Introduction
Python is a great language for prototyping and building applications. However, when it comes to performance, it is not the best choice. Python is an interpreted language and it is not compiled. This means that the code is not optimized for the machine it is running on. This is where C comes in. C is a compiled language and it is much faster than Python. So, if you want to write a Python module that is fast, you can write it in C and compile it. This is called a C extension module. In this article, we will see how to build and distribute a Python C extension module using wheels.
Building a C extension module
Let's start by creating a simple C extension module called maths
. In this, we will create a square
function that takes a number and returns its square.
First, create a directory called maths
and create a file called maths.c
inside it. This is where we will write our C code.
#include <Python.h>
int square(int num) {
return num \* num;
}
static PyObject \*py\_square(PyObject \*self, PyObject \*args) {
int n\_num, result;
if (!PyArg\_ParseTuple(args, "i", &n\_num)) {
return NULL;
}
result = square(n\_num);
return Py\_BuildValue("i", result);
}
static PyMethodDef mathsMethods[] = {
{"square", py\_square, METH\_VARARGS, "Function for calculating square in C"},
{NULL, NULL, 0, NULL}
};
static struct PyModuleDef maths = {
PyModuleDef\_HEAD\_INIT,
"maths",
"Custom maths module",
-1,
mathsMethods
};
PyMODINIT\_FUNC PyInit\_maths(void)
{
return PyModule\_Create(&maths);
}
We need to create a setup.py
file to build our module. This file tells Python how to build our module.
from setuptools import setup, Extension
setup(
name="maths",
version="0.1",
ext\_modules=[Extension("maths", ["maths.c"])]
)
Now, we can build our module by running python setup.py build
. This will create a build
directory with a lib
directory inside it. This lib
directory contains our compiled module. We can import this module in Python and use it.
>>> import maths
>>> maths.square(5)
25
Instead of testing our module by importing it in Python, we can also test it by running python setup.py test
. This will run the tests in the test
directory. We can create a test
directory and create a file called test_maths.py
inside it. This is where we will write our tests.
import unittest
import maths
class TestMaths(unittest.TestCase):
def test\_square(self):
self.assertEqual(maths.square(5), 25)
Distributing a C extension module
Now that we have built our module, we can distribute it. We can distribute it as a source distribution or a binary distribution. A source distribution is a zip file that contains the source code of our module. We can distribute our module as a source distribution by running python setup.py sdist
. This will create a dist
directory with a zip file inside it. This zip file contains our source code.
However, source distribution of C extension modules is not recommended. This is because the user needs to have a C compiler installed on their machine to build the module. Most users just want to pip install
the module and use it. So, we need to distribute our module as a binary distribution.
We can use cibuildwheel
package to build wheels across all platforms. We can install it by running pip install cibuildwheel
.
To build a wheel for a specific platform and a specific architecture, we can run cibuildwheel --platform <platform> --architecture <architecture>
. For example, to build a wheel for Linux x86_64, we can run cibuildwheel --platform linux --architecture x86_64
. This will create a wheelhouse
directory with a wheel file inside it. This wheel file contains our compiled module.
cibuildwheel
runs on most CI servers. With proper workflows, we can easily get wheels for all platforms and architectures. We can then upload these wheels to PyPI and users can easily install these wheels.
Conclusion
In this article, we saw how to build and distribute a Python C extension module using wheels. We saw how to build a C extension module and how to distribute it as a binary distribution. We also saw how to use cibuildwheel
to build wheels across all platforms and architectures.
Top comments (0)