postcss 8.4.48 probably broke `stylelint-stylus`, `@stylistic/stylelint-plugin` and `stylelint-plugin-stylus` tests

[Mouvedia]

e.g. for stylelint-plugin-stylus

10/11 2:01AM: passes
10/11 10:19PM: 8.4.48 is released
11/11 1:58AM: fails

https://github.com/stylus/stylelint-stylus/blob/f57fc3379bad2c3eb8454f846ddec11bd609f00d/package.json#L78
https://github.com/postcss/postcss/releases/tag/8.4.48
https://www.npmjs.com/package/stylelint-plugin-stylus?activeTab=dependencies
https://github.com/stylelint/stylelint-ecosystem-tester/actions/workflows/test-package-stylelint-plugin-stylus-8e7.next.yml

[ybiquitous]

I could reproduce an issue in the @stylistic/stylelint-plugin test suite with postcss-html:

		{
			code: `<div>
<style>a {
  color: red;
}</style>


</div>`,
			message: messages.rejected,
			line: 4,
			column: 1,
		},

https://github.com/stylelint-stylistic/stylelint-stylistic/blob/32cff1c6107ca12bd149c81727fd6bc1377d3853/lib/rules/no-missing-end-of-source-newline/index.test.js#L95-L105

We can also see this issue on our demo site as follows.

With postcss@8.4.47, the lint problem's location is 3:2-3. See demo.

With postcss@8.4.48, the lint problem's location is 2:9-10. See demo.

The difference between the two demos is only postcss versions.

I suppose that this commit (postcss/postcss@1a8b261) for rangeBy() is related, but I'm not still sure which package we should fix, e.g. postcss, postcss-html, @stylistic/stylelint-plugin, or stylelint. 🤔

@romainmenke If you have any ideas, please feel free to let us know. 🙏🏼

[Mouvedia]

but I'm not still sure which package we should fix

You could say that it should have been a minor version.
i.e. 8.4.47 → 8.5.0
But that's irrelevant now, if postcss consider it a fix it is a patch.

tl;dr: I have opened this issue so that we can track the issues on @stylistic/stylelint-plugin and stylelint-stylus

[romainmenke]

PostCSS is large enough that any change is a breaking change for someone.
This type of breakage was expected with those changes.

---

If I understand the issue correctly then this is roughly what is happening.

postcss-html (and likely similar packages) have node.source.input.css as the content of <style>...</style> block while also having node.source.start relative to the whole document.

We can't have both, those are contradictory.
This is best reported upstream in affected packages so that their respective communities can provide patches.

postcss-html and similar packages use the Document node to represent multiple Root nodes found in a non-css file. Like html, jsx, ... So there needs to be some kind of mapping between both coordinate systems.

With the patches in PostCSS the workarounds used by postcss-html and others likely became incorrect.

A minimal repro for postcss-html:

const assert = require('assert');
const syntax = require("./lib/index.js")();
const document = syntax
	.parse(`<div>
<style>
a {
  color: red;
}</style>

</div>`, { syntax, from: 'foo' });

assert.deepStrictEqual(
	document.first.rangeBy({ index: document.first.source.input.css.length - 1 }),
	{ end: { column: 2, line: 5 }, start: { column: 1, line: 5 } }
);

I don't currently have much time and this is unlikely to improve over the next few months.
So any help with reporting this upstream would be greatly appreciated 🙇

[Mouvedia]

@firefoxic @ota-meshi

[ybiquitous]

Thank you for your help. Let me investigate it more.
Once I identify the cause, I'll open an issue upstream. 👍🏼

[romainmenke]

I've also filed an issue in PostCSS as I think we can make a small change there to make it easier for postcss-html and other to update their code:

postcss/postcss#1995

[Mouvedia]

@ai @ajcwebdev
see postcss/postcss#1996

[ai]

I am going to release it today =^_^=

[Mouvedia]

@firefoxic we are down from 2 tests failing to 1.
i.e. 14/01/25 vs 15/01/25