The future of this fork/plugin

[JounQin]

I'm thinking about maintaining this package as a complete independent plugin from upstream because a lot of issues can not be fixed easily without API changes, see

https://github.com/import-js/eslint-plugin-import/issues/1479

https://github.com/import-js/eslint-plugin-import/issues/2108

https://github.com/import-js/eslint-plugin-import/issues/2111

https://github.com/import-js/eslint-plugin-import/issues/2717

It would affect 3rd party resolvers like https://github.com/import-js/eslint-import-resolver-typescript/pull/248

@SukkaW do you want to collaborate here?

[43081j]

when i was considering forking the same plugin, i came to the conclusion it makes more sense to diverge than staying in sync.

there's a lot of extra logic, architecture, etc that could easily fall away these days IMO.

i'd rather you make it independent so you can be set free on it rather than working within constraints of the original plugin

just my two cents

if you need any help, too, i'd be happy to. i work on a fair amount of things in the eslint ecosystem and was about to do my own fork until i discovered this one!

[JounQin]

Benefits

1. No limitations anymore, I'll rewrite the whole package with TypeScript, #23 can be fixed as an effect, cc @Shinigami92
2. Replace all the legacy codes to modern JavaScript/TypeScript
3. Add features we want

Downsides

1. Less contributors at the first place
2. Maybe difficult to support both plugins, what's the future of https://github.com/import-js/eslint-import-resolver-typescript
3. Per 2, should we support TypeScript as first-class, so basically we combined the two packages, is that worth or needed considering not everyone uses TypeScript and package size difference

[SukkaW]

eslint-plugin-import has been missing a real update for years. We should take this opportunity to drop all Node.js <= 14 support.

[JounQin]

eslint-plugin-import has been missing a real update for years. We should take this opportunity to drop all Node.js <= 14 support.

@SukkaW

Yes, we'll rewrite the whole code base into modern JavaScript/TypeScript features.

Do you want to join to maintain this project?

Next steps:

1. I'll transfer this repository into @un-ts org because it will be changed as a TypeScript based library
2. If you want to join, which level do you want to be involved, org level or repository level
3. I'll start to setup the whole code base in a few days, the progress would just be similar as https://github.com/prettier/pretty-quick/compare/v3.1.3...v4.0.0

[43081j]

a few things i think could be cleaned up or thrown away too:

• throw away memo-parser, feels like an experiment that got dumped in the repo and doesn't really belong here anymore
• eslint-module-utils is a "package" that gets pulled from disk (utils/), seems unnecessary. probably good to just move this stuff into src/utils
• check that those utils are actually needed
• consider dropping the concept of a "resolver"

i don't believe there's any need for this resolver architecture anymore. feels like an over-complication of things, especially splitting them out into independent packages.

i think you should be able to use this plugin and it supports every day resolution out of the box (standard resolution browsers/node-esm use, and node resolution). anything else is an edge case you could support through config, but definitely don't need an extra package for.

if we use enhanced-resolve, that itself is configurable for all those cases. doesn't seem to be a reason to have resolvers anymore, just configure enhanced-resolve differently for those edge cases.

[JounQin]

i think you should be able to use this plugin and it supports every day resolution out of the box (standard resolution browsers/node-esm use, and node resolution). anything else is an edge case you could support through config, but definitely don't need an extra package for.

if we use enhanced-resolve, that itself is configurable for all those cases. doesn't seem to be a reason to have resolvers anymore, just configure enhanced-resolve differently for those edge cases.

And that's why I'm using enhanced-resolve in https://github.com/import-js/eslint-import-resolver-typescript, I'll continue to use that package in the new core. While we still need to support tsconfig features like paths, allowArbitraryExtensions, customConditions and moduleSuffixes, etc.

I mean the new plugin will be a TypeScript-first plugin.

[Shinigami92]

Just want to leave my opinion here. I really like the idea of rewriting everything to TS and enhance the project in a good way.

Right now I don't feel like that I can over much time 🙁 But if you want, you can invite me into the org 🙂 and I can help when I have and find time

Another thing: I dislike the plugin-name "i", because it is not clear what "i" means. Is it possible to e.g. publish as @un-es/eslint-plugin-import or something like that?

[JounQin]

The idea of eslint-plugin-i comes from eslint-plugin-n.

Personally, I don't think we should rename to a new brand considering it's already a little bit well known.

[Shinigami92]

The idea of eslint-plugin-i comes from eslint-plugin-n.

Personally, I don't think we should rename to a new brand considering it's already a little bit well known.

I also hate n, what does "n" stand for? I would need to look this up in the Readme or Repo...

[JounQin]

I also hate n, what does "n" stand for? I would need to look this up in the Readme or Repo...

