Segmentation fault (core dumped)

[rcerljenko]

What steps are needed to reproduce the bug?

Just install latest version.

What Stylelint configuration is needed to reproduce the bug?

NA

How did you run Stylelint?

stylelint --cache <path>

Which version of Stylelint or dependencies are you using?

15.6.3

What did you expect to happen?

Run without errors.

What actually happened?

It crashes with this exception: Segmentation fault (core dumped)

Does the bug relate to non-standard syntax?

No

Proposal to fix the bug

No response

[younky-yang]

same issue happens when I updated to the latest version of 15.6.3.

[ybiquitous]

@rcerljenko Thanks for the report. I doubt this is a problem of Cosmiconfig:

• Bump cosmiconfig from 8.1.3 to 8.2.0 #6893
• Segmentation fault with 8.2.0 cosmiconfig/cosmiconfig#307

Can you confirm a workaround to downgrade Cosmiconfig to 8.1.3?

[jlherren]

Downgrading to cosmiconfig 8.1.3 did solve the issue for me.

[ghost]

最新バージョン 15.6.3 に更新したときにも同じ問題が発生します。

Me too.

[romainmenke]

I added this in my package.json to temporarily avoid the issues :

(I use npm)

  "overrides": {
    "stylelint": {
      "cosmiconfig": "8.1.3"
    }
  },

[tjx666]

I also occur this problem:

"stylelint": "^15.6.3"

This issue can occasionally be reproduced on my MBP, but it can be reproduced every time on the CI Linux.

[tjx666]

I can confirmed lock "cosmiconfig": "8.1.3" works

[lucgagan]

Confirmed that:

  "pnpm": {
    "overrides": {
      "cosmiconfig": "8.1.3"
    }
  },

fixes the issue

[d-fischer]

Hi, cosmiconfig maintainer here.

For some reason, within stylelint, any call to import() crashes. Do you have any idea why that could be? In general, even though stylelint is written in CJS. it shouldn't have any problems calling dynamic import()...

I have tried running the full prettier test suite, no issues with that... I'm not sure what's different here.

[romainmenke]

This seems relevant : https://github.com/stylelint/stylelint/blob/main/bin/stylelint.js#L6

If I remove that, the issue goes away.

---

This seems to match some of the keywords above : zertosh/v8-compile-cache#30

[alexander-akait]

Oh, yes v8-compile-cache and dynamic imports broken

[alexander-akait]

It is an old problem zertosh/v8-compile-cache#30

[alexander-akait]

There is no good solution, because vm for ESM is still experimantal and still have problems with dynamic imports and memory leaking, so there is only one solution, remove v8-compile-cache usage

[romainmenke]

This still fails.

Is it possible that jest also isn't compatible?

[alexander-akait]

Yeah, there is nodejs/node#35889 and nodejs/node#35889 (comment), ESM is still not stable as we want 😞

[d-fischer]

Well, uh, that makes sense. cosmiconfig migrated to vitest for this reason. Apparently, the switch is relatively straightforward.

[romainmenke]

Given that the issue doesn't seem trivial to resolve I think it might be best to first help those users who are affected by this.

@d-fischer Would you consider doing a rollback of the changes in 8.2 as they seem to be breaking? A semver major would be less disruptive for a feature addition like this, as users of Stylelint wouldn't get that next major unless a change is also made here first.

If that is not possible I think we should release a version of Stylelint with cosmiconfig pinned to 8.1.3. This will takes the pressure of and ensure that there is only a limited proliferation of "overrides": { "stylelint": { "cosmiconfig": "8.1.3" } }.

When the issues are resolved we can unpin the version and users won't have to take manual actual.

@alexander-akait @ybiquitous thoughts?

[alexander-akait]

@d-fischer I'll tell you a secret - vitest has the same problem, because leaking and broken dynamic import cache in v8, not in node.js or jest

[alexander-akait]

@romainmenke

If that is not possible I think we should release a version of Stylelint with cosmiconfig pinned to 8.1.3. This will takes the pressure of and ensure that there is only a limited proliferation of "overrides": { "stylelint": { "cosmiconfig": "8.1.3" } }.

yeah, it is the best solution right now - pin it and open an issue (or rename this) and do a patch release

[d-fischer]

@romainmenke

yeah, it is the best solution right now - pin it and open an issue (or rename this) and do a patch release

The removal of v8-compiler-cache would still be required, I think. I'm updating #6901 with the pin.

