Skip to content

Tutorial: Doing Python development with Pixi#

In this tutorial, we will show you how to create a simple Python project with pixi. We will show some of the features that pixi provides, that are currently not a part of pdm, poetry etc.

Why is this useful?#

Pixi builds upon the conda ecosystem, which allows you to create a Python environment with all the dependencies you need. This is especially useful when you are working with multiple Python interpreters and bindings to C and C++ libraries. For example, GDAL from PyPI does not have binary C dependencies, but the conda package does. On the other hand, some packages are only available through PyPI, which pixi can also install for you. Best of both world, let's give it a go!

pixi.toml and pyproject.toml#

We support two manifest formats: pyproject.toml and pixi.toml. In this tutorial, we will use the pyproject.toml format because it is the most common format for Python projects.

Let's get started#

Let's start out by making a directory and creating a new pyproject.toml file.

pixi init pixi-py --pyproject

This gives you the following pyproject.toml:

[project]
name = "pixi-py"
version = "0.1.0"
description = "Add a short description here"
authors = [{name = "Tim de Jager", email = "tim@prefix.dev"}]
requires-python = ">= 3.11"
dependencies = []

[build-system]
requires = ["setuptools"]
build-backend = "setuptools.build_meta"

[tool.pixi.project]
channels = ["conda-forge"]
platforms = ["osx-arm64"]

[tool.pixi.pypi-dependencies]
pixi-py = { path = ".", editable = true }

[tool.pixi.tasks]

Let's add the Python project to the tree:

cd pixi-py # move into the project directory
mkdir pixi_py
touch pixi_py/__init__.py
cd pixi-py
mkdir pixi_py
type nul > pixi_py\__init__.py

We now have the following directory structure:

.
├── pixi_py
   ├── __init__.py
└── pyproject.toml

We've used a flat-layout here but pixi supports both flat- and src-layouts.

What's in the pyproject.toml?#

Okay, so let's have a look at what's sections have been added and how we can modify the pyproject.toml.

These first entries were added to the pyproject.toml file:

# Main pixi entry
[tool.pixi.project]
channels = ["conda-forge"]
# This is your machine platform by default
platforms = ["osx-arm64"]

The channels and platforms are added to the [tool.pixi.project] section. Channels like conda-forge manage packages similar to PyPI but allow for different packages across languages. The keyword platforms determines what platform the project supports.

The pixi_py package itself is added as an editable dependency. This means that the package is installed in editable mode, so you can make changes to the package and see the changes reflected in the environment, without having to re-install the environment.

# Editable installs
[tool.pixi.pypi-dependencies]
pixi-py = { path = ".", editable = true }

In pixi, unlike other package managers, this is explicitly stated in the pyproject.toml file. The main reason being so that you can choose which environment this package should be included in.

Managing both conda and PyPI dependencies in pixi#

Our projects usually depend on other packages.

$ pixi add black
Added black

This will result in the following addition to the pyproject.toml:

# Dependencies
[tool.pixi.dependencies]
black = ">=24.4.2,<24.5"

But we can also be strict about the version that should be used with pixi add black=24, resulting in

[tool.pixi.dependencies]
black = "24.*"

Now, let's add some optional dependencies:

pixi add --pypi --feature test pytest

Which results in the following fields added to the pyproject.toml:

[project.optional-dependencies]
test = ["pytest"]

After we have added the optional dependencies to the pyproject.toml, pixi automatically creates a feature, which can contain a collection of dependencies, tasks, channels, and more.

Sometimes there are packages that aren't available on conda channels but are published on PyPI. We can add these as well, which pixi will solve together with the default dependencies.

$ pixi add black --pypi
Added black
Added these as pypi-dependencies.

which results in the addition to the dependencies key in the pyproject.toml

dependencies = ["black"]

When using the pypi-dependencies you can make use of the optional-dependencies that other packages make available. For example, black makes the cli dependencies option, which can be added with the --pypi keyword:

$ pixi add black[cli] --pypi
Added black[cli]
Added these as pypi-dependencies.

which updates the dependencies entry to

dependencies = ["black[cli]"]
Optional dependencies in pixi.toml

This tutorial focuses on the use of the pyproject.toml, but in case you're curious, the pixi.toml would contain the following entry after the installation of a PyPI package including an optional dependency:

[pypi-dependencies]
black = { version = "*", extras = ["cli"] }

Installation: pixi install#

Now let's install the project with pixi install:

$ pixi install
 Project in /path/to/pixi-py is ready to use!

