Unhandled Exception on macbook m1.

[weinbe58]

Hello, I'm just trying to make sure I am not doing something really dumb here.

I basically want to have a function that uses std::variant types to handle the dispatching on the c++ backend but as an entry point use numpy arrays. In that sense every single pybind function will start with casting a numpy array to a c++ variant of typed numpy arrays from pybind11. In some cases I do not want to support all types for the variants, in which case I need to throw an error if the incorrect type is passed in. One problem I am having now is that on certain platforms I am getting an unhandled c++ exception when in principle it should throw a c++ exception and pybind11 should dispatch that exception to a python error. I have a minimal example of where this happens and I want to make sure it is not something dumb that I am doing:

in the extension I have the following simple function:

#include <pybind11/pybind11.h>

namespace py = pybind11;

py::array_t<int8_t> test (py::array arr){

  const int type_num = arr.dtype().num();

  switch (type_num) {
    case py::detail::npy_api::NPY_INT8_:
      return py::array_t<int8_t>(arr);
    default:
      throw py::value_error("test");
  }
}


PYBIND11_MODULE(ext, m) {
  m.def("test", &test);
}

In this case ext.test(np.zeros(...,dtype=np.int8)) should be (and is) fine while any other type array should fail. This will give me a zsh: trace trap error when I run this with a bad argument while in principle it should just throw a Python error.

So the weird thing is that this only happens on my macbook m1 while on my other computers or in a linux/windows platform it works just fine.