Nvm, I misunderstood.

[d-fischer]

@d-fischer I'll tell you a secret - vitest has the same problem, because leaking and broken dynamic import cache in v8, not in node.js or jest

Switching from jest to vitest fixed the problems of cosmiconfig's test suite with dynamic imports, that's all I can say.

[d-fischer]

My concern with that is that it might be passing the issue to users of Stylelint.

So how likely do you think it is that people will invoke stylelint as part of their test suite?

The new ESM loader of cosmiconfig is pretty much exactly what prettier has been shipping for a while, plus an additional try-catch (which doesn't catch this error because it's a segfault, sadly). I have not tested it yet, but it should be pretty much the exact same situation, but prettier didn't ever get a report about that.

To be honest, I don't have any clue how to fix this problem without just getting rid of ESM loading again. This is an ecosystem issue.

[d-fischer]

@d-fischer Is there any reason not to use the sync loaders?

Should be fine; as cosmiconfig uses await, a non-promise result should be handled just fine. The only (obscure) breakage I could imagine would be a file that includes a then property at the root level, which might just crash with an obscure error. If you want to be extra sure about that corner case, I guess you could do something like

'.js': (path, content) => Promise.resolve(defaultLoadersSync['.js'](path, content)),

[ybiquitous]

Thanks for the quick fix #6902.

To confirm, if people try using .stylelintrc.mjs (ESM), can #6902 prevent this problem?

[d-fischer]

To confirm, if people try using .stylelintrc.mjs (ESM), can #6902 prevent this problem?

I guess you would still have to remove that weird cache package for that to work.

[romainmenke]

To confirm, if people try using .stylelintrc.mjs (ESM), can #6902 prevent this problem?

No, or at least not intentionally.

I am unsure what the behavior was before when people tried to use .stylelintrc.mjs.

Is this something that was supported and should be tested?
Is this something that wasn't supported and should be handled by a nice error message?

[ybiquitous]

Currently, I don't think we should support .mjs explicitly. See also below (CLI help and doc):

stylelint/lib/cli.js

Lines 115 to 118 in f1ace61

	- a "stylelint" property in "package.json"
	- a ".stylelintrc" file (with or without filename extension:
	".json", ".yaml", ".yml", ".js", and ".cjs" are available)
	- a "stylelint.config.js" file exporting a JS object

stylelint/docs/user-guide/configure.md

Lines 5 to 8 in f1ace61

	- `stylelint` property in `package.json`
	- `.stylelintrc` file
	- `.stylelintrc.{cjs,js,json,yaml,yml}` file
	- `stylelint.config.{cjs,js}` file exporting a JS object

But, we note that some people may try using .mjs if Cosmiconfig 8.2 is installed. Let's consider it later, though.

[romainmenke]

Yeah, should just work if we remove v8-compiler-cache as @d-fischer said and could be tested with a simpler separate script that doesn't rely on jest :)

[alexander-akait]

Just infromation - we use this https://github.com/webpack-contrib/postcss-loader/blob/master/src/utils.js#L37 for js/mjs/cjs and everything works for a lot of developers

[alexander-akait]

Shorty - many tools and bundlers handle import(...) and most of the time it's expected. BUT sometimes you don't want to do it, that is why you can use non analizable code like new Function(...)

[ybiquitous]

I agree with removing v8-compiler-cache if there are no more concerns. 👍🏼

[ybiquitous]

Great thanks, @romainmenke! I've merged #6902.

I'll do a patch release today. See also #6903.

[ybiquitous]

https://github.com/stylelint/stylelint/releases/tag/15.7.0 is released! 🎉
The segmentation fault should not occur. 👍🏼

[ybiquitous]

I've opened follow-up issues:

• Remove v8-compile-cache #6904
• Support .mjs configuration files #6905

So, I'm closing this issue since it's no longer needed. Feel free to comment if you still have any concerns.

Thank you so much for helping us resolve this problem!