Wheel build for 8.0.0 on macos x86_64 is failing tests

[eriknw]

See: https://github.com/GraphBLAS/python-suitesparse-graphblas/actions/runs/5035542365/jobs/9031130433

Raw log attached for posterity: macos_x86_64_log.txt

Looks like something bad is happening in test_matrix_binfile_read_write.

[eriknw]

I'm not sure if there is anything useful here, but I dumped the JIT config for the failing wheel build: https://github.com/GraphBLAS/python-suitesparse-graphblas/actions/runs/5161271671/jobs/929812306

===================================
  Printing default JIT configurations
  -----------------------------------
  JIT_C_CONTROL on
  JIT_USE_CMAKE False
  JIT_C_COMPILER_NAME /Applications/Xcode_14.2.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/cc
  JIT_C_COMPILER_FLAGS -arch x86_64 -O3 -DNDEBUG -Xclang -fopenmp -I/usr/local/opt/libomp/include  -fPIC  -arch x86_64  -isysroot /Applications/Xcode_14.2.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX13.1.sdk 
  JIT_C_LINKER_FLAGS -L/usr/local/opt/libomp/lib -dynamiclib 
  JIT_C_LIBRARIES  -lm -ldl -lomp
  JIT_C_CMAKE_LIBS m;dl;omp
  JIT_C_PREFACE 
  JIT_ERROR_LOG 
  JIT_CACHE_PATH /Users/runner/.SuiteSparse/GrB8.0.1
  ===================================

[alugowski]

Anecdotally I noticed that, at least for C++, builds on macOS seem to be a lot less tolerant of things like out-of-bounds memory accesses. Even reads. Usually on Linux and Windows you have to write to an entirely different page to trigger a segfault, it seems like surprisingly often a single byte out of range is enough to crash on AppleClang/macOS.

So maybe this is an upstream bug in eWiseMult that is just being caught on this platform. The JIT is new code, after all.