Skip to content

Latest commit

 

History

History
131 lines (92 loc) · 5.06 KB

README.md

File metadata and controls

131 lines (92 loc) · 5.06 KB
Raw

bazel-mypy-integration

Integrate MyPy type-checking into your Python Bazel builds.

⚠️ This software is now in 'production' use, but still in the PRE-RELEASE PHASE and under active development. Please give it a try, and ⭐️ or watch the repo to follow progress ⚠️

mypy is an incremental type system for Python. Incremental type systems arguably become essential when Python codebases grow to be very large [1, 2, 3, 4]. As Bazel is a build system designed for very large codebases, it makes sense that they should work together to help a Python codebase scale.

With bazel-mypy-integration, type errors are flagged as bazel build errors, so that teams can be sure their type-checking is being respected.

Usage

Aspect-based

It's recommended that you register this integration's Aspect to run everytime you build Python code. To do that you can add the following to your .bazelrc:

build --aspects @mypy_integration//:mypy.bzl%mypy_aspect
build --output_groups=+mypy

But if you want to invoke the integration directly, you can do so with:

bazel build --aspects @mypy_integration//:mypy.bzl%mypy_aspect --output_groups=mypy  //my/python/code/...

If there's a typing error in your Python code, then your build will fail. You'll see something like:

ERROR: /Users/jonathonbelotti/Code/thundergolfer/bazel-mypy-integration/examples/hangman/BUILD:1:1: 
MyPy hangman/hangman_dummy_out failed (Exit 1) hangman_mypy_exe failed: error executing command bazel-out/darwin-fastbuild/bin/hangman/hangman_mypy_exe

Use --sandbox_debug to see verbose messages from the sandbox
hangman/hangman.py:52: error: Syntax error in type annotation
hangman/hangman.py:52: note: Suggestion: Use Tuple[T1, ..., Tn] instead of (T1, ..., Tn)
Found 1 error in 1 file (checked 1 source file)
INFO: Elapsed time: 2.026s, Critical Path: 1.85s
INFO: 1 process: 1 darwin-sandbox.
FAILED: Build did NOT complete successfully

mypy_test rule-based

An alternative to registering the Aspect is to use the mypy_test rule which will run MyPy when the target run with bazel test.

load("@mypy_integration//:mypy.bzl", "mypy_test")

mypy_test(
    name = "foo_mypy_test",
    deps = [
        ":foo", # py_[library, binary, test] target
    ],
)

If there's a typing error in your Python code, then the test will fail. Using --test_output=all will ensure the MyPy error is visible in the console.

Installation

mypy_integration expects the user to provide the mypy dependency. Given not every mypy version is compatible to all Python versions and mypy's transitive dependencies can differ based on the Python version, mypy_integration cannot offer a mypy which satisfies all potential users.

1. Provide mypy to mypy_integration. You can do so in 2 ways:

  • Add to your Bazel command --@mypy_integration//:mypy=<your_target_providing_mypy>
  • Add to your bazelrc build --@mypy_integration//:mypy=<your_target_providing_mypy>

❣️ Ensure that your selected MyPy version is compatible with your Python version. Incompatibilities can produce obscure looking errors.

2. Finally, if using the Bazel Aspect, add the following to your .bazelrc so that MyPy checking is run whenever Python code is built:

build --aspects @mypy_integration//:mypy.bzl%mypy_aspect
build --output_groups=+mypy

2b. If using the Bazel rule, you'll add to a BUILD file something like:

load("@mypy_integration//:mypy.bzl", "mypy_test")

py_binary(
    name = "foo",
    srcs = glob(["foopy"]),
    main = "foo.py",
    python_version = "PY3",
    deps = [],
)

mypy_test(
    name = "foo_mypy",
    deps = [
        ":foo",
    ],
)

Configuration

To support the MyPy configuration file you can add the following to your .bazelrc:

build --@mypy_integration//:mypy_config=//:mypy.ini

where //:mypy.ini is a valid MyPy config file within your projects workspace.

🛠 Development

Testing

./test.sh runs some basic integration tests. Right now, running the integration in the Bazel workspace in examples/ tests a lot more functionality but can't automatically test failure cases.