Clang 15 compilation issues on macOS

[jakirkham]

When using Clang 15 on macOS, we saw the following issue ( conda-forge/pp-sketchlib-feedstock#40 (comment) ), which is inlined below:

compiler line:

2023-06-08T09:23:51.9438150Z   [ 93%] Building CXX object CMakeFiles/pp_sketchlib.dir/src/random/random_match.cpp.o
2023-06-08T09:23:51.9544530Z   /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_build_env/bin/x86_64-apple-darwin13.4.0-clang++ -DPYTHON_EXT -Dpp_sketchlib_EXPORTS -I/Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_build_env/include -I/Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/work/vendor/highfive/include -I/Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/work/src -isystem /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeh/lib/python3.10/site-packages/pybind11/include -isystem /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeh/include/python3.10 -isystem /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeh/include/eigen3 -march=core2 -mtune=haswell -mssse3 -ftree-vectorize -fPIC -fPIE -fstack-protector-strong -O2 -pipe -stdlib=libc++ -fvisibility-inlines-hidden -fmessage-length=0 -isystem /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeh/include -fdebug-prefix-map=/Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/work=/usr/local/src/conda/pp-sketchlib-2.1.1 -fdebug-prefix-map=/Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_h_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeh=/usr/local/src/conda-prefix -DVERSION_INFO=\"2.1.1\" -Wall -DEIGEN_USE_BLAS -O3 -DNDEBUG -ffast-math -funroll-loops -m64 -std=gnu++14 -flto=thin -isysroot /Applications/Xcode_13.2.1.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.9.sdk -mmacosx-version-min=10.9 -fPIC -fvisibility=hidden -fopenmp=libomp -MD -MT CMakeFiles/pp_sketchlib.dir/src/random/random_match.cpp.o -MF CMakeFiles/pp_sketchlib.dir/src/random/random_match.cpp.o.d -o CMakeFiles/pp_sketchlib.dir/src/random/random_match.cpp.o -c /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/work/src/random/random_match.cpp

error:

2023-06-08T09:24:01.3657440Z   In file included from /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/work/src/random/random_match.cpp:12:
2023-06-08T09:24:01.3662070Z   In file included from /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_build_env/bin/../include/c++/v1/random:1689:
2023-06-08T09:24:01.3666480Z   /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_build_env/bin/../include/c++/v1/__random/discrete_distribution.h:216:5: error: static assertion failed due to requirement '__libcpp_random_is_valid_urng<Xoshiro, void>::value':
2023-06-08T09:24:01.3668000Z       static_assert(__libcpp_random_is_valid_urng<_URNG>::value, "");
2023-06-08T09:24:01.3669030Z       ^             ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2023-06-08T09:24:01.3670680Z   /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_build_env/bin/../include/c++/v1/__random/discrete_distribution.h:117:17: note: in instantiation of function template specialization 'std::discrete_distribution<>::operator()<Xoshiro>' requested here
2023-06-08T09:24:01.3672060Z           {return (*this)(__g, __p_);}
2023-06-08T09:24:01.3672980Z                   ^
2023-06-08T09:24:01.3674510Z   /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/work/src/random/random_match.cpp:414:43: note: in instantiation of function template specialization 'std::discrete_distribution<>::operator()<Xoshiro>' requested here
2023-06-08T09:24:01.3675780Z         random_seq_buffer << RCMAP[base_dist(generator)];
2023-06-08T09:24:01.3677300Z                                             ^
2023-06-08T09:24:01.3678550Z   In file included from /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/work/src/random/random_match.cpp:12:
2023-06-08T09:24:01.3679990Z   In file included from /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_build_env/bin/../include/c++/v1/random:1682:
2023-06-08T09:24:01.3681310Z   In file included from /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_build_env/bin/../include/c++/v1/__random/bernoulli_distribution.h:14:
2023-06-08T09:24:01.3683720Z   /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_build_env/bin/../include/c++/v1/__random/uniform_real_distribution.h:119:5: error: static assertion failed due to requirement '__libcpp_random_is_valid_urng<Xoshiro, void>::value':
2023-06-08T09:24:01.3685090Z       static_assert(__libcpp_random_is_valid_urng<_URNG>::value, "");
2023-06-08T09:24:01.3686050Z       ^             ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
2023-06-08T09:24:01.3687950Z   /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_build_env/bin/../include/c++/v1/__random/uniform_real_distribution.h:84:17: note: in instantiation of function template specialization 'std::uniform_real_distribution<>::operator()<Xoshiro>' requested here
2023-06-08T09:24:01.3689270Z           {return (*this)(__g, __p_);}
2023-06-08T09:24:01.3690160Z                   ^
2023-06-08T09:24:01.3691750Z   /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_build_env/bin/../include/c++/v1/__random/discrete_distribution.h:219:65: note: in instantiation of function template specialization 'std::uniform_real_distribution<>::operator()<Xoshiro>' requested here
2023-06-08T09:24:01.3693660Z              _VSTD::upper_bound(__p.__p_.begin(), __p.__p_.end(), __gen(__g)) -
2023-06-08T09:24:01.3694590Z                                                                   ^
2023-06-08T09:24:01.3696190Z   /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/_build_env/bin/../include/c++/v1/__random/discrete_distribution.h:117:17: note: in instantiation of function template specialization 'std::discrete_distribution<>::operator()<Xoshiro>' requested here
2023-06-08T09:24:01.3697770Z           {return (*this)(__g, __p_);}
2023-06-08T09:24:01.3698840Z                   ^
2023-06-08T09:24:01.3700190Z   /Users/runner/miniforge3/conda-bld/pp-sketchlib_1686215809509/work/src/random/random_match.cpp:414:43: note: in instantiation of function template specialization 'std::discrete_distribution<>::operator()<Xoshiro>' requested here
2023-06-08T09:24:01.3701530Z         random_seq_buffer << RCMAP[base_dist(generator)];
2023-06-08T09:24:01.3702440Z                                             ^

---

Think this is due to libcxx 15.0.0+ (used by clang++ 15.0.0+ on macOS) being stricter about random number generators implementations

The result_type defined by Xoshiro is size_t, which libcxx expect to match the return type of operator(). However Xoshiro::operator() returns uint64_t. Hence this check fails

Maybe Xoshiro::result_type should be uint64_t? Or perhaps another solution to align these could be pursued?

[jakirkham]

cc @johnlees (for vis)

[jakirkham]

Am curious. Has this been looked into?

[johnlees]

Yes sorry I haven't had a chance to fix this, but I will do so this week so we can fix the conda recipe issues!

[jakirkham]

Thanks John! 🙏

Hoping the exploration I did before (and wrote-up above) makes this somewhat easier

[johnlees]

Working on this in #93 but facing a CI issue which I suspect is caused by inconsistent HDF5 versions, will take me a bit longer to resolve as a result

[jakirkham]

Thanks John! 🙏

Commented in the PR: #93 (comment)