pyproject.toml in pixi#
We support the use of the pyproject.toml as our manifest file in pixi.
This allows the user to keep one file with all configuration.
The pyproject.toml file is a standard for Python projects.
We don't advise to use the pyproject.toml file for anything else than python projects, the pixi.toml is better suited for other types of projects.
Initial setup of the pyproject.toml file#
When you already have a pyproject.toml file in your project, you can run pixi init in a that folder. Pixi will automatically
- Add a
[tool.pixi.project]section to the file, with the platform and channel information required by pixi; - Add the current project as an editable pypi dependency;
- Add some defaults to the
.gitignoreand.gitattributesfiles.
If you do not have an existing pyproject.toml file , you can run pixi init --format pyproject in your project folder. In that case, pixi will create a pyproject.toml manifest from scratch with some sane defaults.
Python dependency#
The pyproject.toml file supports the requires_python field.
Pixi understands that field and automatically adds the version to the dependencies.
This is an example of a pyproject.toml file with the requires_python field, which will be used as the python dependency:
[project]
name = "my_project"
requires-python = ">=3.9"
[tool.pixi.project]
channels = ["conda-forge"]
platforms = ["linux-64", "osx-arm64", "osx-64", "win-64"]
Which is equivalent to:
[project]
name = "my_project"
channels = ["conda-forge"]
platforms = ["linux-64", "osx-arm64", "osx-64", "win-64"]
[dependencies]
python = ">=3.9"
Dependency section#
The pyproject.toml file supports the dependencies field.
Pixi understands that field and automatically adds the dependencies to the project as [pypi-dependencies].
This is an example of a pyproject.toml file with the dependencies field:
[project]
name = "my_project"
requires-python = ">=3.9"
dependencies = [
"numpy",
"pandas",
"matplotlib",
]
[tool.pixi.project]
channels = ["conda-forge"]
platforms = ["linux-64", "osx-arm64", "osx-64", "win-64"]
Which is equivalent to:
[project]
name = "my_project"
channels = ["conda-forge"]
platforms = ["linux-64", "osx-arm64", "osx-64", "win-64"]
[pypi-dependencies]
numpy = "*"
pandas = "*"
matplotlib = "*"
[dependencies]
python = ">=3.9"
You can overwrite these with conda dependencies by adding them to the dependencies field:
[project]
name = "my_project"
requires-python = ">=3.9"
dependencies = [
"numpy",
"pandas",
"matplotlib",
]
[tool.pixi.project]
channels = ["conda-forge"]
platforms = ["linux-64", "osx-arm64", "osx-64", "win-64"]
[tool.pixi.dependencies]
numpy = "*"
pandas = "*"
matplotlib = "*"
This would result in the conda dependencies being installed and the pypi dependencies being ignored. As pixi takes the conda dependencies over the pypi dependencies.
Optional dependencies#
If your python project includes groups of optional dependencies, pixi will automatically interpret them as pixi features of the same name with the associated pypi-dependencies.
You can add them to pixi environments manually, or use pixi init to setup the project, which will create one environment per feature. Self-references to other groups of optional dependencies are also handled.
For instance, imagine you have a project folder with a pyproject.toml file similar to:
[project]
name = "my_project"
dependencies = ["package1"]
[project.optional-dependencies]
test = ["pytest"]
all = ["package2","my_project[test]"]
Running pixi init in that project folder will transform the pyproject.toml file into:
[project]
name = "my_project"
dependencies = ["package1"]
[project.optional-dependencies]
test = ["pytest"]
all = ["package2","my_project[test]"]
[tool.pixi.project]
channels = ["conda-forge"]
platforms = ["linux-64"] # if executed on linux
[tool.pixi.environments]
default = {features = [], solve-group = "default"}
test = {features = ["test"], solve-group = "default"}
all = {features = ["all", "test"], solve-group = "default"}
In this example, three environments will be created by pixi:
- default with 'package1' as pypi dependency
- test with 'package1' and 'pytest' as pypi dependencies
- all with 'package1', 'package2' and 'pytest' as pypi dependencies
All environments will be solved together, as indicated by the common solve-group, and added to the lock file. You can edit the [tool.pixi.environments] section manually to adapt it to your use case (e.g. if you do not need a particular environment).
Example#
As the pyproject.toml file supports the full pixi spec with [tool.pixi] prepended an example would look like this:
[project]
name = "my_project"
requires-python = ">=3.9"
dependencies = [
"numpy",
"pandas",
"matplotlib",
"ruff",
]
[tool.pixi.project]
channels = ["conda-forge"]
platforms = ["linux-64", "osx-arm64", "osx-64", "win-64"]
[tool.pixi.dependencies]
compilers = "*"
cmake = "*"
[tool.pixi.tasks]
start = "python my_project/main.py"
lint = "ruff lint"
[tool.pixi.system-requirements]
cuda = "11.0"
[tool.pixi.feature.test.dependencies]
pytest = "*"
[tool.pixi.feature.test.tasks]
test = "pytest"
[tool.pixi.environments]
test = ["test"]
Build-system section#
The pyproject.toml file normally contains a [build-system] section. Pixi will use this section to build and install the project if it is added as a pypi path dependency.
If the pyproject.toml file does not contain any [build-system] section, pixi will fall back to uv's default, which is equivalent to the below:
[build-system]
requires = ["setuptools >= 40.8.0"]
build-backend = "setuptools.build_meta:__legacy__"
Including a [build-system] section is highly recommended. If you are not sure of the build-backend you want to use, including the [build-system] section below in your pyproject.toml is a good starting point.
pixi init --format pyproject defaults to hatchling.
The advantages of hatchling over setuptools are outlined on its website.