test(core): add cuda.core.__all__ vs public docs consistency check

cuda_core/cuda/core/__init__.py

@@ -69,45 +69,51 @@ class _PatchedProperty(metaclass=_PatchedPropMeta):
 
 
 from cuda.core import checkpoint, system, utils
-from cuda.core._context import Context, ContextOptions
-from cuda.core._device import Device
-from cuda.core._device_resources import (
-    DeviceResources,
-    SMResource,
-    SMResourceOptions,
-    WorkqueueResource,
-    WorkqueueResourceOptions,
+from cuda.core._context import *
+from cuda.core._context import __all__ as _context_all
+from cuda.core._device import *
+from cuda.core._device import __all__ as _device_all
+from cuda.core._device_resources import *
+from cuda.core._device_resources import __all__ as _device_resources_all
+from cuda.core._event import *
+from cuda.core._event import __all__ as _event_all
+from cuda.core._graphics import *
+from cuda.core._graphics import __all__ as _graphics_all
+from cuda.core._host import *
+from cuda.core._host import __all__ as _host_all
+from cuda.core._launch_config import *
+from cuda.core._launch_config import __all__ as _launch_config_all
+from cuda.core._launcher import *
+from cuda.core._launcher import __all__ as _launcher_all
+from cuda.core._linker import *
+from cuda.core._linker import __all__ as _linker_all
+from cuda.core._memory import *
+from cuda.core._memory import __all__ as _memory_all
+from cuda.core._module import *
+from cuda.core._module import __all__ as _module_all
+from cuda.core._program import *
+from cuda.core._program import __all__ as _program_all
+from cuda.core._stream import *
+from cuda.core._stream import __all__ as _stream_all
+from cuda.core._tensor_map import *
+from cuda.core._tensor_map import __all__ as _tensor_map_all
+
+__all__ = (
+    _context_all
+    + _device_all
+    + _device_resources_all
+    + _event_all
+    + _graphics_all
+    + _host_all
+    + _launch_config_all
+    + _launcher_all
+    + _linker_all
+    + _memory_all
+    + _module_all
+    + _program_all
+    + _stream_all
+    + _tensor_map_all
 )
-from cuda.core._event import Event, EventOptions
-from cuda.core._graphics import GraphicsResource
-from cuda.core._host import Host
-from cuda.core._launch_config import LaunchConfig
-from cuda.core._launcher import launch
-from cuda.core._linker import Linker, LinkerOptions
-from cuda.core._memory import (
-    Buffer,
-    DeviceMemoryResource,
-    DeviceMemoryResourceOptions,
-    GraphMemoryResource,
-    LegacyPinnedMemoryResource,
-    ManagedBuffer,
-    ManagedMemoryResource,
-    ManagedMemoryResourceOptions,
-    MemoryResource,
-    PinnedMemoryResource,
-    PinnedMemoryResourceOptions,
-    VirtualMemoryResource,
-    VirtualMemoryResourceOptions,
-)
-from cuda.core._module import Kernel, ObjectCode
-from cuda.core._program import Program, ProgramOptions
-from cuda.core._stream import (
-    LEGACY_DEFAULT_STREAM,
-    PER_THREAD_DEFAULT_STREAM,
-    Stream,
-    StreamOptions,
-)
-from cuda.core._tensor_map import TensorMapDescriptor, TensorMapDescriptorOptions
 
 # isort: split
 # Texture/surface types live under the cuda.core.texture namespace (not the

cuda_core/cuda/core/_device.pyi

@@ -1023,4 +1023,5 @@ class Device:
         .. versionadded:: 1.1.0
         """
 _tls = threading.local()
-_lock = threading.Lock()
+_lock = threading.Lock()
+__all__ = ['Device']

cuda_core/cuda/core/_device.pyx

@@ -61,6 +61,8 @@ _tls = threading.local()
 _lock = threading.Lock()
 cdef bint _is_cuInit = False
 
+__all__ = ['Device']
+
 
 cdef class DeviceProperties:
     """

cuda_core/cuda/core/_event.pyi

@@ -180,6 +180,7 @@ class IPCEventDescriptor:
 
     def __reduce__(self) -> tuple[object, ...]:
         ...
+__all__ = ['Event', 'EventOptions']
 
 def _reduce_event(event: Event) -> tuple[object, ...]:
     ...

cuda_core/cuda/core/_event.pyx

@@ -43,6 +43,8 @@ if TYPE_CHECKING:
     import cuda.bindings.driver  # no-cython-lint
     from cuda.core._device import Device
 
+__all__ = ['Event', 'EventOptions']
+
 
 @dataclass
 cdef class EventOptions:

cuda_core/cuda/core/_host.py

@@ -6,6 +6,8 @@
 import threading
 from typing import ClassVar
 
+__all__ = ["Host"]
+
 
 class Host:
     """Host (CPU) location for managed-memory operations.