We now have a new directory called .pixi in the project root. This directory contains the environment that was created when we ran pixi install. The environment is a conda environment that contains the dependencies that we specified in the pyproject.toml file. We can also install the test environment with pixi install -e test. We can use these environments for executing code.

We also have a new file called pixi.lock in the project root. This file contains the exact versions of the dependencies that were installed in the environment across platforms.

What's in the environment?#

Using pixi list, you can see what's in the environment, this is essentially a nicer view on the lock file:

$ pixi list
Package          Version       Build               Size       Kind   Source
bzip2            1.0.8         h93a5062_5          119.5 KiB  conda  bzip2-1.0.8-h93a5062_5.conda
black            24.4.2                            3.8 MiB    pypi   black-24.4.2-cp312-cp312-win_amd64.http.whl
ca-certificates  2024.2.2      hf0a4a13_0          152.1 KiB  conda  ca-certificates-2024.2.2-hf0a4a13_0.conda
libexpat         2.6.2         hebf3989_0          62.2 KiB   conda  libexpat-2.6.2-hebf3989_0.conda
libffi           3.4.2         h3422bc3_5          38.1 KiB   conda  libffi-3.4.2-h3422bc3_5.tar.bz2
libsqlite        3.45.2        h091b4b1_0          806 KiB    conda  libsqlite-3.45.2-h091b4b1_0.conda
libzlib          1.2.13        h53f4e23_5          47 KiB     conda  libzlib-1.2.13-h53f4e23_5.conda
ncurses          6.4.20240210  h078ce10_0          801 KiB    conda  ncurses-6.4.20240210-h078ce10_0.conda
openssl          3.2.1         h0d3ecfb_1          2.7 MiB    conda  openssl-3.2.1-h0d3ecfb_1.conda
python           3.12.3        h4a7b5fc_0_cpython  12.6 MiB   conda  python-3.12.3-h4a7b5fc_0_cpython.conda
readline         8.2           h92ec313_1          244.5 KiB  conda  readline-8.2-h92ec313_1.conda
tk               8.6.13        h5083fa2_1          3 MiB      conda  tk-8.6.13-h5083fa2_1.conda
tzdata           2024a         h0c530f3_0          117 KiB    conda  tzdata-2024a-h0c530f3_0.conda
pixi-py          0.1.0                                        pypi   . (editable)
xz               5.2.6         h57fd34a_0          230.2 KiB  conda  xz-5.2.6-h57fd34a_0.tar.bz2

Python

The Python interpreter is also installed in the environment. This is because the Python interpreter version is read from the requires-python field in the pyproject.toml file. This is used to determine the Python version to install in the environment. This way, pixi automatically manages/bootstraps the Python interpreter for you, so no more brew, apt or other system install steps.

Here, you can see the different conda and Pypi packages listed. As you can see, the pixi-py package that we are working on is installed in editable mode. Every environment in pixi is isolated but reuses files that are hard-linked from a central cache directory. This means that you can have multiple environments with the same packages but only have the individual files stored once on disk.

We can create the default and test environments based on our own test feature from the optional-dependency:

# Environments
[tool.pixi.environments]
default = { solve-group = "default" }
test = { features = ["test"], solve-group = "default" }
Solve Groups

Solve groups are a way to group dependencies together. This is useful when you have multiple environments that share the same dependencies. For example, maybe pytest is a dependency that influences the dependencies of the default environment. By putting these in the same solve group, you ensure that the versions in test and default are exactly the same.

