Allow configuration to be defined on a file-scoped basis

[Czaki]

It will be nice to allow define banned-api rules per file pattern.

In general, there are API that may be useful in tests but should not be used in the library itself: For example, warnings.catch_warnings because of https://github.com/python/cpython/issues/73858.

On the other hand, I would like to block using numpy.random in tests but allow in the library and/or examples (as it leads to random test failure).

If you think it is useful, I could try to contribute (learning rust).

[MichaReiser]

That makes sense. Have you tried using a hierarchical configuration to achieve your goal? The way this works is that you create a new ruff.toml in the test directory that extends your default configuration. The test-specific configuration can override individual settings.

[Czaki]

Unfortuantelly in this project (napari), test are not stored in a separate directory but in _test modules. And I'm only one of the maintainers. (and this simplifies searching for a test of a given function).

[charliermarsh]

Really wishing we had ESLint-style file overrides, as in:

{
  "rules": {
    "quotes": ["error", "double"]
  },

  "overrides": [
    {
      "files": ["bin/*.js", "lib/*.js"],
      "excludedFiles": "*.test.js",
      "rules": {
        "quotes": ["error", "single"]
      }
    }
  ]
}

[Czaki]

cibuildwheel project use toml tables to implement overwrite rules inside pyproject.toml https://github.com/pypa/cibuildwheel/ https://cibuildwheel.readthedocs.io/en/stable/options/#overrides

Maybe it could be an inspiration?

[MichaReiser]

Really wishing we had ESLint-style file overrides, as in:

{
  "rules": {
    "quotes": ["error", "double"]
  },

  "overrides": [
    {
      "files": ["bin/*.js", "lib/*.js"],
      "excludedFiles": "*.test.js",
      "rules": {
        "quotes": ["error", "single"]
      }
    }
  ]
}

I've mixed feelings about overrides. It is powerful but can become confusing if you mix it with extend. E.g. what happens if you have:

{
   "rules": {
     "quotes": ["error", "double"]
   },
 
   "overrides": [
     {
       "files": ["bin/*.js", "lib/*.js"],
       "excludedFiles": "*.test.js",
       "rules": {
         "quotes": ["error", "single"]
       }
     }
   ]
}

and

{
	"extend": "config-above.json",
   "rules": {
     "quotes": ["error", "double"]
   }
}

What does now bind stronger, the parent configuration because it's in the overrides section or the extend? I think we would need to change our configuration more drastically to make "overrides" feel intuitive, similar to what eslint does with the new flat configuration

[charliermarsh]

(Making this issue more general -- it would be nice to support something like ESLint's flat configuration, in which you can define configuration that only applies to a subset of files.)

[DetachHead]

i'm fine with hierarchical configuration, but the only problem is that i wish it would inherit the flake8-tidy-imports.banned-api options from parent directories:

# ./pyproject.toml
[tool.ruff.flake8-tidy-imports.banned-api]
"foo".msg = "banned project-wide"

# ./tests/ruff.toml
extend = "../pyproject.toml"

[flake8-tidy-imports.banned-api]
"bar".msg = "only banned in tests"

# tests/test_foo.py
import foo # no error, because the `banned-api` the `tests` directory doesn't inherit from the global one

[BT-rmartin]

In my case #7890 we want to configure isort on file-scope basis, and in our particular need the hierarchical solution is not feasible, as we want it for every init.py file. Is there any progress on this topic @charliermarsh ?

[BT-rmartin]

@charliermarsh Is there any progress on configure isort on file-scope basis?

[jorenham]

see https://github.com/astral-sh/ruff/issues/16057 for another unique use-case of this

[irdkwmnsb]

I'm not sure if that's the feature I want, but my usecase is as follows:
Coming from a big legacy project I have multiple files with long variable names where it's ok to have wider lines.
And the formatter wants to split those lines into multiple lines, which ends up being not so pretty.

One specific example of this is such code:

if (
    (something1) is not None
    and (something2) is not None
    and (something3) is not None
):

With the default value of 88 the formatter will split up those lines to have and and is not None to be sometimes on one line, and sometimes on multiple, since the length of something1 expression can be different. If I set the line-length parameter to some large number the code looks neat and every is not None statement is placed on a separate line. But I don't want such large line-length on the whole project, so having a feature to be able to change settings on a per-file basis would be nice.

[MichaReiser]

@irdkwmnsb you can already achieve that if all legacy files have a common ancestor directory by adding a ruff.toml in that ancestor directory.

You need this feature if the files are side by side to non-legacy files.

[arogozhnikov]

I have two use-cases for this feature, line-length in notebooks vs plain python files (they are side by side), and an additional 'type' of python file, which also should have a different line-length (also side by side with rest of codebase)