Skip to main content
Version: 2.15 (deprecated)

Custom `python_artifact()` kwargs

How to add your own logic to python_artifact().

Pants can build Python distributions, such as wheels and sdists, from information you provide in a python_distribution target.

When doing so, and if you don't provide your own setup.py file, Pants generates one and passes it the kwargs provided in the provides=python_artifact(...) field to the setup(...) call (Pants also generates some of the kwargs, such as install_requires and namespace_packages by analyzing your code).

It's fairly common to want to generate more of the kwargs dynamically. For example, you may want to:

  • Reduce boilerplate by not repeating common kwargs across BUILD files.
  • Read from the file system to dynamically determine kwargs, such as the long_description or version.
  • Run processes like git to dynamically determine kwargs like version.

You can write a plugin to add custom kwarg generation logic.

Note: there may only be at most one applicable plugin per target customizing the kwargs for the setup() function.

Example

See here for an example that Pants uses internally for its python_distribution targets. This plugin demonstrates reading from the file system to set the version and long_description kwargs, along with adding hardcoded kwargs.

  1. Set up a subclass of SetupKwargsRequest

Set the class method is_applicable() to determine whether your implementation should be used for the particular python_distribution target. If False, Pants will use the default implementation which simply uses the explicitly provided python_artifact from the BUILD file.

In this example, we will always use our custom implementation:

from pants.backend.python.goals.setup_py import SetupKwargsRequest
from pants.engine.target import Target

class CustomSetupKwargsRequest(SetupKwargsRequest):
    @classmethod
    def is_applicable(cls, _: Target) -> bool:
        return True

This example will only use our plugin implementation for python_distribution targets defined in the folder src/python/project1.

class CustomSetupKwargsRequest(SetupKwargsRequest):
    @classmethod
    def is_applicable(cls, target: Target) -> bool:
        return target.address.spec.startswith("src/python/project1")

Then, register your new SetupKwargsRequest with a UnionRule so that Pants knows your implementation exists:

from pants.engine.rules import collect_rules
from pants.engine.unions import UnionRule

...

def rules():
    return [
      	*collect_rules(),
        UnionRule(SetupKwargsRequest, CustomSetupKwargsRequest),
    ]
Consider defining custom python_distribution target types

If you don't want to always use a single custom implementation, an effective approach could be to create custom python_distribution target types so that your users decide which implementation they want to use in their BUILD files.

For example, a user could do this:

pants_python_distribution(
   name="my-dist",
   dependencies=[...],
   provides=python_artifact(...)
)

pants_contrib_python_distribution(
   name="my-dist",
   dependencies=[...],
   provides=python_artifact(...)
)

To support this workflow, create new target types.

class PantsPythonDistribution(Target):
   alias = "pants_python_distribution"
   core_fields = PythonDistribution.core_fields

class PantsContribPythonDistribution(Target):
   alias = "pants_contrib_python_distribution"
   core_fields = PythonDistribution.core_fields

Then, for each SetupKwargsRequest subclass, check which target type was used:

class PantsSetupKwargsRequest(SetupKwargsRequest):
    @classmethod
    def is_applicable(cls, target: Target) -> bool:
        return isinstance(target, PantsPythonDistribution)
  1. Create a rule with your logic

Your rule should take as a parameter the SetupKwargsRequest from step 1. This type has two fields: target: Target and explicit_kwargs: dict[str, Any]. You can use these fields to get more information on the target you are generating a setup.py for.

Your rule should return SetupKwargs, which takes two arguments: kwargs: dict[str, Any] and address: Address.

For example, this will simply hardcode a kwarg:

from pants.backend.python.goals.setup_py import SetupKwargs
from pants.engine.rules import rule

@rule
async def setup_kwargs_plugin(request: CustomSetupKwargsRequest) -> SetupKwargs:
    return SetupKwargs(
        {**request.explicit_kwargs, "plugin_demo": "hello world"}, address=request.target.address
    )

Update your plugin's register.py to activate this file's rules.

pants-plugins/python_plugins/register.py
from python_plugins import custom_python_artifact

def rules():
   return custom_python_artifact.rules()

Then, run pants package path/to:python_distribution and inspect the generated setup.pyto confirm that your plugin worked correctly.

Often, you will want to read from a file in your project to set kwargs like version or long_description. Use await Get(DigestContents, PathGlobs) to do this (see File system):

from pants.backend.python.goals.setup_py import SetupKwargs
from pants.engine.fs import DigestContents, GlobMatchErrorBehavior, PathGlobs
from pants.engine.rules import rule

@rule
async def setup_kwargs_plugin(request: CustomSetupKwargsRequest) -> SetupKwargs:
    digest_contents = await Get(
        DigestContents,
        PathGlobs(
            ["project/ABOUT.rst"],
            description_of_origin="`python_artifact()` plugin",
            glob_match_error_behavior=GlobMatchErrorBehavior.error,
        ),
    )
    about_page_content = digest_contents[0].content.decode()
    return SetupKwargs(
        {**request.explicit_kwargs, "long_description": "\n".join(about_page_content)},
        address=request.target.address
    )

It can be helpful to allow users to add additional kwargs to their BUILD files for you to consume in your plugin. For example, this plugin adds a custom long_description_path field, which gets popped and replaced by the plugin with a normalized long_description kwarg:

python_distribution(
    name="mydist",
    dependencies=[...],
    provides=python_artifact(
        name="mydist",
        ...
        long_description_path="README.md",
    ),
    generate_setup = True,
    sdist = False,
)
import os.path

from pants.backend.python.goals.setup_py import SetupKwargs
from pants.engine.fs import DigestContents, GlobMatchErrorBehavior, PathGlobs
from pants.engine.rules import rule

@rule
async def setup_kwargs_plugin(request: CustomSetupKwargsRequest) -> SetupKwargs:
    original_kwargs = request.explicit_kwargs.copy()
    long_description_relpath = original_kwargs.pop("long_description_file", None)
    if not long_description_relpath:
        raise ValueError(
            f"The python_distribution target {request.target.address} did not include "
            "`long_description_file` in its python_artifact's kwargs. Our plugin requires this! "
            "Please set to a path relative to the BUILD file, e.g. `ABOUT.md`."
        )

    build_file_path = request.target.address.spec_path
    long_description_path = os.path.join(build_file_path, long_description_relpath)
    digest_contents = await Get(
        DigestContents,
        PathGlobs(
            [long_description_path],
            description_of_origin=f"the 'long_description_file' kwarg in {request.target.address}",
            glob_match_error_behavior=GlobMatchErrorBehavior.error,
        ),
    )
    description_content = digest_contents[0].content.decode()
    return SetupKwargs(
        {**original_kwargs, "long_description": "\n".join(description_content)},
        address=request.target.address
    )

Refer to these guides for additional things you may want to do in your plugin: