ENH/WIP: Index[bool]

Author: jbrockmendel

pandas/_libs/index.pyi

@@ -41,6 +41,7 @@ class ObjectEngine(IndexEngine): ...
 class DatetimeEngine(Int64Engine): ...
 class TimedeltaEngine(DatetimeEngine): ...
 class PeriodEngine(Int64Engine): ...
+class BoolEngine(Uint8Engine): ...
 
 class BaseMultiIndexCodesEngine:
     levels: list[np.ndarray]

pandas/_libs/index.pyx

@@ -795,3 +795,9 @@ cdef class BaseMultiIndexCodesEngine:
 
 # Generated from template.
 include "index_class_helper.pxi"
+
+
+cdef class BoolEngine(UInt8Engine):
+    cdef _check_type(self, object val):
+        if not util.is_bool_object(val):
+            raise KeyError(val)

pandas/conftest.py

@@ -529,7 +529,8 @@ def _create_mi_with_dt64tz_level():
     "num_uint8": tm.makeNumericIndex(100, dtype="uint8"),
     "num_float64": tm.makeNumericIndex(100, dtype="float64"),
     "num_float32": tm.makeNumericIndex(100, dtype="float32"),
-    "bool": tm.makeBoolIndex(10),
+    "bool-object": tm.makeBoolIndex(10).astype(object),
+    "bool-dtype": Index(np.random.randn(10) < 0),
     "categorical": tm.makeCategoricalIndex(100),
     "interval": tm.makeIntervalIndex(100),
     "empty": Index([]),

pandas/core/algorithms.py

@@ -219,9 +219,6 @@ def _reconstruct_data(
     elif is_bool_dtype(dtype):
         values = values.astype(dtype, copy=False)
 
-        # we only support object dtypes bool Index
-        if isinstance(original, ABCIndex):
-            values = values.astype(object, copy=False)
     elif dtype is not None:
         if is_datetime64_dtype(dtype):
             dtype = np.dtype("datetime64[ns]")

pandas/core/dtypes/common.py

@@ -1330,7 +1330,7 @@ def is_bool_dtype(arr_or_dtype) -> bool:
         # we don't have a boolean Index class
         # so its object, we need to infer to
         # guess this
-        return arr_or_dtype.is_object() and arr_or_dtype.inferred_type == "boolean"
+        return arr_or_dtype.inferred_type == "boolean"
     elif isinstance(dtype, ExtensionDtype):
         return getattr(dtype, "_is_boolean", False)
 

pandas/core/indexes/base.py

@@ -481,6 +481,10 @@ def __new__(
             if data.dtype.kind in ["i", "u", "f"]:
                 # maybe coerce to a sub-class
                 arr = data
+            elif data.dtype.kind == "b":
+                # No special subclass, and Index._ensure_array won't do this
+                #  for us.
+                arr = np.asarray(data)
             else:
                 arr = com.asarray_tuplesafe(data, dtype=np.dtype("object"))
 
@@ -672,7 +676,7 @@ def _with_infer(cls, *args, **kwargs):
             # "Union[ExtensionArray, ndarray[Any, Any]]"; expected
             # "ndarray[Any, Any]"
             values = lib.maybe_convert_objects(result._values)  # type: ignore[arg-type]
-            if values.dtype.kind in ["i", "u", "f"]:
+            if values.dtype.kind in ["i", "u", "f", "b"]:
                 return Index(values, name=result.name)
 
         return result
@@ -837,6 +841,8 @@ def _engine(self) -> libindex.IndexEngine:
         # to avoid a reference cycle, bind `target_values` to a local variable, so
         # `self` is not passed into the lambda.
         target_values = self._get_engine_target()
+        if target_values.dtype == bool:
+            return libindex.BoolEngine(target_values)
         return self._engine_type(target_values)
 
     @final
@@ -2548,6 +2554,8 @@ def _is_all_dates(self) -> bool:
         """
         Whether or not the index values only consist of dates.
         """
+        if self.dtype.kind == "b":
+            return False
         return is_datetime_array(ensure_object(self._values))
 
     @cache_readonly
@@ -7048,7 +7056,7 @@ def _maybe_cast_data_without_dtype(
             FutureWarning,
             stacklevel=3,
         )
-    if result.dtype.kind in ["b", "c"]:
+    if result.dtype.kind in ["c"]:
         return subarr
     result = ensure_wrapped_if_datetimelike(result)
     return result

pandas/core/tools/datetimes.py

@@ -925,6 +925,8 @@ def to_datetime(
             result = convert_listlike(arg, format)
     else:
         result = convert_listlike(np.array([arg]), format)[0]
+        if isinstance(arg, bool) and isinstance(result, np.bool_):
+            result = bool(result)  # TODO: avoid this kludge.
 
