Incremental adoption
How to incrementally add Pants to an existing repository.
Recommended steps
If you have an existing repository, we recommend incrementally adopting to reduce the surface area of change, which reduces risk.
Incremental adoption also allows you to immediately start benefitting from Pants, then deepen adoption at your own pace, instead of postponing benefit until you are ready to make dramatic change all at once.
We would love to help you with adopting Pants. Please reach out through Slack.
1. A basic pants.toml
Follow the Getting Started guide to install Pants and set up an initial pants.toml
. Validate that running ./pants count-loc '**'
works properly. If you want to exclude a specific folder at first, you can use the pants_ignore
option.
Add the relevant backends to [GLOBAL].backend_packages
.
2. Set up formatters/linters with basic BUILD files
Formatters and linters are often the simplest to get working because—for all tools other than Pylint— you do not need to worry about things like dependencies and third-party requirements.
First, run ./pants tailor
to generate BUILD files. This tells Pants which files to operate on, and will allow you to set additional metadata over time like test timeouts and dependencies on resources.
Then, activate the Linters and formatters you'd like to use. Hook up the fmt
and lint
goals to your CI.
3. Set up tests
To get tests working, you will first need to set up source roots and third-party dependencies.
Pants's dependency inference will infer most dependencies for you by looking at your import statements. However, some dependencies cannot be inferred, such as resources.
Try running ./pants test ::
to see if any tests fail. Sometimes, your tests will fail with Pants even if they pass with your normal setup because tests are more isolated than when running Pytest/unittest directly:
- Tests run in a sandbox, meaning they can only access dependencies that Pants knows about. If you have a missing file or missing import, run
./pants dependencies path/to/my_test.py
and./pants dependencies --transitive path/to/my_test.py
to confirm what you are expecting is known by Pants. If not, see Troubleshooting / common issues for reasons dependency inference can fail. - Test files are isolated from each other. If your tests depended on running in a certain order, they may now fail. This requires rewriting your tests to remove the shared global state.
You can port your tests incrementally with the skip_tests
field:
python_tests(
name="tests",
# Skip all tests in this folder.
skip_tests=True,
# Or, use `overrides` to only skip some test files.
overrides={
"dirutil_test.py": {"skip_tests": True},
("osutil_test.py", "strutil.py"): {"skip_tests": True},
},
)
./pants test ::
will only run the relevant tests. You can combine this with ./pants peek
to get a list of test files that should be run with your original test runner:
./pants filter --target-type=python_test :: | \
xargs ./pants peek | \
jq -r '.[] | select(.skip_tests== true) | .["sources"][]'
You may want to speed up your CI by having Pants only run tests for changed files.
4. Set up ./pants package
You can use ./pants package
to package your code into various formats, such as a PEX binary, a wheel, an AWS Lambda, or a zip/tar archive.
We recommend manually verifying that this step is working how you'd like by inspecting the built packages. Alternatively, you can write automated tests that will call the equivalent of ./pants package
for you, and insert the built package into your test environment.
5. Check out writing a plugin
Pants is highly extensible. In fact, all of Pants's core functionality is implemented using the exact same API used by plugins.
Check out Plugins Overview. We'd also love to help in the #plugins channel on Slack.
Some example plugins that users have written:
- Cython support
- Building a Docker image with packages built via
./pants package
- Custom
setup.py
logic to compute theversion
dynamically - Jupyter support
Migrating from other BUILD tools? Set custom BUILD file names
If you're migrating from another system that already uses the name BUILD
, such as Bazel or Please, you have a few ways to avoid conflicts:
First, by default Pants recognizes BUILD.extension
for any extension
as a valid BUILD file. So you can use a name like BUILD.pants
without changing configuration.
Second, you can configure Pants to use a different set of file names entirely:
[GLOBAL]
build_patterns = ["PANTSBUILD", "PANTSBUILD.*"]
[tailor]
build_file_name = "PANTSBUILD"
And finally you can configure Pants to not look for BUILD files in certain locations. This can be helpful, for example, if you use Pants for some languages and another tool for other languages:
[GLOBAL]
build_ignore = ["src/cpp"]