Passing float16 ndarray from Python to C++ with pybind11 results in undefined symbol error

[Zening-Li]

I am attempting to pass a float16 ndarray from Python to C++ using pybind11. While the compilation succeeds, running the Python code results in an error regarding an undefined symbol. Below are the relevant code snippets and environment details.

Error Message:

check_dtype.cpython-310-x86_64-linux-gnu.so: undefined symbol: _ZNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEE15_M_replace_coldEPcmPKcmm

C++ Code:

#include <iostream>
#include <stdfloat>
#include <pybind11/pybind11.h>
#include <pybind11/numpy.h>

namespace py = pybind11;

void check_array(py::array_t<std::float16_t> arr) {
    auto r = arr.unchecked<1>();  // Read-only access to NumPy array
    for (ssize_t i = 0; i < r.shape(0); ++i) {
        std::cout << "Element " << i << " has a size of " << sizeof(r(i)) << " bytes." << std::endl;
    }
}

PYBIND11_MODULE(check_dtype, m) {
    m.def("check_array", &check_array, "A function to check the byte size of array elements.");
}

Python Code:

import numpy as np
import check_dtype

a = np.array([1.0, 2.0, 3.0], dtype=np.float16)
check_dtype.check_array(a)

Environment:

• G++ version: 13.3.0
• Python version: 3.10.15
• pybind11 version: 2.13.6
• OS: Ubuntu 22.04

Has anyone encountered this issue before or has insights on how to resolve the undefined symbol error? Any help would be greatly appreciated!