Add macros KVIKIO_EXPECT and KVIKIO_FAIL to improve exception handling

[kingcrimsontianyu]

This PR follows cuDF's enhanced exception handling approach and adds the macros KVIKIO_EXPECT and KVIKIO_FAIL to KvikIO.

The benefit is that the file name and line number where the exception is thrown are automatically included in the exception message. This also leads to cleaner code.

This PR does not add the stacktrace capability as in cuDF's exception handling. This feature should be added in the future.

Closes #654

[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.

[kingcrimsontianyu]

/ok to test

[kingcrimsontianyu]

/ok to test

[kingcrimsontianyu]

/ok to test

[kingcrimsontianyu]

/ok to test

[kingcrimsontianyu]

/ok to test

[kingcrimsontianyu]

Doxygen documentation on macros has been added in this PR:

[kingcrimsontianyu]

@madsbk @bdice I'm wondering if you have any thought about this CI failure: https://github.com/rapidsai/kvikio/actions/runs/13708284889/job/38339415663?pr=653

Additional observations

KvikIO_CUDA_SUPPORT=ON (default)

The tests in test_cufile_driver.py all passed on my PC with the following setup:

• build-all -DBUILD_TESTS=ON -j 16
• dev container CUDA 11.8-pip or 12.8-pip

KvikIO_CUDA_SUPPORT=OFF

However, if I add -DKvikIO_CUDA_SUPPORT=OFF to the build-all command, the build will fail under CUDA 11.8-pip or 12.8-pip due to an nvcomp issue:

[1/27] Transpiling cufile_driver.pyx to cufile_driver.cxx
      [2/27] Transpiling future.pyx to future.cxx
      [3/27] Transpiling defaults.pyx to defaults.cxx
      [4/27] Transpiling libnvcomp_ll.pyx to libnvcomp_ll.cxx
      [5/27] Building CXX object kvikio/_lib/CMakeFiles/libnvcomp_ll.dir/libnvcomp_ll.cxx.o
      FAILED: kvikio/_lib/CMakeFiles/libnvcomp_ll.dir/libnvcomp_ll.cxx.o
      /usr/bin/sccache /usr/bin/g++ -DBS_THREAD_POOL_ENABLE_PAUSE=1 -DKVIKIO_LIBCURL_FOUND -Dlibnvcomp_ll_EXPORTS -isystem /usr/include/python3.10 -isystem /home/coder/kvikio/cpp/include -isystem /home/coder/kvikio/cpp/build/pip/cuda-11.8/release/_deps/bs_thread_pool-src/include -isystem /home/coder/kvikio/python/kvikio/build/pip/cuda-11.8/release/_deps/nvcomp_proprietary_binary-src/include -O3 -DNDEBUG -fPIC -MD -MT kvikio/_lib/CMakeFiles/libnvcomp_ll.dir/libnvcomp_ll.cxx.o -MF kvikio/_lib/CMakeFiles/libnvcomp_ll.dir/libnvcomp_ll.cxx.o.d -o kvikio/_lib/CMakeFiles/libnvcomp_ll.dir/libnvcomp_ll.cxx.o -c /home/coder/kvikio/python/kvikio/build/pip/cuda-11.8/release/kvikio/_lib/libnvcomp_ll.cxx
      /home/coder/kvikio/python/kvikio/build/pip/cuda-11.8/release/kvikio/_lib/libnvcomp_ll.cxx:1053:10: fatal error: cuda_runtime.h: No such file or directory
       1053 | #include "cuda_runtime.h"
            |          ^~~~~~~~~~~~~~~~
      compilation terminated.
      [6/27] Transpiling file_handle.pyx to file_handle.cxx
      [7/27] Transpiling arr.pyx to arr.cxx
      [8/27] Transpiling libnvcomp.pyx to libnvcomp.cxx
      [9/27] Transpiling buffer.pyx to buffer.cxx
      [10/27] Transpiling remote_handle.pyx to remote_handle.cxx
      [11/27] Building CXX object kvikio/_lib/CMakeFiles/cufile_driver.dir/cufile_driver.cxx.o
      [12/27] Building CXX object kvikio/_lib/CMakeFiles/future.dir/future.cxx.o
      [13/27] Building CXX object kvikio/_lib/CMakeFiles/defaults.dir/defaults.cxx.o

[kingcrimsontianyu]

This fix ensures a successful build with -DKvikIO_CUDA_SUPPORT=OFF option.
Related to the merged PR: #637

[bdice]

Okay, so the cuda_runtime.h issue is all sorted out?

[kingcrimsontianyu]

Sorry for the confusion. I updated my comment above. The cuda_runtime.h issue still exists.

#define CU_MEMHOSTALLOC_PORTABLE here is a fix for the C++-part of the build:

# This is successful now
build-kvikio-cpp  -DKvikIO_CUDA_SUPPORT=OFF -DBUILD_TESTS=ON -j 16

build-all -DKvikIO_CUDA_SUPPORT=OFF -DBUILD_TESTS=ON -j 16 still fails with the cuda_runtime.h issue.

[kingcrimsontianyu]

This return {} and ditto below is to avoid a compile warning (error):

no return statement in function returning...

[madsbk]

Looks great, thanks @kingcrimsontianyu

[kingcrimsontianyu]

The CI issue on CUDA 11.8 is now fixed.

According to test_wheel.sh

# If running CUDA 11.8 on arm64, we skip tests marked "cufile" since
# cuFile didn't support arm until 12.4

So the failed test should have been decorated by the marker @pytest.mark.cufile.

PS: The local build fail with build-all -DKvikIO_CUDA_SUPPORT=OFF -DBUILD_TESTS=ON -j 16 needs further investigation.
PPS: The local build fail also occurs to branch-25.04.

[kingcrimsontianyu]

/merge