Allow CMake 4

[bdice]

This PR attempts to remove the upper bound on CMake, so we can allow CMake 4.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[bdice]

/ok to test

[bdice]

@alexbarghi-nv Are you familiar with the errors here? I didn't immediately see an issue or PR that sounded related.

https://github.com/rapidsai/cugraph-gnn/actions/runs/17685435951/job/50269677353?pr=307

FAILED tests/pylibwholegraph/test_wholememory_binding.py::test_dlpack
FAILED tests/pylibwholegraph/test_wholememory_tensor.py::test_wholememory_tensor

[alexbarghi-nv]

@alexbarghi-nv Are you familiar with the errors here? I didn't immediately see an issue or PR that sounded related.

https://github.com/rapidsai/cugraph-gnn/actions/runs/17685435951/job/50269677353?pr=307

FAILED tests/pylibwholegraph/test_wholememory_binding.py::test_dlpack
FAILED tests/pylibwholegraph/test_wholememory_tensor.py::test_wholememory_tensor

No, I haven't seen these errors before. Other PRs passed.

[bdice]

Hmm. It seems strange that a CMake change would cause this test failure. Maybe I can rerun CI.

CU call='cuDeviceGetAttribute(&flag, CU_DEVICE_ATTRIBUTE_HANDLE_TYPE_FABRIC_SUPPORTED, currentDev)' at file=/tmp/conda-bld-output/bld/rattler-build_libwholegraph/work/cpp/src/wholememory/communicator.cpp line=505 failed with invalid argument
WholeMemory failure at file=/tmp/conda-bld-output/bld/rattler-build_libwholegraph/work/cpp/src/wholememory/memory_handle.cpp line=1929: WholeMemory FATAL at file=/tmp/conda-bld-output/bld/rattler-build_libwholegraph/work/cpp/src/wholememory/memory_handle.cpp line=525: Create host shared memory from IPC key -299100082 failed, Reason=File exists

[alexbarghi-nv]

@bdice upon further inspection it looks like the system may have run out of shared memory during a test.

[alexbarghi-nv]

Hmm. It seems strange that a CMake change would cause this test failure. Maybe I can rerun CI.

CU call='cuDeviceGetAttribute(&flag, CU_DEVICE_ATTRIBUTE_HANDLE_TYPE_FABRIC_SUPPORTED, currentDev)' at file=/tmp/conda-bld-output/bld/rattler-build_libwholegraph/work/cpp/src/wholememory/communicator.cpp line=505 failed with invalid argument
WholeMemory failure at file=/tmp/conda-bld-output/bld/rattler-build_libwholegraph/work/cpp/src/wholememory/memory_handle.cpp line=1929: WholeMemory FATAL at file=/tmp/conda-bld-output/bld/rattler-build_libwholegraph/work/cpp/src/wholememory/memory_handle.cpp line=525: Create host shared memory from IPC key -299100082 failed, Reason=File exists

I don't think CMake is to blame here. I would rerun.

[bdice]

/merge

[vyasr]

The CMake pin was there because of #267. We should probably go through and address rapidsai/build-planning#194 now that we have newer CMake versions out since that was one of the two sufficient conditions to allow it.