     #  error: Incompatible return value type (got "Union[Timestamp, NaTType,
     # Series, Index]", expected "Union[DatetimeIndex, Series, float, str,

pandas/core/util/hashing.py

@@ -311,7 +311,7 @@ def _hash_ndarray(
 
     # First, turn whatever array this is into unsigned 64-bit ints, if we can
     # manage it.
-    elif isinstance(dtype, bool):
+    elif dtype == bool:
         vals = vals.astype("u8")
     elif issubclass(dtype.type, (np.datetime64, np.timedelta64)):
         vals = vals.view("i8").astype("u8", copy=False)

pandas/tests/indexes/common.py

@@ -492,7 +492,11 @@ def test_fillna(self, index):
         # GH 11343
         if len(index) == 0:
             return
-        elif isinstance(index, NumericIndex) and is_integer_dtype(index.dtype):
+        elif (
+            isinstance(index, NumericIndex)
+            and is_integer_dtype(index.dtype)
+            or index.dtype == bool
+        ):
             return
         elif isinstance(index, MultiIndex):
             idx = index.copy(deep=True)

pandas/tests/indexes/test_base.py

@@ -320,15 +320,21 @@ def test_view_with_args(self, index):
             "unicode",
             "string",
             pytest.param("categorical", marks=pytest.mark.xfail(reason="gh-25464")),
-            "bool",
+            "bool-object",
+            "bool-dtype",
             "empty",
         ],
         indirect=True,
     )
     def test_view_with_args_object_array_raises(self, index):
-        msg = "Cannot change data-type for object array"
-        with pytest.raises(TypeError, match=msg):
-            index.view("i8")
+        if index.dtype == bool:
+            msg = "When changing to a larger dtype"
+            with pytest.raises(ValueError, match=msg):
+                index.view("i8")
+        else:
+            msg = "Cannot change data-type for object array"
+            with pytest.raises(TypeError, match=msg):
+                index.view("i8")
 
     @pytest.mark.parametrize("index", ["int", "range"], indirect=True)
     def test_astype(self, index):
@@ -587,7 +593,8 @@ def test_append_empty_preserve_name(self, name, expected):
         "index, expected",
         [
             ("string", False),
-            ("bool", False),
+            ("bool-object", False),
+            ("bool-dtype", False),
             ("categorical", False),
             ("int", True),
             ("datetime", False),
@@ -602,7 +609,8 @@ def test_is_numeric(self, index, expected):
         "index, expected",
         [
             ("string", True),
-            ("bool", True),
+            ("bool-object", True),
+            ("bool-dtype", False),
             ("categorical", False),
             ("int", False),
             ("datetime", False),
@@ -617,7 +625,8 @@ def test_is_object(self, index, expected):
         "index, expected",
         [
             ("string", False),
-            ("bool", False),
+            ("bool-object", False),
+            ("bool-dtype", False),
             ("categorical", False),
             ("int", False),
             ("datetime", True),

pandas/tests/indexes/test_index_new.py

@@ -73,7 +73,7 @@ def test_constructor_dtypes_to_object(self, cast_index, vals):
             index = Index(vals)
 
         assert type(index) is Index
-        assert index.dtype == object
+        assert index.dtype == bool
 
     def test_constructor_categorical_to_object(self):
         # GH#32167 Categorical data and dtype=object should return object-dtype

pandas/tests/indexes/test_numpy_compat.py

@@ -51,7 +51,7 @@ def test_numpy_ufuncs_basic(index, func):
         with tm.external_error_raised((TypeError, AttributeError)):
             with np.errstate(all="ignore"):
                 func(index)
-    elif isinstance(index, NumericIndex):
+    elif isinstance(index, NumericIndex) or index.dtype == bool:
         # coerces to float (e.g. np.sin)
         with np.errstate(all="ignore"):
             result = func(index)
@@ -89,7 +89,7 @@ def test_numpy_ufuncs_other(index, func, request):
         with tm.external_error_raised(TypeError):
             func(index)
 
-    elif isinstance(index, NumericIndex):
+    elif isinstance(index, NumericIndex) or index.dtype == bool:
         # Results in bool array
         result = func(index)
         assert isinstance(result, np.ndarray)

pandas/tests/series/indexing/test_setitem.py

@@ -696,7 +696,7 @@ def test_index_where(self, obj, key, expected, val, request):
         mask[key] = True
 
         res = Index(obj).where(~mask, val)
-        tm.assert_index_equal(res, Index(expected))
+        tm.assert_index_equal(res, Index(expected, dtype=expected.dtype))
 
     def test_index_putmask(self, obj, key, expected, val):
         if Index(obj).dtype != obj.dtype:
@@ -707,7 +707,7 @@ def test_index_putmask(self, obj, key, expected, val):
         mask[key] = True
 
         res = Index(obj).putmask(mask, val)
-        tm.assert_index_equal(res, Index(expected))
+        tm.assert_index_equal(res, Index(expected, dtype=expected.dtype))
 
 
 @pytest.mark.parametrize(

pandas/tests/series/methods/test_drop.py

@@ -54,7 +54,8 @@ def test_drop_with_ignore_errors():
 
     # GH 8522
     s = Series([2, 3], index=[True, False])
-    assert s.index.is_object()
+    assert not s.index.is_object()
+    assert s.index.dtype == bool
     result = s.drop(True)
     expected = Series([3], index=[False])
     tm.assert_series_equal(result, expected)