The default environment is created when you run pixi install. The test environment is created from the optional dependencies in the pyproject.toml file. You can execute commands in this environment with e.g. pixi run -e test python

Getting code to run#

Let's add some code to the pixi-py package. We will add a new function to the pixi_py/__init__.py file:

from rich import print

def hello():
    return "Hello, [bold magenta]World[/bold magenta]!", ":vampire:"

def say_hello():
    print(*hello())

Now add the rich dependency from PyPI using: pixi add --pypi rich.

Let's see if this works by running:

pixi r python -c "import pixi_py; pixi_py.say_hello()"
Hello, World! 🧛
Slow?

This might be slow(2 minutes) the first time because pixi installs the project, but it will be near instant the second time.

Pixi runs the self installed Python interpreter. Then, we are importing the pixi_py package, which is installed in editable mode. The code calls the say_hello function that we just added. And it works! Cool!

Testing this code#

Okay, so let's add a test for this function. Let's add a tests/test_me.py file in the root of the project.

Giving us the following project structure:

.
├── pixi.lock
├── pixi_py
   ├── __init__.py
├── pyproject.toml
└── tests/test_me.py
from pixi_py import hello

def test_pixi_py():
    assert hello() == ("Hello, [bold magenta]World[/bold magenta]!", ":vampire:")

Let's add an easy task for running the tests.

$ pixi task add --feature test test "pytest"
 Added task `test`: pytest .

So pixi has a task system to make it easy to run commands. Similar to npm scripts or something you would specify in a Justfile.

Pixi tasks

Tasks are actually a pretty cool pixi feature that is powerful and runs in a cross-platform shell. You can do caching, dependencies and more. Read more about tasks in the tasks section.

$ pixi r test
 Pixi task (test): pytest .
================================================================================================= test session starts =================================================================================================
platform darwin -- Python 3.12.2, pytest-8.1.1, pluggy-1.4.0
rootdir: /private/tmp/pixi-py
configfile: pyproject.toml
collected 1 item

test_me.py .                                                                                                                                                                                                    [100%]

================================================================================================== 1 passed in 0.00s =================================================================================================

Neat! It seems to be working!

Test vs Default environment#

The interesting thing is if we compare the output of the two environments.

pixi list -e test
# v.s. default environment
pixi list

Is that the test environment has:

package          version       build               size       kind   source
...
pytest           8.1.1                             1.1 mib    pypi   pytest-8.1.1-py3-none-any.whl
...

But the default environment is missing this package. This way, you can finetune your environments to only have the packages that are needed for that environment. E.g. you could also have a dev environment that has pytest and ruff installed, but you could omit these from the prod environment. There is a docker example that shows how to set up a minimal prod environment and copy from there.

Replacing PyPI packages with conda packages#

Last thing, pixi provides the ability for pypi packages to depend on conda packages. Let's confirm this with pixi list:

$ pixi list
Package          Version       Build               Size       Kind   Source
...
pygments         2.17.2                            4.1 MiB    pypi   pygments-2.17.2-py3-none-any.http.whl
...

Let's explicitly add pygments to the pyproject.toml file. Which is a dependency of the rich package.

pixi add pygments

This will add the following to the pyproject.toml file:

[tool.pixi.dependencies]
pygments = ">=2.17.2,<2.18"

We can now see that the pygments package is now installed as a conda package.

$ pixi list
Package          Version       Build               Size       Kind   Source
...
pygments         2.17.2        pyhd8ed1ab_0        840.3 KiB  conda  pygments-2.17.2-pyhd8ed1ab_0.conda

This way, PyPI dependencies and conda dependencies can be mixed and matched to seamlessly interoperate.

$  pixi r python -c "import pixi_py; pixi_py.say_hello()"
Hello, World! 🧛

And it still works!

Conclusion#

In this tutorial, you've seen how easy it is to use a pyproject.toml to manage your pixi dependencies and environments. We have also explored how to use PyPI and conda dependencies seamlessly together in the same project and install optional dependencies to manage Python packages.

Hopefully, this provides a flexible and powerful way to manage your Python projects and a fertile base for further exploration with Pixi.

Thanks for reading! Happy Coding 🚀

Any questions? Feel free to reach out or share this tutorial on X, join our Discord, send us an e-mail or follow our GitHub.