PR feedback refactor

1. Changed fixture to provide a function that empties the stack of
   contexts. The function has hidden max_iters bound. If exceeded,
   a RuntimeError is raised

2. Modified _device_unset_current utility function to return a boolean.
   True is returned is a context was popped, False if the stack was
   already empty.

Author: oleksandr-pavlyk

cuda_core/tests/conftest.py

@@ -31,29 +31,45 @@ def init_cuda():
     device = Device()
     device.set_current()
     yield
-    _device_unset_current()
+    _ = _device_unset_current()
 
 
-def _device_unset_current():
+def _device_unset_current() -> bool:
+    """Pop current CUDA context.
+
+    Returns True if context was popped, False it the stack was empty.
+    """
     ctx = handle_return(driver.cuCtxGetCurrent())
     if int(ctx) == 0:
         # no active context, do nothing
-        return
+        return False
     handle_return(driver.cuCtxPopCurrent())
     if hasattr(_device._tls, "devices"):
         del _device._tls.devices
+    return True
 
 
 @pytest.fixture(scope="function")
 def deinit_cuda():
     # TODO: rename this to e.g. deinit_context
     yield
-    _device_unset_current()
+    _ = _device_unset_current()
 
 
 @pytest.fixture(scope="function")
-def deinit_context_function():
-    return _device_unset_current
+def deinit_all_contexts_function():
+    def pop_all_contexts():
+        max_iters = 256
+        for _ in range(max_iters):
+            if _device_unset_current():
+                # context was popped, continue until stack is empty
+                continue
+            # no active context, we are ready
+            break
+        else:
+            raise RuntimeError(f"Number of iterations popping current CUDA contexts, exceded {max_iters}")
+
+    return pop_all_contexts
 
 
 # samples relying on cffi could fail as the modules cannot be imported

cuda_core/tests/test_module.py

@@ -234,7 +234,7 @@ class ExpectedStruct(ctypes.Structure):
 
 
 @skipif_testing_with_compute_sanitizer
-def test_num_args_error_handling(deinit_context_function, cuda12_prerequisite_check):
+def test_num_args_error_handling(deinit_all_contexts_function, cuda12_prerequisite_check):
     if not cuda12_prerequisite_check:
         pytest.skip("Test requires CUDA 12")
     src = "__global__ void foo(int a) { }"
@@ -244,14 +244,9 @@ def test_num_args_error_handling(deinit_context_function, cuda12_prerequisite_ch
         name_expressions=("foo",),
     )
     krn = mod.get_kernel("foo")
-    # Unset current context using function from conftest
-    while True:
-        deinit_context_function()
-        ctx = handle_return(driver.cuCtxGetCurrent())
-        if int(ctx) == 0:
-            # no active context, we are ready
-            break
-    # with no context, cuKernelGetParamInfo would report
+    # empty driver's context stack using function from conftest
+    deinit_all_contexts_function()
+    # with no current context, cuKernelGetParamInfo would report
     # exception which we expect to handle by raising
     with pytest.raises(CUDAError):
         # assignment resolves linter error "B018: useless expression"