Whittle down failing tests

Author: jbrockmendel

pandas/_testing/__init__.py

@@ -307,6 +307,8 @@ def box_expected(expected, box_cls, transpose: bool = True):
             expected = pd.concat([expected] * 2, ignore_index=True)
     elif box_cls is np.ndarray or box_cls is np.array:
         expected = np.array(expected)
+        if expected.dtype.kind in "iufb" and pd.get_option("mode.pdep16_data_types"):
+            expected = pd.array(expected, copy=False)
     elif box_cls is to_array:
         expected = to_array(expected)
     else:

pandas/core/arrays/base.py

@@ -22,6 +22,8 @@
 
 import numpy as np
 
+from pandas._config import get_option
+
 from pandas._libs import (
     algos as libalgos,
     lib,
@@ -2420,7 +2422,12 @@ def _where(self, mask: npt.NDArray[np.bool_], value) -> Self:
         result = self.copy()
 
         if is_list_like(value):
-            val = value[~mask]
+            if np.ndim(value) == 1 and len(value) == 1:
+                # test_where.test_broadcast if we change to use nullable...
+                #  maybe this should be handled at a higher level?
+                val = value[0]
+            else:
+                val = value[~mask]
         else:
             val = value
 
@@ -2655,6 +2662,10 @@ def _groupby_op(
         if op.how in op.cast_blocklist:
             # i.e. how in ["rank"], since other cast_blocklist methods don't go
             #  through cython_operation
+            if get_option("mode.pdep16_data_types"):
+                from pandas import array as pd_array
+
+                return pd_array(res_values)
             return res_values
 
         if isinstance(self.dtype, StringDtype):

pandas/core/arrays/datetimelike.py

@@ -1701,6 +1701,8 @@ def _groupby_op(
         if op.how in op.cast_blocklist:
             # i.e. how in ["rank"], since other cast_blocklist methods don't go
             #  through cython_operation
+            # if get_option("mode.pdep16_data_types"):
+            #    return pd_array(res_values)  # breaks bc they dont support 2D
             return res_values
 
         # We did a view to M8[ns] above, now we go the other direction

pandas/core/arrays/timedeltas.py

@@ -507,6 +507,10 @@ def __mul__(self, other) -> Self:
             # numpy >= 2.1 may not raise a TypeError
             # and seems to dispatch to others.__rmul__?
             raise TypeError(f"Cannot multiply with {type(other).__name__}")
+        if isinstance(result, type(self)):
+            # e.g. if other is IntegerArray
+            assert result.dtype == self.dtype
+            return result
         return type(self)._simple_new(result, dtype=result.dtype)
 
     __rmul__ = __mul__

pandas/core/frame.py

@@ -141,6 +141,7 @@
 )
 from pandas.core.arrays.sparse import SparseFrameAccessor
 from pandas.core.construction import (
+    array as pd_array,
     ensure_wrapped_if_datetimelike,
     sanitize_array,
     sanitize_masked_array,
@@ -4411,6 +4412,14 @@ def _iset_item_mgr(
     def _set_item_mgr(
         self, key, value: ArrayLike, refs: BlockValuesRefs | None = None
     ) -> None:
+        if get_option("mode.pdep16_data_types"):
+            # TODO: possibly handle this at a lower level?
+            if (
+                isinstance(value, np.ndarray)
+                and value.dtype.kind in "iufb"
+                and value.dtype != np.float16
+            ):
+                value = pd_array(value, copy=False)
         try:
             loc = self._info_axis.get_loc(key)
         except KeyError:

pandas/core/groupby/generic.py

@@ -2166,9 +2166,10 @@ def _cython_transform(
         )
 
         def arr_func(bvalues: ArrayLike) -> ArrayLike:
-            return self._grouper._cython_operation(
+            blk_res = self._grouper._cython_operation(
                 "transform", bvalues, how, 1, **kwargs
             )
+            return blk_res
 
         res_mgr = mgr.apply(arr_func)
 

