Integrate MyPy type-checking into your Python Bazel builds.
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.
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
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.
1. Create a file that will specify the version of mypy
to use. You will pass the Bazel label for
this file to the deps()
function in @mypy_integration//repositories:deps.bzl
, which below is named
mypy_integration_deps(...)
:
mypy==0.790
β£οΈ Ensure that your selected MyPy version is compatible with your Python version. Incompatibilities can produce obscure looking errors.
(In the examples/
Bazel workspace this file is specified in tools/typing/
)
2. Next, copy the WORKSPACE
snippet
This can be found in the releases page for the release you use.
Note that by default mypy_integration_deps
will default to passing "python3"
as the interpreter used at install,
but this can be overridden by setting python_interpreter
or python_interpreter_target
(but not both).
3. 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
3b. 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",
],
)
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.
./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.