cuda_core/cuda/core/_launch_config.pyi

@@ -66,6 +66,7 @@ class LaunchConfig:
     def __hash__(self) -> int:
         ...
 _LAUNCH_CONFIG_ATTRS = ('grid', 'cluster', 'block', 'shmem_size', 'is_cooperative')
+__all__ = ['LaunchConfig']
 
 def _to_native_launch_config(config: LaunchConfig) -> object:
     """Convert LaunchConfig to native driver CUlaunchConfig.

cuda_core/cuda/core/_launch_config.pyx

@@ -15,6 +15,8 @@ from cuda.core._utils.cuda_utils import (
 
 _LAUNCH_CONFIG_ATTRS = ('grid', 'cluster', 'block', 'shmem_size', 'is_cooperative')
 
+__all__ = ['LaunchConfig']
+
 
 cdef class LaunchConfig:
     """Customizable launch options.

cuda_core/cuda/core/_launcher.pyi

@@ -8,6 +8,7 @@ from cuda.core._stream import Stream
 from cuda.core.graph import GraphBuilder
 from cuda.core.typing import IsStreamType
 
+__all__ = ['launch']
 
 def launch(stream: Stream | GraphBuilder | IsStreamType, config: LaunchConfig, kernel: Kernel, *kernel_args) -> None:
     """Launches a :obj:`~_module.Kernel`

cuda_core/cuda/core/_launcher.pyx

@@ -24,6 +24,8 @@ if TYPE_CHECKING:
     from cuda.core.graph import GraphBuilder
     from cuda.core.typing import IsStreamType
 
+__all__ = ['launch']
+
 
 def launch(
     stream: Stream | GraphBuilder | IsStreamType,

cuda_core/cuda/core/_memory/__init__.py

@@ -1,13 +1,34 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: Apache-2.0
 
 from ._buffer import *
+from ._buffer import __all__ as _buffer_all
 from ._device_memory_resource import *
+from ._device_memory_resource import __all__ as _device_memory_resource_all
 from ._graph_memory_resource import *
+from ._graph_memory_resource import __all__ as _graph_memory_resource_all
 from ._ipc import *
+from ._ipc import __all__ as _ipc_all
 from ._legacy import *
-from ._managed_buffer import ManagedBuffer
+from ._legacy import __all__ as _legacy_all
+from ._managed_buffer import *
+from ._managed_buffer import __all__ as _managed_buffer_all
 from ._managed_memory_resource import *
+from ._managed_memory_resource import __all__ as _managed_memory_resource_all
 from ._pinned_memory_resource import *
+from ._pinned_memory_resource import __all__ as _pinned_memory_resource_all
 from ._virtual_memory_resource import *
+from ._virtual_memory_resource import __all__ as _virtual_memory_resource_all
+
+__all__ = (
+    _buffer_all
+    + _device_memory_resource_all
+    + _graph_memory_resource_all
+    + _ipc_all
+    + _legacy_all
+    + _managed_buffer_all
+    + _managed_memory_resource_all
+    + _pinned_memory_resource_all
+    + _virtual_memory_resource_all
+)

cuda_core/cuda/core/_memory/_managed_buffer.py

@@ -25,6 +25,8 @@
     from cuda.core._stream import Stream
     from cuda.core.graph import GraphBuilder
 
+__all__ = ["ManagedBuffer"]
+
 
 _INT_SIZE = 4
 

cuda_core/cuda/core/_stream.pyi

@@ -212,6 +212,7 @@ class Stream:
             Newly created graph builder object.
 
         """
+__all__ = ['LEGACY_DEFAULT_STREAM', 'PER_THREAD_DEFAULT_STREAM', 'Stream', 'StreamOptions']
 LEGACY_DEFAULT_STREAM: Stream = Stream._legacy_default()
 PER_THREAD_DEFAULT_STREAM: Stream = Stream._per_thread_default()
 

cuda_core/cuda/core/_stream.pyx

@@ -47,6 +47,9 @@ if TYPE_CHECKING:
     from cuda.core._device import Device
     from cuda.core.graph import GraphBuilder
 
+__all__ = ['LEGACY_DEFAULT_STREAM', 'PER_THREAD_DEFAULT_STREAM', 'Stream', 'StreamOptions']
+
+
 @dataclass
 cdef class StreamOptions:
     """Customizable :obj:`~_stream.Stream` options.

