Fix #1169: correct NumPy skip conditions for DLPack host writes (#2238)

Writing into a host-accessible array obtained via np.from_dlpack requires
NumPy 2.2.5+, because earlier versions return a read-only array
(numpy GH #28632). Several tests and examples guarded these writes at
NumPy 2.1.0, so on NumPy 2.1.0-2.2.4 they would error instead of skip.

Audit all NumPy version guards in cuda_core and bump the write-into-DLPack
sites to 2.2.5 with a consistent reason matching the existing correct guard
in test_launcher.py. Guards that only read DLPack arrays (e.g.
strided_memory_view_constructors.py) or skip for an unrelated NumPy fix
(test_utils.py, numpy#26501) are left at 2.1.

Also fix examples/memory_ops.py, which used a plain string comparison
(np.__version__ < "2.1.0") that is unreliable for multi-digit versions;
replace it with np.lib.NumpyVersion to match the other examples, and
standardize test_graph_builder.py off a hand-rolled tuple skipif onto
requires_module. Update affected PEP 723 dependency pins to numpy>=2.2.5.

Signed-off-by: Arsalan Shakil <shakil.arsalan@yahoo.com>

Author: ArsalanShakil

cuda_core/examples/graph_update.py

@@ -10,7 +10,7 @@
 # ################################################################################
 
 # /// script
-# dependencies = ["cuda_bindings", "cuda_core", "nvidia-cuda-nvrtc", "numpy>=2.1"]
+# dependencies = ["cuda_bindings", "cuda_core", "nvidia-cuda-nvrtc", "numpy>=2.2.5"]
 # ///
 
 import sys
@@ -44,8 +44,11 @@ def build_increment_graph(device, kernel, target_ptr):
 
 
 def main():
-    if np.lib.NumpyVersion(np.__version__) < "2.1.0":
-        print("This example requires NumPy 2.1.0 or later", file=sys.stderr)
+    # Writing into the pinned host array imported via DLPack requires NumPy
+    # 2.2.5+ (np.from_dlpack returns a read-only array on earlier versions;
+    # see numpy GH #28632).
+    if np.lib.NumpyVersion(np.__version__) < "2.2.5":
+        print("This example requires NumPy 2.2.5 or later", file=sys.stderr)
         sys.exit(1)
 
     device = Device()

cuda_core/examples/memory_ops.py

@@ -5,13 +5,14 @@
 # ################################################################################
 #
 # This example demonstrates memory resources for allocation and management,
-# copying data between device and pinned memory, and DLPack interop. Requires
-# NumPy 2.1.0+.
+# copying data between device and pinned memory, and DLPack interop. Writing
+# into the pinned host array imported via DLPack requires NumPy 2.2.5+ (arrays
+# from np.from_dlpack are read-only on earlier versions; see numpy GH #28632).
 #
 # ################################################################################
 
 # /// script
-# dependencies = ["cuda_bindings", "cuda_core", "nvidia-cuda-nvrtc", "cupy-cuda13x"]
+# dependencies = ["cuda_bindings", "cuda_core", "nvidia-cuda-nvrtc", "cupy-cuda13x", "numpy>=2.2.5"]
 # ///
 
 import sys
@@ -47,8 +48,8 @@
 
 
 def main():
-    if np.__version__ < "2.1.0":
-        print("This example requires NumPy 2.1.0 or later", file=sys.stderr)
+    if np.lib.NumpyVersion(np.__version__) < "2.2.5":
+        print("This example requires NumPy 2.2.5 or later", file=sys.stderr)
         sys.exit(1)
 
     dev = Device()

cuda_core/examples/memory_pool_resources.py

@@ -11,7 +11,7 @@
 # ################################################################################
 
 # /// script
-# dependencies = ["cuda_bindings", "cuda_core", "nvidia-cuda-nvrtc", "numpy>=2.1"]
+# dependencies = ["cuda_bindings", "cuda_core", "nvidia-cuda-nvrtc", "numpy>=2.2.5"]
 # ///
 
 import sys
@@ -43,8 +43,11 @@
 
 
 def main():
-    if np.lib.NumpyVersion(np.__version__) < "2.1.0":
-        print("This example requires NumPy 2.1.0 or later", file=sys.stderr)
+    # Writing into the managed/pinned host arrays imported via DLPack requires
+    # NumPy 2.2.5+ (np.from_dlpack returns a read-only array on earlier
+    # versions; see numpy GH #28632).
+    if np.lib.NumpyVersion(np.__version__) < "2.2.5":
+        print("This example requires NumPy 2.2.5 or later", file=sys.stderr)
         sys.exit(1)
 
     device = Device()

cuda_core/tests/graph/test_device_launch.py

@@ -76,7 +76,7 @@ def _compile_device_launcher_kernel():
     Device().compute_capability.major < 9,
     reason="Device-side graph launch requires Hopper (sm_90+) architecture",
 )
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 def test_device_launch_basic(init_cuda):
     """Test basic device-side graph launch functionality.
 
