Skip to main content
Version: 2.3 (deprecated)

1.30.x

Some highlights:

  • The Pants daemon (pantsd) is now enabled by default for improved performance.
  • Added experimental support for Python dependency inference. See below.
  • Pants now logs when certain steps of your build are done. This improves, in particular, the experience when you have --no-dynamic-ui enabled, such as in CI.

See here for a detailed change log.

Experimental dependency inference feature​

Pants can now read your Python source files to infer the dependencies field for your Python targets, meaning that you can now leave off the dependencies field for most of your BUILD files.

How to activate​

Add dependency_inference = true to your pants.toml, like this:

pants.toml
[GLOBAL]
dependency_inference = true

To test that it is working, find a sample target, then delete the dependencies field from its BUILD file and run ./pants dependencies path/to:target.

Pants will assume that each of your dependencies exposes a module with the same name; for example, the requirement Django>=2.0 would expose the module "django". However, sometimes the module is different, like setuptools exposing pkg_resources.

If you are using a requirements.txt and python_requirements() target, teach Pants about any unusual modules like this:

3rdparty/BUILD
python_library(
  module_mapping={
    "ansicolors": ["colors"],
    "beautifulsoup4": ["bs4"],
    "setuptools": ["pkg_resources"],
  },
)

For inline python_requirement_library targets, configure like this:

3rdparty/BUILD
python_requirement_library(
  name='setuptools',
  requirements=[
    python_requirement('setuptools', modules=['pkg_resources']),
  ],
)

Known limitations​

Does not work with the v1 engine​

You must be solely using the v2 engine for dependency inference to work. Otherwise, when you run v1 tasks, Pants will not know what your dependencies are.

Performance tuning​

We have not yet closely tuned the performance. The performance should be acceptable—and the results will be cached through the Pants daemon (pantsd)—but dependency inference will result in a slowdown compared to explicit targets.

May find cycles in your code​

Dependency inference sometimes reveals cycles between your targets that you did not know about. Run ./pants dependencies --transitive :: to see if you have any.

If you have cycles, you will need to manually fix these cycles by either creating new targets or moving around code

No way to exclude inferred dependencies​

Sometimes, dependency inference may infer something that you do not like. Currently, there is not a way to ignore the inference.

We are working on a feature to ignore a dependency by prefixing the value with !, like !helloworld/project:util.