I don't think anyone would use @un-es/eslint-plugin-import before knowing this is the active more elegant fork, I may prefer to use the non-fork version eslint-plugin-import.

n or i, on another hand, indicates that it's much lightweight somehow. If you know what it does at once, you'll never forget it.

[ljharb]

To clarify, eslint-module-utils is a separate package; the repo is a kind of hybrid monorepo. You’re right about the memo parser, that isn’t used.

A great many users need the webpack resolver, or use the plugin with ecosystems that don’t use standard node resolution, so just be aware of who’s being excluded if you choose to drop the concept of resolvers.

[43081j]

i don't believe we need a resolver architecture/plugin system for that

nothing is stopping us providing configurable resolution using a single built-in resolver. i think it was an over-engineering of a much simpler thing. enhanced-resolve plus some configuration will achieve everything we need.

just my opinion though. sure you believed the opposite when you built it, which is fair enough. each to their own

[ljharb]

I didn’t build it.

[JounQin]

enhanced-resolve is good enough to handle webpack things. eslint-module-utils is needless anymore.

[JounQin]

I've started the migration at #26

[fisker]

Really glad to see someone start fork it, I had really bad experience contribute to the original repo once, and decide never do it again..

I dislike the plugin-name "i".

Same here.

[JounQin]

Progress: I've successfully removed all resolver concepts by using enhanced-resolve directly at f747071 (#26) as @43081j proposed. And eslint . run in this project itself is just working, the next step will be focusing on testing.

[JounQin]

@fisker @Shinigami92

Do you think @eslinter/eslint-plugin-import or @eslint-community/eslint-plugin-import would be accepted?

I own @eslinter but I don't know how to join @eslint-community, and eslint-community is a bit too long to myself, like @typescript-eslint, I'd prefer @tseslint for example.

Although it's too late as @bradzacher said, but we still have the chance.

---

Another interesting thing is that, I own https://github.com/eslinter/eslint-plugin-jsx and the npm package now, so I'll start to adapt @jsx-eslint packages like https://github.com/jsx-eslint/eslint-plugin-react into eslint-plugin-jsx when I finished the migration of eslint-plugin-import.

I really hope @Rel1cx can join in this case.

[43081j]

doing such good work already 🙏 ill have a read through the PR this weekend if i can

on the naming - i do agree it'd be nice if it had a "full" name (rather than i) but i understand it already has a fair amount of users. if it were upto me, i'd probably call it eslint-plugin-import under a scope.

the eslint community org would be good to put this into but you're right that the scope is pretty lengthy. even if you don't adopt that name, it may still be worth joining that org on github though (assuming it is "official" and actively managed, and doesn't remove any ownership from you).

[fisker]

Any chance to have eslint-plugin-module or eslint-plugin-modules?

[JounQin]

Any chance to have eslint-plugin-module or eslint-plugin-modules?

@fisker I'll try to contact these two package owners.

[Shinigami92]

@eslint-community/eslint-plugin-import would be nice, @eslint-community is a trusted community group

[fisker]

Does @eslint-community/eslint-plugin-import even work in legacy ESLint config?

It should be eslint-plugin-___ or @___/eslint-plugin, am I wrong? If it works, the plugin name will be @eslint-community/import?

[JounQin]

Does @eslint-community/eslint-plugin-import even work in legacy ESLint config?

Yes. The users can always use the plugin full name or @eslint-community/import for shorthand.

[fisker]

{
   '@eslint-community/import/consistent-type-specifier-style': 'off'
}

Great rule name! 😄

[JounQin]

Great rule name! 😄

That's why I'd prefer a short package name.

[JounQin]

Progress:

Test Suites: 25 failed, 29 passed, 54 total
Tests:       257 failed, 2226 passed, 2483 total

[JounQin]

cc @MichaelDeBoey @ota-meshi for @eslint-community

[MichaelDeBoey]

@JounQin Although I agree with the choices made in this fork and I don't get why @ljharb doesn't want to add them (I understand the reasoning of BC, but I don't agree with people still using Node v4, especially not for running their linter), I'm not sure if we want to have a fork of a package that's still actively maintained but has different opinions tbh 🤔

I'll talk to the rest of the core @eslint-community team to decide what to do with this case, as normally we're only taking over (widely dependent upon) ESLint-related packages that are either widely dependent upon and/or that are abandoned by the original author or where the original author wants to give it a safer place so it doesn't get abandoned if they're unable to continue maintenance for whatever reason

[ljharb]

fwiw I'm highly likely to do a breaking change in the ~first~ last part of this year to drop eslint < 8, and the associated node versions.

[43081j]

I'm not sure if we want to have a fork of a package that's still actively maintained but has different opinions

choices are good. if it doesn't make sense to have two in the eslint-community org, maybe that's fair enough. but out in the wild, it'd be nice to see multiple options sometimes, regardless of why they exist. just fwiw.

i'm likely to move many repos to this plugin, away from the other, whatever the outcome of the naming and wherever it lives. some will be similar. so maybe the whole eslint-community thing is something for future rather than now (if it becomes depended upon enough).

