Working with pointer arrays in pybind11

Pybind11 provides a convenient interface for C++ API. This header-only library is lightweight, easy to build, and include in a CMake project. In this post, I'm going to tackle a specific problem of working with pointer arrays in your C++ API.

To get started with Pybind11, check our getting started guide

Getting Started with Pybind11

Get started with Pybind11, a powerful C++ library for creating Python bindings. Easily integrate Python and C++ for high-performance computing

ScicodingJake

The problem of working with pointers arises from the non-bound size constraints of a pointer type.  For example, consider the following function declaration

void process(double *data);

It is impossible for pybind11 to tell whether the input pointer should be a pointer to an array or a scalar.

A simple workaround for this is to implement a wrapper around the Python API call for the process function and declare the data parameter as std::vector, which tells pybind11 to expect a Python array as input.

#include <pybind11/pybind11.h>
#include <pybind11/stl.h>

class Example {
	public:
    	Example() = {}
        void process(double *data, int len) {
            for (int i = 0; i < len; i++) {
            	data[i] *= 2;
            }
        }
};

namespace py = pybind11;

PYBIND11_MODULE(pyexample, m) {
    py::class_<Example>(m, "Example")
        .def(py::init<>())
        .def("process", [](Example &m, std::vector<double> data) {
            m.process(data.data(), data.size());
            return data;
        });
}

Note that our intention is not to make changes to the C++ API, and thus, we pass a pointer to the memory space of std::vector as the first parameter and size as the second. It is left to the user to make sure that sufficient space is allocated.