pandas/core/indexes/base.py

@@ -577,7 +577,11 @@ def __new__(
                 raise ValueError("Index data must be 1-dimensional") from err
             raise
         arr = ensure_wrapped_if_datetimelike(arr)
-        if arr.dtype.kind in "iufb" and get_option("mode.pdep16_data_types"):
+        if (
+            arr.dtype.kind in "iufb"
+            and arr.dtype != np.float16
+            and get_option("mode.pdep16_data_types")
+        ):
             arr = pd_array(arr, copy=False)
 
         klass = cls._dtype_to_subclass(arr.dtype)
@@ -6937,7 +6941,8 @@ def insert(self, loc: int, item) -> Index:
                 return self.astype(dtype).insert(loc, item)
 
         try:
-            if isinstance(arr, ExtensionArray):
+            if isinstance(arr, ExtensionArray) and not isinstance(self, ABCRangeIndex):
+                # RangeIndex's _simple_new expects a range object
                 res_values = arr.insert(loc, item)
                 return type(self)._simple_new(res_values, name=self.name)
             else:

pandas/core/indexes/datetimelike.py

@@ -557,7 +557,7 @@ def _wrap_range_setop(self, other, res_i8) -> Self:
         # because test_setops_preserve_freq fails with _validate_frequency raising.
         # This raising is incorrect, as 'on_freq' is incorrect. This will
         # be fixed by GH#41493
-        res_values = res_i8.values.view(self._data._ndarray.dtype)
+        res_values = np.asarray(res_i8.values).view(self._data._ndarray.dtype)
         result = type(self._data)._simple_new(
             # error: Argument "dtype" to "_simple_new" of "DatetimeArray" has
             # incompatible type "Union[dtype[Any], ExtensionDtype]"; expected

pandas/core/indexes/range.py

@@ -18,6 +18,8 @@
 
 import numpy as np
 
+from pandas._config import get_option
+
 from pandas._libs import (
     index as libindex,
     lib,
@@ -44,7 +46,10 @@
 
 from pandas.core import ops
 import pandas.core.common as com
-from pandas.core.construction import extract_array
+from pandas.core.construction import (
+    array as pd_array,
+    extract_array,
+)
 from pandas.core.indexers import check_array_indexer
 import pandas.core.indexes.base as ibase
 from pandas.core.indexes.base import (
@@ -278,7 +283,10 @@ def _data(self) -> np.ndarray:  # type: ignore[override]
 
         The constructed array is saved in ``_cache``.
         """
-        return np.arange(self.start, self.stop, self.step, dtype=np.int64)
+        data = np.arange(self.start, self.stop, self.step, dtype=np.int64)
+        if get_option("mode.pdep16_data_types"):
+            return pd_array(data)
+        return data
 
     def _get_data_as_items(self) -> list[tuple[str, int]]:
         """return a list of tuples of start, stop, step"""

pandas/core/internals/blocks.py

@@ -361,6 +361,11 @@ def reduce(self, func) -> list[Block]:
         else:
             res_values = result.reshape(-1, 1)
 
+        if self.values.dtype == object:
+            if res_values.dtype.kind == "f":
+                # TODO: this is kludgy; does it mean there is a problem
+                #  at a higher level?
+                res_values = res_values.astype(object)
         nb = self.make_block(res_values)
         return [nb]
 
@@ -2226,6 +2231,8 @@ def new_block_2d(
     klass = get_block_type(values.dtype)
 
     values = maybe_coerce_values(values)
+    if isinstance(values, np.ndarray):
+        assert values.dtype == np.float16 or values.dtype.kind not in "iufb"
     return klass(values, ndim=2, placement=placement, refs=refs)
 
 
@@ -2241,6 +2248,8 @@ def new_block(
     # - check_ndim/ensure_block_shape already checked
     # - maybe_coerce_values already called/unnecessary
     klass = get_block_type(values.dtype)
+    if isinstance(values, np.ndarray):
+        assert values.dtype == np.float16 or values.dtype.kind not in "iufb"
     return klass(values, ndim=ndim, placement=placement, refs=refs)