[import/order] Regression in behavior with parent and sibling

[JohnnyDevNull]

Since Version 2.27.0 we've recognized a change in over 500 files with the ordering rule:

        "import/order": [
          "error",
          {
            "groups": ["builtin", "external", "parent", "sibling", "index"],
            "newlines-between": "always",
            "alphabetize": {
              "order": "asc",
              "caseInsensitive": true
            }
          }
        ],

For example:

@my-local-package/mysub-path
@my-local-package/mysub/path

They are now handled different than before and leads to hundres of linting errors where no before.

[JohnnyDevNull]

This eventually relates to #2683 and #2682

[ljharb]

cc @Pearce-Ropion

[Pearce-Ropion]

According to the 347d78b change, this is not a bug. The change was made to specifically address this specific use case (as far as I understand it)

[ljharb]

@JohnnyDevNull how many of those changes are fixable with the autofixer?

[fvsch]

We use a similar configuration to @JohnnyDevNull in a large codebase and the import-order rule considered this order valid in 2.26:

import abc from '../../abc'
import def from '../def'
import ghi from './ghi'
import jkl from './jkl'

But in 2.27 it's forcing the following order, which doesn't make sense to us:

import ghi from './ghi'
import jkl from './jkl'
import abc from '../../abc'
import def from '../def'

It looks like our only workaround right now would be to change our config to put "parent" and "sibling" into different groups, forcing adding a blank line between parent and sibling imports.

This regression does seem to come from this change:

According to the 347d78b change, this is not a bug. The change was made to specifically address this specific use case (as far as I understand it)

Looking at #2396, it looks like the intent was to sort forward slashes (/) before hyphens (-), so that alphabetical order would be:

something
something/subpath
something-else

But the change also triggered sorting the / character before dots (.), resulting in the issues reported here for parent and sibling imports in the same alphabetized group.

So the importsSorter function should probably be expanded on a bit to sort dots before slashes, hopefully fixing this while still passing the tests from #2396.

[tjx666]

I had lock the plugin version to 2.26.o to fix this:

[Pearce-Ropion]

Quick progress update, I have a fix for this working locally. I'm working on adding a few more unit test cases to maintain code coverage. I should have a PR up by the end of the week.

[notaphplover]

According to the 347d78b change, this is not a bug. The change was made to specifically address this specific use case (as far as I understand it)

I would hazard to say it's a bug since the expected behavior, as described in the docs is not the actual one.

[jefflage]

Quick progress update, I have a fix for this working locally. I'm working on adding a few more unit test cases to maintain code coverage. I should have a PR up by the end of the week.

@Pearce-Ropion How's this coming along?