I am not sure if the bug comes from xtensor-python or pybind11 but as far as I can tell it is specific to xtensor-python.

Assume that we have the following cpp code:

m.def("foo", [](xt::pyarray<float> array){return 0;});

with m a pybind11 module.

Then, the following python instruction leads to a segfault:

foo("bar")

Note that this does not happen when using pybind11 wrapper for numpy arrays.
Trying to call the following definition:

m.def("foo", [](pybind11::array_t<float> array){return 0;});

with a string argument produces the expected error:

TypeError: foo(): incompatible function arguments. The following argument types are supported:
    1. (arg0: numpy.ndarray[float32]) -> int

Also note that I remember that this was working fine a few months ago (before summer).