[ljharb]

It def makes sense to wait until sufficient adoption before moving anything into that org.

[JounQin]

@MichaelDeBoey

I'm not sure if we want to have a fork of a package that's still actively maintained but has different opinions tbh

That's the biggest problem, a lot of issues can not be resolved due to this problem

I'll talk to the rest of the core https://github.com/eslint-community team to decide what to do with this case

Glad to hear. And I want to clarify, we'll move forward w/wo joining in @eslint-community, and I own @eslinter, I'll migrate eslnt-plugin-react in next step.

---

@ljharb

fwiw I'm highly likely to do a breaking change i the first part of this year to drop eslint < 8, and the associated node versions.

Sorry, but I can't believe it and we've removed the concept of different resolvers, I don't think that's acceptable for you.

It def makes sense to wait until sufficient adoption before moving anything into that org.

That's OK, we have our own org @eslinter.

[43081j]

I'll migrate eslnt-plugin-react in next step.

did you already start this? i have a branch lying around which does a bunch of it from what i remember

if you want any help, ping me when you make the repo and i'll push the branch for you to look at, at least

(don't want to take this thread off topic so lets catch up once the repo exists rather than here)

[ljharb]

@JounQin true, because that makes it useless for preact and svelte and vue and styled-components etc users.

[JounQin]

@43081j

I haven't personally, but I know a great project https://github.com/Rel1cx/eslint-react, that's why I said I really hope @Rel1cx can join us.

[Rel1cx]

@43081j

I haven't personally, but I know a great project Rel1cx/eslint-react, that's why I said I really hope @Rel1cx can join us.

I strongly support you to do a full rewrite of eslint-plugin-import, maybe called eslint-plugin-esm + eslint-plugin-cjs (two separate packages), or eslint-plugin-module (one package). But as for migrating eslint-plugin-react to eslint-plugin-jsx, my point of view is consistent with this reply https://github.com/Rel1cx/eslint-react/issues/52#issuecomment-1785153861.

[JounQin]

@Rel1cx In my view, some rules are universal, some rules are framework specific.

For react rules, we can have jsx/react/x rules, for vue rules, we can have jsx/vue/x rules, etc, and we can have jsx/x rules for universal cases.

So in eslint-plugin-jsx, @eslint-react could be an dependency.

[Rel1cx]

You could make an eslint-plugin-jsx but that only makes sense if the rules in it are truly framework-agnostic, and considering that most of the natural framework-agnostic rules have already been migrated to @stylistic/eslint-plugin-jsx, there aren't that many left.

Where JSX fits into the ecosystem of front-end languages and frameworks:

JS in Vue TS in Vue TSX in Vue

JS in Solid TS in Solid TSX in Solid

...

In essence, JSX is just a syntax extension, how to lint TSX here is the same as TS, JS here, it's up to the frameworks, making an eslint-plugin-jsx that doesn't just take care of the correctness of JSX itself, but also has to be configured (or prefixed) to support the behaviour of all the specific frameworks that have used JSX in the past, are using JSX now, and will be in the future is equivalent to making a @typescript-eslint/eslint-plugin but not only for the correctness of TS itself but also configured (or prefixed) to support the behaviour of all the frameworks that use TS, React, Preact, Solid, Vue, etc.

Don't think of JSX as anything special, it's basically the same thing as JS and TS.

[43081j]

Just my two cents, I roughly agree.

• there should be a jsx plugin which lints jsx syntax (similar to a html validator), but nothing framework specific
• there should be a plugin for each framework with the framework specific rules, whether using jsx or not

[JounQin]

OK, like @jsx-eslint, I own @eslint-jsx, we can make framework specific plugins at that org.

[JounQin]

Progress:

Test Suites: 21 failed, 33 passed, 54 total
Tests:       171 failed, 2108 passed, 2279 total

[JounQin]

After thinking, I think eslint-plugin-import-x could be the best choice just like https://github.com/eslint-community/eslint-plugin-es-x.

[43081j]

i think that makes sense, since future forks of other plugins could follow the same naming convention too. nice to have some consistency there

[JounQin]

I've published eslint-plugin-import-x as the successor of eslint-plugin-i, it should work as previous for now, except it's not an alias anymore, all settings are renamed into import-x/ prefix.

The whole progress was tough, #26 was unable to continue because I chose to migrate codes instead tests first.

But finally I migrated babel-core into @babel/core, babel-eslint into @babel/eslint-parser, etc., successfully at #30. And I also migrated mocha/sinon to jest.

But I haven't removed the resolver concept: let's continue discussing at #40.

---

In the next step, I'll try to migrate the whole code base into TypeScript.