Bitmask Backed MaskedArray

.pre-commit-config.yaml

@@ -46,6 +46,10 @@ repos:
     -   id: codespell
         types_or: [python, rst, markdown, cython, c]
         additional_dependencies: [tomli]
+        exclude: |
+            (?x)
+            ^pandas/_libs/include/pandas/vendored/nanoarrow.h
+            |pandas/_libs/src/vendored/nanoarrow.c
 -   repo: https://github.com/MarcoGorelli/cython-lint
     rev: v0.15.0
     hooks:
@@ -74,7 +78,11 @@ repos:
     rev: 1.6.1
     hooks:
     -   id: cpplint
-        exclude: ^pandas/_libs/include/pandas/vendored/klib
+        exclude: |
+            (?x)
+            ^pandas/_libs/include/pandas/vendored/klib
+            |pandas/_libs/include/pandas/vendored/nanoarrow.h
+            |pandas/_libs/src/vendored/nanoarrow.c
         args: [
             --quiet,
             '--extensions=c,h',

asv_bench/benchmarks/array.py

@@ -28,6 +28,17 @@ def time_from_float_array(self):
         pd.array(self.values_float, dtype="boolean")
 
 
+class BooleanArrayMem:
+    def setup_cache(self):
+        N = 250_000
+        data = np.array([True] * N)
+        mask = np.array([False] * N)
+        return [pd.arrays.BooleanArray(data, mask)] * 500
+
+    def peakmem_array(self, arrays):
+        return [~x for x in arrays]
+
+
 class IntegerArray:
     def setup(self):
         N = 250_000

pandas/_libs/arrays.pyi

@@ -3,10 +3,13 @@ from typing import Sequence
 import numpy as np
 
 from pandas._typing import (
+    ArrayLike,
     AxisInt,
     DtypeObj,
+    PositionalIndexer,
     Self,
     Shape,
+    type_t,
 )
 
 class NDArrayBacked:
@@ -38,3 +41,35 @@ class NDArrayBacked:
     def _concat_same_type(
         cls, to_concat: Sequence[Self], axis: AxisInt = ...
     ) -> Self: ...
+
+class BitmaskArray:
+    parent: Self
+    def __init__(self, data: np.ndarray | Self) -> None: ...
+    def __len__(self) -> int: ...
+    def __setitem__(self, key: PositionalIndexer, value: ArrayLike | bool) -> None: ...
+    def __getitem__(self, key: PositionalIndexer) -> bool: ...
+    def __invert__(self) -> Self: ...
+    def __and__(self, other: np.ndarray | Self | bool) -> np.ndarray: ...
+    def __or__(self, other: np.ndarray | Self | bool) -> np.ndarray: ...
+    def __xor__(self, other: np.ndarray | Self | bool) -> np.ndarray: ...
+    def __getstate__(self) -> dict: ...
+    def __setstate__(self, other: dict) -> None: ...
+    def __iter__(self): ...
+    @classmethod
+    def concatenate(cls, objs: list[Self], axis: int) -> Self: ...
+    @property
+    def size(self) -> int: ...
+    @property
+    def nbytes(self) -> int: ...
+    @property
+    def bytes(self) -> bytes: ...
+    @property
+    def shape(self) -> tuple[int, ...]: ...
+    @property
+    def dtype(self) -> type_t[bool]: ...
+    def any(self) -> bool: ...
+    def all(self) -> bool: ...
+    def sum(self) -> int: ...
+    def take_1d(self, indices: np.ndarray, axis: int) -> Self: ...
+    def copy(self) -> Self: ...
+    def to_numpy(self) -> np.ndarray: ...