REF: use _validate_fill_value in Index.insert

pandas/core/dtypes/cast.py

@@ -20,7 +20,7 @@
 
 import numpy as np
 
-from pandas._libs import lib, tslib, tslibs
+from pandas._libs import lib, missing as libmissing, tslib, tslibs
 from pandas._libs.tslibs import (
     NaT,
     OutOfBoundsDatetime,
@@ -584,6 +584,9 @@ def maybe_promote(dtype, fill_value=np.nan):
             dtype = np.dtype(np.object_)
         elif is_integer(fill_value) or (is_float(fill_value) and not isna(fill_value)):
             dtype = np.dtype(np.object_)
+        elif is_valid_nat_for_dtype(fill_value, dtype):
+            # e.g. pd.NA, which is not accepted by Timestamp constructor
+            fill_value = np.datetime64("NaT", "ns")
         else:
             try:
                 fill_value = Timestamp(fill_value).to_datetime64()
@@ -597,6 +600,9 @@ def maybe_promote(dtype, fill_value=np.nan):
         ):
             # TODO: What about str that can be a timedelta?
             dtype = np.dtype(np.object_)
+        elif is_valid_nat_for_dtype(fill_value, dtype):
+            # e.g pd.NA, which is not accepted by the  Timedelta constructor
+            fill_value = np.timedelta64("NaT", "ns")
         else:
             try:
                 fv = Timedelta(fill_value)
@@ -670,7 +676,7 @@ def maybe_promote(dtype, fill_value=np.nan):
                 # e.g. mst is np.complex128 and dtype is np.complex64
                 dtype = mst
 
-    elif fill_value is None:
+    elif fill_value is None or fill_value is libmissing.NA:
         if is_float_dtype(dtype) or is_complex_dtype(dtype):
             fill_value = np.nan
         elif is_integer_dtype(dtype):
@@ -680,7 +686,8 @@ def maybe_promote(dtype, fill_value=np.nan):
             fill_value = dtype.type("NaT", "ns")
         else:
             dtype = np.dtype(np.object_)
-            fill_value = np.nan
+            if fill_value is not libmissing.NA:
+                fill_value = np.nan
     else:
         dtype = np.dtype(np.object_)
 

pandas/core/indexes/base.py

@@ -34,6 +34,7 @@
 
 from pandas.core.dtypes.cast import (
     maybe_cast_to_integer_array,
+    maybe_promote,
     validate_numeric_casting,
 )
 from pandas.core.dtypes.common import (
@@ -4162,24 +4163,6 @@ def _string_data_error(cls, data):
             "to explicitly cast to a numeric type"
         )
 
-    @final
-    def _coerce_scalar_to_index(self, item):
-        """
-        We need to coerce a scalar to a compat for our index type.
-
-        Parameters
-        ----------
-        item : scalar item to coerce
-        """
-        dtype = self.dtype
-
-        if self._is_numeric_dtype and isna(item):
-            # We can't coerce to the numeric dtype of "self" (unless
-            # it's float) if there are NaN values in our output.
-            dtype = None
-
-        return Index([item], dtype=dtype, **self._get_attributes_dict())
-
     def _to_safe_for_reshape(self):
         """
         Convert to object if we are a categorical.
@@ -5490,8 +5473,22 @@ def insert(self, loc: int, item):
         """
         # Note: this method is overridden by all ExtensionIndex subclasses,
         #  so self is never backed by an EA.
+
+        try:
+            item = self._validate_fill_value(item)
+        except TypeError:
+            if is_scalar(item):
+                dtype, item = maybe_promote(self.dtype, item)
+            else:
+                # maybe_promote would raise ValueError
+                dtype = np.dtype(object)
+
+            return self.astype(dtype).insert(loc, item)
+
         arr = np.asarray(self)
-        item = self._coerce_scalar_to_index(item)._values
+
+        # Use Index constructor to ensure we get tuples cast correctly.
+        item = Index([item], dtype=self.dtype)._values
         idx = np.concatenate((arr[:loc], item, arr[loc:]))
         return Index(idx, name=self.name)
 

pandas/core/indexes/numeric.py

@@ -16,6 +16,7 @@
     is_float,
     is_float_dtype,
     is_integer_dtype,
+    is_number,
     is_numeric_dtype,
     is_scalar,
     is_signed_integer_dtype,
@@ -120,15 +121,30 @@ def _validate_fill_value(self, value):
             # force conversion to object
             # so we don't lose the bools
             raise TypeError
-        elif isinstance(value, str) or lib.is_complex(value):
-            raise TypeError
         elif is_scalar(value) and isna(value):
             if is_valid_nat_for_dtype(value, self.dtype):
                 value = self._na_value
+                if self.dtype.kind != "f":
+                    # raise so that caller can cast
+                    raise TypeError
             else:
                 # NaT, np.datetime64("NaT"), np.timedelta64("NaT")
                 raise TypeError
 
+        elif is_scalar(value):
+            if not is_number(value):
+                # e.g. datetime64, timedelta64, datetime, ...
+                raise TypeError
+
+            elif lib.is_complex(value):
+                # at least until we have a ComplexIndx
+                raise TypeError
+
+            elif is_float(value) and self.dtype.kind != "f":
+                if not value.is_integer():
+                    raise TypeError
+                value = int(value)
+
         return value
 
     def _convert_tolerance(self, tolerance, target):
@@ -164,15 +180,6 @@ def _is_all_dates(self) -> bool:
         """
         return False
 
-    @doc(Index.insert)
-    def insert(self, loc: int, item):
-        try:
-            item = self._validate_fill_value(item)
-        except TypeError:
-            return self.astype(object).insert(loc, item)
-
-        return super().insert(loc, item)
-
     def _union(self, other, sort):
         # Right now, we treat union(int, float) a bit special.
         # See https://github.com/pandas-dev/pandas/issues/26778 for discussion

pandas/tests/dtypes/cast/test_promote.py

@@ -110,6 +110,8 @@ def _assert_match(result_fill_value, expected_fill_value):
         assert res_type == ex_type or res_type.__name__ == ex_type.__name__
 
     match_value = result_fill_value == expected_fill_value
+    if match_value is pd.NA:
+        match_value = False
 
     # Note: type check above ensures that we have the _same_ NA value
     # for missing values, None == None (which is checked
@@ -569,7 +571,7 @@ def test_maybe_promote_any_with_object(any_numpy_dtype_reduced, object_dtype):
     _check_promote(dtype, fill_value, expected_dtype, exp_val_for_scalar)
 
 
-@pytest.mark.parametrize("fill_value", [None, np.nan, NaT])
+@pytest.mark.parametrize("fill_value", [None, np.nan, NaT, pd.NA])
 def test_maybe_promote_any_numpy_dtype_with_na(any_numpy_dtype_reduced, fill_value):
     dtype = np.dtype(any_numpy_dtype_reduced)
 
@@ -597,7 +599,10 @@ def test_maybe_promote_any_numpy_dtype_with_na(any_numpy_dtype_reduced, fill_val
     else:
         # all other cases cast to object, and use np.nan as missing value
         expected_dtype = np.dtype(object)
-        exp_val_for_scalar = np.nan
+        if fill_value is pd.NA:
+            exp_val_for_scalar = pd.NA
+        else:
+            exp_val_for_scalar = np.nan
 
     _check_promote(dtype, fill_value, expected_dtype, exp_val_for_scalar)