fix(library): alternative implementation that cleans up using a finalizer

Author: cpcloud

cuda_pathfinder/cuda/pathfinder/_dynamic_libs/load_dl_linux.py

@@ -38,11 +38,20 @@ def _load_libdl() -> ctypes.CDLL:
 LIBDL.dlerror.argtypes = []
 LIBDL.dlerror.restype = ctypes.c_char_p
 
+LIBDL.dlclose.argtypes = [ctypes.c_void_p]
+LIBDL.dlclose.restype = ctypes.c_int
+
 # First appeared in 2004-era glibc. Universally correct on Linux for all practical purposes.
 RTLD_DI_LINKMAP = 2
 RTLD_DI_ORIGIN = 6
 
 
+def unload_dl(handle: ctypes.c_void_p) -> None:
+    result = LIBDL.dlclose(handle)
+    if result:
+        raise RuntimeError(LIBDL.dlerror())
+
+
 class _LinkMapLNameView(ctypes.Structure):
     """
     Prefix-only view of glibc's `struct link_map` used **solely** to read `l_name`.

cuda_pathfinder/cuda/pathfinder/_dynamic_libs/load_dl_windows.py

@@ -46,6 +46,9 @@
 kernel32.AddDllDirectory.argtypes = [ctypes.wintypes.LPCWSTR]
 kernel32.AddDllDirectory.restype = ctypes.c_void_p  # DLL_DIRECTORY_COOKIE
 
+kernel32.FreeLibrary.argtypes = [ctypes.wintypes.HMODULE]
+kernel32.FreeLibrary.restype = ctypes.c_bool
+
 
 def ctypes_handle_to_unsigned_int(handle: ctypes.wintypes.HMODULE) -> int:
     """Convert ctypes HMODULE to unsigned int."""
@@ -157,3 +160,9 @@ def load_with_abs_path(libname: str, found_path: str) -> LoadedDL:
         raise RuntimeError(f"Failed to load DLL at {found_path}: Windows error {error_code}")
 
     return LoadedDL(found_path, False, ctypes_handle_to_unsigned_int(handle))
+
+
+def unload_dl(handle: ctypes.c_void_p) -> None:
+    result = kernel32.FreeLibrary(handle)
+    if not result:
+        raise RuntimeError(f"Failed to load windows DLL with error code: {ctypes.GetLastError()}")

cuda_pathfinder/cuda/pathfinder/_dynamic_libs/load_nvidia_dynamic_lib.py

@@ -1,9 +1,11 @@
 # SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
+import ctypes
 import functools
 import struct
 import sys
+import weakref
 
 from cuda.pathfinder._dynamic_libs.find_nvidia_dynamic_lib import _FindNvidiaDynamicLib
 from cuda.pathfinder._dynamic_libs.load_dl_common import LoadedDL, load_dependencies
@@ -14,12 +16,14 @@
         check_if_already_loaded_from_elsewhere,
         load_with_abs_path,
         load_with_system_search,
+        unload_dl,
     )
 else:
     from cuda.pathfinder._dynamic_libs.load_dl_linux import (
         check_if_already_loaded_from_elsewhere,
         load_with_abs_path,
         load_with_system_search,
+        unload_dl,
     )
 
 
@@ -117,4 +121,13 @@ def load_nvidia_dynamic_lib(libname: str) -> LoadedDL:
             f" Currently running: {pointer_size_bits}-bit Python"
             f" {sys.version_info.major}.{sys.version_info.minor}"
         )
-    return _load_lib_no_cache(libname)
+
+    library = _load_lib_no_cache(libname)
+
+    # Ensure that the library is unloaded after GC runs on `library`
+    #
+    # We only need the address, so the rest of whatever is in `library` is free
+    # to be cleaned up. The integer address is immutable, so it gets copied
+    # upon being referenced here
+    weakref.finalize(library, unload_dl, ctypes.c_void_p(library._handle_uint))
+    return library