@@ -128,7 +128,7 @@ def test_device_launch_basic(init_cuda):
     Device().compute_capability.major < 9,
     reason="Device-side graph launch requires Hopper (sm_90+) architecture",
 )
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 def test_device_launch_multiple(init_cuda):
     """Test that device-side graph launch can be executed multiple times.
 

cuda_core/tests/graph/test_graph_builder.py

@@ -121,7 +121,7 @@ def test_graph_is_join_required(init_cuda):
     gb.end_building().complete()
 
 
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 def test_graph_repeat_capture(init_cuda):
     mod = compile_common_kernels()
     add_one = mod.get_kernel("add_one")
@@ -341,7 +341,7 @@ def read_byte(data):
     assert result[0] == 0xAB
 
 
-@pytest.mark.skipif(tuple(int(i) for i in np.__version__.split(".")[:2]) < (2, 1), reason="need numpy 2.1.0+")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 def test_graph_child_graph(init_cuda):
     mod = compile_common_kernels()
     add_one = mod.get_kernel("add_one")

cuda_core/tests/graph/test_graph_builder_conditional.py

@@ -17,7 +17,7 @@
 @pytest.mark.parametrize(
     "condition_value", [True, False, ctypes.c_bool(True), ctypes.c_bool(False), np.bool_(True), np.bool_(False), 1, 0]
 )
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 def test_graph_conditional_if(init_cuda, condition_value):
     mod = compile_conditional_kernels(type(condition_value))
     add_one = mod.get_kernel("add_one")
@@ -81,7 +81,7 @@ def test_graph_conditional_if(init_cuda, condition_value):
 @pytest.mark.parametrize(
     "condition_value", [True, False, ctypes.c_bool(True), ctypes.c_bool(False), np.bool_(True), np.bool_(False), 1, 0]
 )
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 def test_graph_conditional_if_else(init_cuda, condition_value):
     mod = compile_conditional_kernels(type(condition_value))
     add_one = mod.get_kernel("add_one")
@@ -153,7 +153,7 @@ def test_graph_conditional_if_else(init_cuda, condition_value):
 
 
 @pytest.mark.parametrize("condition_value", [0, 1, 2, 3])
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 def test_graph_conditional_switch(init_cuda, condition_value):
     mod = compile_conditional_kernels(type(condition_value))
     add_one = mod.get_kernel("add_one")
@@ -244,7 +244,7 @@ def test_graph_conditional_switch(init_cuda, condition_value):
 
 
 @pytest.mark.parametrize("condition_value", [True, False, 1, 0])
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 def test_graph_conditional_while(init_cuda, condition_value):
     mod = compile_conditional_kernels(type(condition_value))
     add_one = mod.get_kernel("add_one")

cuda_core/tests/graph/test_graph_definition_mutation.py

@@ -139,7 +139,7 @@ def close(self):
         self._buf.close()
 
 
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 class TestMutateYRig:
     """Tests that mutate the Y-shaped graph built by YRig."""
 
@@ -335,7 +335,7 @@ def test_self_edge(init_cuda):
         node.succ.add(node)
 
 
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 def test_convert_linear_to_fan_in(init_cuda):
     """Chain four computations sequentially, then rewire so all pairs run in
     parallel feeding into a reduce node.

cuda_core/tests/graph/test_graph_update.py

@@ -14,7 +14,7 @@
 
 
 @pytest.mark.parametrize("builder", ["GraphBuilder", "GraphDefinition"])
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 def test_graph_update_kernel_args(init_cuda, builder):
     """Update redirects a kernel to write to a different pointer."""
     mod = compile_common_kernels()
@@ -60,7 +60,7 @@ def build(ptr):
     b.close()
 
 
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 def test_graph_update_conditional(init_cuda):
     """Update swaps conditional switch graphs with matching topology."""
     mod = compile_conditional_kernels(int)

cuda_core/tests/test_launcher.py

@@ -320,7 +320,7 @@ def test_launch_invalid_values(init_cuda):
 
 
 @pytest.mark.parametrize("python_type, cpp_type, init_value", PARAMS)
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 def test_launch_scalar_argument(python_type, cpp_type, init_value):
     dev = Device()
     dev.set_current()
@@ -536,7 +536,7 @@ class MyBool(ctypes.c_bool):
     assert holder.ptr != 0
 
 
-@requires_module(np, "2.1")
+@requires_module(np, "2.2.5", reason="need numpy 2.2.5+ (numpy GH #28632)")
 @pytest.mark.parametrize(
     ("scalar_kind", "np_dtype", "cpp_type", "raw_value"),
     [