cuda_core/cuda/core/_tensor_map.pyi

@@ -284,6 +284,7 @@ class TensorMapDescriptor:
 
     def __repr__(self) -> str:
         ...
+__all__ = ['TensorMapDescriptor', 'TensorMapDescriptorOptions']
 _TMA_DT_UINT8 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT8)
 _TMA_DT_UINT16 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT16)
 _TMA_DT_UINT32 = int(cydriver.CU_TENSOR_MAP_DATA_TYPE_UINT32)

cuda_core/cuda/core/_tensor_map.pyx

@@ -47,6 +47,8 @@ try:
 except ImportError:
     ml_bfloat16 = None
 
+__all__ = ['TensorMapDescriptor', 'TensorMapDescriptorOptions']
+
 
 class TensorMapDataType(enum.IntEnum):
     """Data types for tensor map descriptors.

cuda_core/cuda/core/graph/__init__.py

@@ -2,7 +2,19 @@
 #
 # SPDX-License-Identifier: Apache-2.0
 
+from . import _graph_builder, _graph_definition, _graph_node, _subclasses
 from ._graph_builder import *
 from ._graph_definition import *
 from ._graph_node import *
 from ._subclasses import *
+
+# Aggregate the star-imported submodule exports so ``cuda.core.graph`` carries
+# an explicit ``__all__`` derived from its parts (no manual list to drift).
+__all__ = [
+    *_graph_builder.__all__,
+    *_graph_definition.__all__,
+    *_graph_node.__all__,
+    *_subclasses.__all__,
+]
+
+del _graph_builder, _graph_definition, _graph_node, _subclasses

cuda_core/cuda/core/system/__init__.py

@@ -12,8 +12,10 @@
 
 __all__ = [
     "CUDA_BINDINGS_NVML_IS_COMPATIBLE",
+    "get_driver_branch",
     "get_kernel_mode_driver_version",
     "get_num_devices",
+    "get_nvml_version",
     "get_process_name",
     "get_user_mode_driver_version",
 ]
@@ -40,7 +42,6 @@
     from .exceptions import *
     from .exceptions import __all__ as _exceptions_all
 
-    __all__.append("get_nvml_version")
     __all__.extend(_device_all)
     __all__.extend(_system_events_all)
     __all__.extend(_exceptions_all)

cuda_core/docs/source/api_nvml.rst

@@ -45,3 +45,11 @@ Types
 
    Device
    NvlinkInfo
+
+Constants
+---------
+
+.. autosummary::
+   :toctree: generated/
+
+   CUDA_BINDINGS_NVML_IS_COMPATIBLE

cuda_core/docs/source/api_private.rst

@@ -43,10 +43,10 @@ CUDA runtime
 
    :template: autosummary/cyclass.rst
 
+   DeviceResources
    _device.DeviceProperties
-   _device_resources.DeviceResources
-   _memory._ipc.IPCAllocationHandle
-   _memory._ipc.IPCBufferDescriptor
+   IPCAllocationHandle
+   IPCBufferDescriptor
    _memory._managed_buffer.AccessedBySetProxy
 
 
@@ -125,3 +125,36 @@ NVML
    system.typing.TemperatureThresholds
    system.typing.ThermalController
    system.typing.ThermalTarget
+
+   system.NvmlError
+   system.UninitializedError
+   system.InvalidArgumentError
+   system.NotSupportedError
+   system.NoPermissionError
+   system.AlreadyInitializedError
+   system.NotFoundError
+   system.InsufficientSizeError
+   system.InsufficientPowerError
+   system.DriverNotLoadedError
+   system.TimeoutError
+   system.IrqIssueError
+   system.LibraryNotFoundError
+   system.FunctionNotFoundError
+   system.CorruptedInforomError
+   system.GpuIsLostError
+   system.ResetRequiredError
+   system.OperatingSystemError
+   system.LibRmVersionMismatchError
+   system.InUseError
+   system.MemoryError
+   system.NoDataError
+   system.VgpuEccNotSupportedError
+   system.InsufficientResourcesError
+   system.FreqNotSupportedError
+   system.ArgumentVersionMismatchError
+   system.DeprecatedError
+   system.NotReadyError
+   system.GpuNotFoundError
+   system.InvalidStateError
+   system.ResetTypeNotSupportedError
+   system.UnknownError

cuda_core/pixi.toml

@@ -21,6 +21,7 @@ pytest-randomly = "*"
 pytest-repeat = "*"
 pytest-rerunfailures = "*"
 cloudpickle = "*"
+docutils = "*"
 psutil = "*"
 pyglet = "*"
 

cuda_core/pyproject.toml

@@ -69,6 +69,7 @@ test = [
     "pytest-timeout==2.4.0",
     "cloudpickle==3.1.2",
     "psutil==7.2.2",
+    "docutils==0.23",
     # TODO: remove the Python 3.15 guard once 3.15 is officially supported
     "cffi==2.0.0; python_version < '3.15'",
 ]