Write faster C extensions for Python with Cython

Write faster C extensions for Python with Cython

Learn more about solving common Python problems in our series covering seven PyPI libraries.

Hand drawing out the word "code"

Get the newsletter

Join the 85,000 open source advocates who receive our giveaway alerts and article roundups.

Python is one of the most popular programming languages in use today—and for good reasons: it's open source, it has a wide range of uses (such as web programming, business applications, games, scientific programming, and much more), and it has a vibrant and dedicated community supporting it. This community is the reason we have such a large, diverse range of software packages available in the Python Package Index (PyPI) to extend and improve Python and solve the inevitable glitches that crop up.

In this series, we'll look at seven PyPI libraries that can help you solve common Python problems. First up: Cython, a language that simplifies writing C extensions for Python.


Python is fun to use, but sometimes, programs written in it can be slow. All the runtime dynamic dispatching comes with a steep price: sometimes it's up to 10-times slower than equivalent code written in a systems language like C or Rust.

Moving pieces of code to a completely new language can have a big cost in both effort and reliability: All that manual rewrite work will inevitably introduce bugs. Can we have our cake and eat it too?

To have something to optimize for this exercise, we need something slow. What can be slower than an accidentally exponential implementation of the Fibonacci sequence?

def fib(n):
    if n < 2:
        return 1
    return fib(n-1) + fib(n-2)

Since a call to fib results in two calls, this beautifully inefficient algorithm takes a long time to execute. For example, on my new laptop, fib(36) takes about 4.5 seconds. These 4.5 seconds will be our baseline as we explore how Python's Cython extension can help.

The proper way to use Cython is to integrate it into setup.py. However, a quick and easy way to try things out is with pyximport. Let's put the fib code above in fib.pyx and run it using Cython.

>>> import pyximport; pyximport.install()
>>> import fib
>>> fib.fib(36)

Just using Cython with no code changes reduced the time the algorithm takes on my laptop to around 2.5 seconds. That's a reduction of almost 50% runtime with almost no effort; certainly, a scrumptious cake to eat and have!

Putting in a little more effort, we can make things even faster.

cpdef int fib(int n):
    if n < 2:
        return 1
    return fib(n - 1) + fib(n - 2)

We moved the code in fib to a function defined with cpdef and added a couple of type annotations: it takes an integer and returns an integer.

This makes it much faster—around 0.05 seconds. It's so fast that I may start suspecting my measurement methods contain noise: previously, this noise was lost in the signal.

So, the next time some of your Python code spends too long on the CPU, maybe spinning up some fans in the process, why not see if Cython can fix things?

In the next article in this series, we'll look at Black, a project that automatically corrects format errors in your code.

OpenStack source code (Python) in VIM

Learn more about solving common Python problems in our series covering seven PyPI libraries.


About the author

Moshe sitting down, head slightly to the side. His t-shirt has Guardians of the Galaxy silhoutes against a background of sound visualization bars.
Moshe Zadka - Moshe has been involved in the Linux community since 1998, helping in Linux "installation parties". He has been programming Python since 1999, and has contributed to the core Python interpreter. Moshe has been a DevOps/SRE since before those terms existed, caring deeply about software reliability, build reproducibility and other such things. He has worked in companies as small as three people and as big as tens of thousands -- usually some place around where software meets system administration...