Enabling Python support

How to enable Pants's bundled Python backend package.

---

Example Python repository

See here for examples of Pants's Python functionality.

Enable the Python backend like this:

pants.toml

[GLOBAL]
...
backend_packages = [
  'pants.backend.python'
]

You should now see some new goals available:

Shell

$ ./pants help goals
...

package       Create a distributable package.

repl          Open a REPL with the specified code loadable.

...

run           Runs a binary target.
...

test          Run tests.

Have content in your __init__.py files?

Pants automatically uses all relevant __init__.py files, even if dependency inference does not include the files and you don't add it to the dependencies fields of your targets.

This works if you have empty __init__.py files, like most Python projects do; but if you have actual code in your __init__.py files, you should turn on this option in your pants.toml:

[python-infer]
inits = true

This option will cause Pants to infer "proper" dependencies on any ancestor __init__.py file. If you run ./pants dependencies project/util/foo.py, you should see project/__init__.py and project/util/__init__.py show up. This will ensure that any of the dependencies of your __init__.py files are included.

Using Python 3.8's := walrus operator? Tweak the ./pants script

For dependency inference to work properly when using Python 3.8+ syntax, you need to tweak the ./pants script to run the tool with Python 3.8, rather than 3.6 or 3.7. Change the function set_supported_python_versions to only use 3.8: https://github.com/pantsbuild/setup/blob/2f079cbe4fc6a1d9d87decba51f19d7689aee69e/pants#L135-L141. If you do not have that function in your ./pants script, you likely need to update the script by copying the file from the above GitHub link.

This will be fixed in 2.2 so that Pants can always parse your code, regardless of what interpreter Pants is run with.

macOS users: you may need to change interpreter search paths

By default, Pants will look at both your $PATH and—if you use Pyenv—your $(pyenv root)/versions folder when discovering Python interpreters. Your $PATH likely includes the system Pythons at /usr/bin/python and /usr/bin/python3, which are known to have many issues like failing to install some dependencies.

Pants will prefer new Python versions, like 3.6.10 over 3.6.3. Because macOS system Pythons are usually very old, they will usually be ignored.

However, if you run into issues, you can set the interpreter_search_paths option in the [python-setup] scope:

[python-setup]
interpreter_search_paths = [
    # This will use all interpreters in `$(pyenv root)/versions`.
    "<PYENV>",
     # Brew usually installs Python here.
    "/usr/local/bin",
]

See here for more information.