[thrust] cuda/std/detail/libcxx/include/tuple error: call of overloaded swap() is ambiguous

[cjmayo]

not for merging - to demonstrate compiling with CCCL.

Since 0577aab compiling with recent CCCL fails, with GCC 13:

/home/runner/work/manifold/manifold/src/polygon/src/polygon.cpp:844:16:   required from here
/usr/include/cuda/std/detail/libcxx/include/tuple:208:7: error: call of overloaded ‘swap(__gnu_cxx::__normal_iterator<{anonymous}::EarClip::Vert*, std::vector<{anonymous}::EarClip::Vert> >&, __gnu_cxx::__normal_iterator<{anonymous}::EarClip::Vert*, std::vector<{anonymous}::EarClip::Vert> >&)’ is ambiguous
  208 |   swap(__x.get(), __y.get());
      |   ~~~~^~~~~~~~~~~~~~~~~~~~~~
In file included from /usr/include/cuda/std/__functional/hash.h:36,
                 from /usr/include/cuda/std/detail/libcxx/include/utility:24,
                 from /usr/include/cuda/std/utility:26,
                 from /usr/include/thrust/pair.h:33,
                 from /usr/include/thrust/binary_search.h:33:
/usr/include/cuda/std/__utility/swap.h:35:81: note: candidate: ‘constexpr cuda::std::__4::__swap_result_t<_Tp> cuda::std::__4::swap(_Tp&, _Tp&) [with _Tp = __gnu_cxx::__normal_iterator<{anonymous}::EarClip::Vert*, std::vector<{anonymous}::EarClip::Vert> >; __swap_result_t<_Tp> = void]’
   35 | inline _LIBCUDACXX_INLINE_VISIBILITY _CCCL_CONSTEXPR_CXX14 __swap_result_t<_Tp> swap(_Tp& __x, _Tp& __y) noexcept(
      |                                                                                 ^~~~
In file included from /usr/include/c++/13/bits/stl_function.h:60,
                 from /usr/include/c++/13/functional:49,
                 from /home/runner/work/manifold/manifold/src/polygon/include/polygon.h:16,
                 from /home/runner/work/manifold/manifold/src/polygon/src/polygon.cpp:15:
/usr/include/c++/13/bits/move.h:189:5: note: candidate: ‘std::_Require<std::__not_<std::__is_tuple_like<_Tp> >, std::is_move_constructible<_Tp>, std::is_move_assignable<_Tp> > std::swap(_Tp&, _Tp&) [with _Tp = __gnu_cxx::__normal_iterator<{anonymous}::EarClip::Vert*, vector<{anonymous}::EarClip::Vert> >; _Require<__not_<__is_tuple_like<_Tp> >, is_move_constructible<_Tp>, is_move_assignable<_Tp> > = void]’
  189 |     swap(_Tp& __a, _Tp& __b)
      |     ^~~~
make[2]: *** [src/polygon/CMakeFiles/polygon.dir/build.make:79: src/polygon/CMakeFiles/polygon.dir/src/polygon.cpp.o] Error 1

I guess that is a result of Replace thrust::tuple implementation with cuda::std::tuple NVIDIA/cccl@61c1fd0
which has caused other problems NVIDIA/cccl@78a61a2
so their problem, but a bit outside my knowledge to understand what we would want.

[elalish]

Thanks, but @pca006132 and I have decided to replace Thrust with PSTL instead of CCCL: #823. Looking to minimize dependencies and maximize where everything compiles. Help would be great though!

[cjmayo]

Getting rid of this dependency is even better.

For the record, sticking with CCCL 2.2.0 is a workaround in the meantime.