flake8-mypy 17.8.0

List of warnings

flake8-mypy reserves T4 for all current and future codes, T being the natural letter for typing-related errors. There are other plugins greedily reserving the entire letter T. To this I say: ¯\_(ツ)_/¯.

T400: any typing note.

T484: any typing error (after PEP 484, geddit?).

T498: internal mypy error.

T499: internal mypy traceback, stderr output, or an unmatched line.

I plan to support more fine-grained error codes for specific mypy errors in the future.

Two levels of type checking

mypy shines when given a full program to analyze. You can then use options like --follow-imports or --disallow-untyped-calls to exercise the full transitive closure of your modules, catching errors stemming from bad API usage or incompatible types. That being said, those checks take time, and require access to the entire codebase. For some tools, like an editor with an open file, or a code review tool, achieving this is not trivial. This is where a more limited approach inside a linter comes in.

Flake8 operates on unrelated files, it doesn’t perform full program analysis. In other words, it doesn’t follow imports. This is a curse and a blessing. We cannot find complex problems and the number of warnings we can safely show without risking false positives is lower. In return, we can provide useful warnings with great performance, usable for realtime editor integration.

As it turns out, in this mode of operation, mypy is still able to provide useful information on the annotations within and at least usage of stubbed standard library and third party libraries. However, for best effects, you will want to use separate configuration for mypy’s standalone mode and for usage as a Flake8 plugin.

Configuration

Due to the reasoning above, by default flake8-mypy will operate with options equivalent to the following:

[mypy]
# Specify the target platform details in config, so your developers are
# free to run mypy on Windows, Linux, or macOS and get consistent
# results.
python_version=3.6
platform=linux

# flake8-mypy expects the two following for sensible formatting
show_column_numbers=True
show_error_context=False

# do not follow imports (except for ones found in typeshed)
follow_imports=skip

# since we're ignoring imports, writing .mypy_cache doesn't make any sense
cache_dir=/dev/null

# suppress errors about unsatisfied imports
ignore_missing_imports=True

# allow untyped calls as a consequence of the options above
disallow_untyped_calls=False

# allow returning Any as a consequence of the options above
warn_return_any=False

# treat Optional per PEP 484
strict_optional=True

# ensure all execution paths are returning
warn_no_return=True

# lint-style cleanliness for typing needs to be disabled; returns more errors
# than the full run.
warn_redundant_casts=False
warn_unused_ignores=False

# The following are off by default since they're too noisy.
# Flip them on if you feel adventurous.
disallow_untyped_defs=False
check_untyped_defs=False

If you disagree with the defaults above, you can specify your own mypy configuration by providing the --mypy-config= command-line option to Flake8 (with the .flake8/setup.cfg equivalent being called mypy_config). The value of that option should be a path to a mypy.ini or setup.cfg compatible file. For full configuration syntax, follow mypy documentation.

For the sake of simplicity and readability, the config you provide will fully replace the one listed above. Values left out will be using mypy’s own defaults.

Remember that for the best user experience, your linter integration mode shouldn’t generally display errors that a full run of mypy wouldn’t. This would be confusing.

Note: chaing the follow_imports option might have surprising effects. If the file you’re linting with Flake8 has other files around it, then in “silent” or “normal” mode those files will be used to follow imports. This includes imports from typeshed.

Tests

Just run:

python setup.py test

OMG, this is Python 3 only!

Yes, so is mypy. Relax, you can run Flake8 with all popular plugins as a tool perfectly fine under Python 3.5+ even if you want to analyze Python 2 code. This way you’ll be able to parse all of the new syntax supported on Python 3 but also effectively all the Python 2 syntax at the same time.

By making the code exclusively Python 3.5+, I’m able to focus on the quality of the checks and re-use all the nice features of the new releases (check out pathlib) instead of wasting cycles on Unicode compatibility, etc.

License

MIT

Change Log

Authors

Glued together by Łukasz Langa.