Skip to main content

Pants 2.8 adds Autoflake & Pyupgrade, Docker publishing, Golang, and Google Cloud Functions

· 6 min read
Eric Arellano

Photo by James Orr / Unsplash

Pants 2.8 adds Autoflake & Pyupgrade, Docker publishing, Golang, PEP 517 publishing, and Google Cloud Functions

We're pleased to announce Pants 2.8.0, the latest release of Pants, the scalable and ergonomic build system.

To update, set pants_version = "2.8.0" in your pants.toml. See upgrade tips, including a new update-build-files goal to automate some of the upgrade.

Autoflake and Pyupgrade​

Pants now supports Autoflake and Pyupgrade, in addition to already supporting Black, isort, Flake8, Bandit, Docformatter, MyPy, Pylint, Yapf, Shfmt, and Shellcheck!

Autoflake removes unused imports automatically, and Pyupgrade modernizes your Python syntax to use new features like f-strings.

Before:

app.py
import os
name = "pantsbuild"
print("Hello %s!" %  name)

After running ./pants fmt app.py with the tools enabled:

name = "pantsbuild"
print(f"Hello {name}!")

Publishing Docker images​

Pants 2.7 added experimental Docker support, which gives you a single command to streamline deploying your Python applications with Docker.

Pants 2.8 expands its Docker support to include publishing images. Now, ./pants publish will first build the Docker image—including copying in any relevant PEX binaries, Python wheels, and files. Pants will then publish the image to the registries you specify.

$ ./pants publish src/docker::
14:41:51.98 [INFO] Built docker image: docker.registry.net:8181/test-example:1.2.5
The push refers to repository [docker.registry.net:8181/test-example]
7c521f914a2d: Layer already exists
...
62a747bf1719: Layer already exists
1.2.5: digest: sha256:961beb6126320d277ad4236f9f12694ad4e077a16f73b896b3292023007a002c size: 2428

✓ docker.registry.net:8181/test-example:1.2.5 published.

Pants 2.8 now also can run your Docker images with ./pants run, making sure that the image is first built with all your binaries and files copied into it.

Google Cloud Functions​

Pants can now create a .zip file understood by Google Cloud Function, allowing you to develop your functions alongside the rest of your projects.

$ ./pants package project/google_cloud_function_example.py
Wrote code bundle to dist/project.zip
  Runtime: python3.8
  Handler: main.handler

This functionality uses the same technology that allows Pants to create AWS Lambda functions.

Experimental Golang support​

Pants now augments Go's (already excellent!) tooling with these unique benefits:

  1. A consistent interface for all languages and tools in your repository.
  2. Integration with Git + advanced project introspection.
  3. Remote caching and execution.
  4. Unlike legacy build systems, these benefits all come with minimal boilerplate.

Check out our announcement post for more, and an example repository to try it out.

PEP-517 packaging support​

As explained in a previous post, Pants can now act as a PEP 517 build frontend.

Pants has long had the ability to generate setup.py files for you, to streamline creating and managing distributions. Pants 2.8 extends this with support for PEP 517, the Python standard for specifying how distributions should be built.

Under this standard, you can specify a "build backend" in a pyproject.toml file, and Pants will install and invoke it appropriately. This means that Pants can now run any distribution build, including building native extensions via setup.py and building Poetry projects.

File-level metadata with overrides​

Since Pants 2.0, Pants has operated at the file level. Unlike legacy build tools, Pants runs tests on individual files, for example, rather than "targets" (usually directories). Likewise, Pants understands metadata like your dependencies at the file level, which gives you fine-grained caching.

All this time, Pants has been dynamically generating one "target" per file, but we hid that from you. Now, we added new targets like python_test and python_source which correspond to individual files. You can manually create those, or let Pants generate the targets for you:

python_tests(
    name="tests",
    # Generates two `python_test` targets.
    sources=["osutil_test.py", "dirutil_test.py"],
)

You can now easily override metadata at the file-level through a new overrides field:

python_tests(
    name="tests",
    overrides={
        "dirutil_test.py": {"timeout": 10},
        ("osutil_test.py", "contextutil_test.py"): {
            "dependencies": [":test_data"]},
        },
)

That is, you get precise, file-level metadata. But without demanding the boilerplate of legacy tools like Bazel.

Impact to existing users​

Improving this modeling required making some other changes:

  • Some targets were renamed, like python_library to python_sources. Run ./pants update-build-files to automatically update your BUILD files. (The old names will still work in Pants 2.8.)
  • conftest.py no longer belongs to the sources of python_tests by default. Instead, it belongs to python_test_utils. Run ./pants tailor to automatically add these new targets.
  • ./pants filter --target-type and ./pants peek both now differentiate between python_tests and python_test targets, for example. Usually you will want to update your scripts to use python_test rather than python_tests and so on.
  • The command line arguments dir: and dir:: now include generated targets for project introspection goals like list and peek.

Other changes​

  • TOML dictionaries work for dict options in pants.toml, e.g.
[tailor]
alias_mapping = { python_sources = "custom_python_sources", python_tests = "custom_python_tests" }
[tailor.alias_mapping]
python_sources = "custom_python_sources"
python_tests = "custom_python_tests"
  • You can set up CLI aliases, like ./pants green :: being an alias for ./pants fmt lint check ::
[cli.alias]
green = "fmt lint check"
  • [coverage-py].fail_under option added to fail if Python test coverage is below a certain number.
  • pex_binary has a script field to allow setting console scripts, which allows running third-party dependencies.
  • pex_binary has a restartable field, which you can set so that Pants will restart the binary when input files have changed. This is particularly useful for auto-reloading Flask and Django servers, for example.
  • Added --tailor-check option as a dry run for automatically adding new targets to your BUILD files.
  • Added [tailor].ignore_paths and [tailor].ignore_adding_targets options to avoid adding targets that are false positives.
  • Added ./pants help tools to list all tools installed by Pants, including their versions.
  • typecheck is deprecated in favor of check, which will run MyPy and compilation for languages like Go and Java.

See the full changelog for more changes.

Upcoming in Pants 2.9​

We are continuing to work on language support: adding more Go features and getting Java/Scala into alpha/beta state.

We also plan to resume the redesign of Python lockfiles started with Pants 2.7's tool lockfiles.

--

Try out our example Python repository, and let us know what you think in Slack!

Thanks​

Thanks to all the contributors to 2.8, including everyone who shared feedback on changes and who tested release candidates! Some shout-outs:

  • Chris Williams for Google Cloud Functions.
  • Andreas Stenius for Docker publishing and CLI aliases.
  • Tom Dyas and Eric Arellano for Go support.
  • Benjy Weinberger for PEP 517 support.
  • Asher Foa for Pyupgrade.
  • Patrick Lawson for Autoflake.
  • Stu Hood for auto-reloading when a PEX binary changes.
  • Nicolas Joseph, Siyan Beverly, Doug Jenkins, Anthony D'Silva, and Ken Howard for continual feedback on Go support + the ideas of --tailor-check and [tailor].ignore_paths.