Add `--per-file-selects` option

[asottile]

In GitLab by @rossmacarthur on Feb 15, 2020, 02:58

Since --per-file-ignores exists I think it makes sense that --per-file-selects exists.

Example: docstrings

For example I have a project that has a package, scripts, as well as tests, and wanted to use flake8-docstrings on just the package part. Being able to specify the following in the config would be useful:

[flake8]
ignore = D
per-file-selects =
    <package-name>/*: D

Example: specific lints from plugins

Additionally, I am unable to select specific lints for the package part, without having to repeat them in the ignores for the rest of the project. Say for example I wanted just 1 lint from three different plugins X001, Y002, Z003 on the package data. It would be nice to use

[flake8]
ignore = X,Y,Z
per-file-selects =
    <package-name>/*: X001,Y002,Z003

But instead I have to do something like

[flake8]
ignore = X,Y,Z
select = X001,Y002,Z003
per-file-ignores =
    ci/*: X001,Y002,Z003
    examples/*: X001,Y002,Z003
    scripts/*: X001,Y002,Z003
    tests/*: X001,Y002,Z003

[asottile]

In GitLab by @asottile on Feb 15, 2020, 06:46

Please demonstrate a use case, I don't think this is necessary

[asottile]

In GitLab by @rossmacarthur on Feb 15, 2020, 11:37

changed the description

[asottile]

In GitLab by @rossmacarthur on Feb 15, 2020, 11:37

changed the description

[asottile]

In GitLab by @rossmacarthur on Feb 15, 2020, 11:37

@asottile I have updated the issue description.

[asottile]

In GitLab by @rossmacarthur on Feb 15, 2020, 11:44

changed the description

[asottile]

In GitLab by @asottile on Feb 15, 2020, 16:51

your use case, couldn't you use per-file-ignores instead?

[sigmavirus24]

Agree that this isn't necessary. Good call closing this out

[MarcoGorelli]

Can I suggest another use case?

In pandas, the structure is:

pandas
├── api
├── arrays
├── compat
├── _config
├── conftest.py
├── core
├── errors
├── __init__.py
├── io
├── _libs
├── plotting
├── _testing
├── testing.py
├── tests
├── tseries
├── _typing.py
├── util
└── _version.py

If I want to add a linting check only for tests, then I have to repeat per-file-ignores 17 times, as opposed to just using per-file-selects once.

@asottile Would this use-case be enough for you to reconsider?

[asottile]

I don't think you need/want per-file-selects in that case -- you can just ban it in the entire codebase (this already passes minus the one case you've identified)

[MarcoGorelli]

this already passes minus the one case you've identified

Not sure what you mean here, there's several cases outside of tests where it doesn't pass (note that the check in the linked PR has files: ^pandas/tests)

[asottile]

I only see the reference to np.testing in the file that's changed in the pr

[MarcoGorelli]

Yes, because the hook has:

         files: ^pandas/tests/

and so the other cases of when the pattern is found (which don't need changing) aren't caught during the pre-commit job.

If I remove that line, I get:

$ pre-commit run np-testing-array-equal -a
Check for usage of numpy testing or array_equal..........................Failed
- hook id: np-testing-array-equal
- exit code: 1

pandas/core/arrays/categorical.py:2268:            return np.array_equal(self._codes, other._codes)
pandas/core/arrays/datetimelike.py:982:            if not np.array_equal(index.asi8, on_freq.asi8):
pandas/core/arrays/datetimelike.py:989:            # GH#11587 the main way this is reached is if the `np.array_equal`
pandas/core/dtypes/cast.py:2054:    if np.array_equal(arr, casted):
pandas/core/dtypes/missing.py:436:            return np.array_equal(left, right)
pandas/core/dtypes/missing.py:470:    return np.array_equal(left, right)
pandas/core/dtypes/missing.py:478:    return np.array_equal(left.view("i8"), right.view("i8"))
pandas/core/generic.py:1446:        numpy.array_equal : Return True if two arrays have the same shape
pandas/core/indexes/datetimelike.py:139:        return np.array_equal(self.asi8, other.asi8)
pandas/core/indexes/multi.py:3532:            if not np.array_equal(self_mask, other_mask):
pandas/core/indexes/numeric.py:330:            if not np.array_equal(data, subarr):
pandas/plotting/_matplotlib/tools.py:373:        if not np.array_equal(ax1_points, ax2.get_position().get_points()):

Ideally, I would add this check to the pandas flake8 extension, and then only enable it in pandas/tests using --per-files-selects

[PeterJCLaw]

A use-case I'd like to have --per-file-selects for is to have stronger lints in some files than others. In our Django codebase we have a lot of Django apps, each of which has an api.py that exposes the main API of that app to the rest of the codebase. I'd like to be able to lint those files to require e.g: that all the public members have docstrings.

Currently I'd need to exhaustively name all the possible file names which aren't api.py and disable docstring related lints from those files.