False
         If an array-like object contains only float or NaN values is
@@ -2179,7 +2180,7 @@ def maybe_convert_objects(ndarray[object] objects, bint try_float=False,
 
     Returns
     -------
-    Array of converted object values to more specific dtypes if applicable.
+    np.ndarray or ExtensionArray
     """
     cdef:
         Py_ssize_t i, n
@@ -2309,7 +2310,10 @@ def maybe_convert_objects(ndarray[object] objects, bint try_float=False,
     if seen.datetimetz_:
         if is_datetime_with_singletz_array(objects):
             from pandas import DatetimeIndex
-            return DatetimeIndex(objects)
+            dti = DatetimeIndex(objects)
+
+            # unbox to DatetimeArray
+            return dti._data
         seen.object_ = True
 
     if not seen.object_:

pandas/core/dtypes/cast.py

@@ -1234,7 +1234,7 @@ def soft_convert_objects(
     numeric: bool = True,
     timedelta: bool = True,
     copy: bool = True,
-):
+) -> ArrayLike:
     """
     Try to coerce datetime, timedelta, and numeric object-dtype columns
     to inferred dtype.
@@ -1249,7 +1249,7 @@ def soft_convert_objects(
 
     Returns
     -------
-    np.ndarray
+    np.ndarray or ExtensionArray
     """
     validate_bool_kwarg(datetime, "datetime")
     validate_bool_kwarg(numeric, "numeric")

pandas/core/internals/blocks.py

@@ -128,7 +128,6 @@ class Block(PandasObject):
 
     __slots__ = ["_mgr_locs", "values", "ndim"]
     is_numeric = False
-    is_float = False
     is_bool = False
     is_object = False
     is_extension = False
@@ -1290,7 +1289,7 @@ def _interpolate(
         data = self.values if inplace else self.values.copy()
 
         # only deal with floats
-        if not self.is_float:
+        if self.dtype.kind != "f":
             if self.dtype.kind not in ["i", "u"]:
                 return [self]
             data = data.astype(np.float64)
@@ -1955,7 +1954,6 @@ def is_bool(self):
 
 class FloatBlock(NumericBlock):
     __slots__ = ()
-    is_float = True
 
     def to_native_types(
         self, na_rep="", float_format=None, decimal=".", quoting=None, **kwargs
@@ -2144,7 +2142,6 @@ def to_native_types(self, na_rep="NaT", **kwargs):
 
 class DatetimeBlock(DatetimeLikeBlockMixin):
     __slots__ = ()
-    is_datetime = True
     fill_value = np.datetime64("NaT", "ns")
     _dtype = fill_value.dtype
     _holder = DatetimeArray

pandas/tests/reshape/concat/test_concat.py

@@ -39,7 +39,7 @@ def test_concat_copy(self):
         result = concat([df, df2, df3], axis=1, copy=False)
 
         for b in result._mgr.blocks:
-            if b.is_float:
+            if b.dtype.kind == "f":
                 assert b.values.base is df._mgr.blocks[0].values.base
             elif b.dtype.kind in ["i", "u"]:
                 assert b.values.base is df2._mgr.blocks[0].values.base
@@ -50,7 +50,7 @@ def test_concat_copy(self):
         df4 = DataFrame(np.random.randn(4, 1))
         result = concat([df, df2, df3, df4], axis=1, copy=False)
         for b in result._mgr.blocks:
-            if b.is_float:
+            if b.dtype.kind == "f":
                 assert b.values.base is None
             elif b.dtype.kind in ["i", "u"]:
                 assert b.values.base is df2._mgr.blocks[0].values.base