Developing a Pants Plugin

This page documents how to develop a Pants plugin, a set of code that defines new Pants functionality. If you develop a new task or target to add to Pants (or to override an existing part of Pants), a plugin gives you a way to register your code with Pants.

Much of Pants' own functionality is organized in plugins; see them in src/python/pants/backend/*.

A plugin registers its functionality with Pants by defining some functions in a register.py file in its top directory. For example, Pants' jvm code registers in src/python/pants/backend/jvm/register.py Pants' backend-loader code assumes your plugin has a register.py file there.

Simple Configuration

If you want to extend Pants without adding any 3rd-party libraries that aren't already referenced by Pants, you can use the following technique using sources stored directly in your repo. All you need to do is name the package where the plugin is defined, and the pythonpath entry to load it from.

In the example below, the stock JvmBinary target will subclassed so that a custom task (not shown) can consume it specifically but disregard regular JvmBinary instances (using isinstance()).

  • Define a home for your plugins. In this example we'll use a top-level plugins/ directory but this is not special.

  • Create an empty plugins/hadoop/targets/__init__.py to define a python package structure for your custom hadoop target type plugin. Also create empty __init__.py files in each directory up to but not including the root directory of your python package layout; in this case, just plugins/hadoop/__init__.py.

  • Create a python module for your custom target type in plugins/hadoop/targets/hadoop_binary.py:

    # plugins/hadoop/targets/hadoop_binary.py
    from pants.backend.jvm.targets.jvm_binary import JvmBinary
    
    class HadoopBinary(JvmBinary):
      pass
    
  • Create plugins/hadoop/register.py to register the elements exposed by your plugin to BUILD file authors:

    # plugins/hadoop/register.py
    
    from pants.build_graph.build_file_aliases import BuildFileAliases
    from hadoop.targets.hadoop_binary import HadoopBinary
    
    def build_file_aliases():
      return BuildFileAliases(
        targets={
          # NB: This allows a HadoopBinary target to be created in a BUILD file using the
          # hadoop_binary alias.
          'hadoop_binary': HadoopBinary,
        },
      )
    
  • In pants.ini, add your new plugin package to the list of backends to load when pants starts. This instructs pants to load a module named hadoop.register.

    [GLOBAL]
    pythonpath: [
        '%(buildroot)s/plugins',
      ]
    backend_packages: [
        'hadoop',
      ]
    

Note that you can also set the PYTHONPATH in your ./pants wrapper script, instead of in pants.ini, if you have other reasons to do so. Either way, pants will look for a register.py file for each backend package you list by prefixing that package with the python path; ie roughly pythonpath + backend package + register.py is tried for each pythonpath prefix until a register.py is found, in this example plugins/hadoop/register.py.

Examples from twitter/commons

For an example of a code repo with plugins to add features to Pants when building in that repo, take a look at twitter/commons, especially its pants-plugins directory.

The repo's pants.ini file has a backend_packages entry listing the plugin packages (packages with register.py files):

[GLOBAL]
pythonpath: [
    "%(buildroot)s/pants-plugins/src/python",
  ]
backend_packages: [
    'twitter.common.pants.jvm.extras',
    'twitter.common.pants.python.commons',
    'pants.contrib.python.checks',
  ]

The ...jvm/extras/register.py file registers a checkstyle goal. To find the code for this task, come back to the pantsbuild/pants repo: Pants defines the Checkstyle task class but doesn't register it. But other Pants workspaces can register it, as twitter/commons illustrates.

Generated by publish_docs from dist/markdown/html/src/docs/howto_plugin.html 2017-02-25T05:12:12.627866