bug(parser): fix parsing of multiline class attribute value (#4242)

[kreativ-los]

Fixes #4242

[Conduitry]

Is the issue that this is fixing just that the scoped .foo styles aren't recognized as applying to the element with the CSS class containing foo and newlines? I didn't see any other problems in #4242. I don't think removing this whitespace during parsing is the right solution, because the user might want that whitespace for certain attributes. As far as I can tell, what needs to be fixed here is how the compiler identifies which selectors might apply to which elements.

edit: For a concrete example, if someone had a title attribute on an element which contained newlines, we'd need to preserve those. We don't want to change how we're parsing attributes.

[ghost]

Yeah, I agree with @Conduitry.
I think this pull request will strip that type of whitespace off of all attributes, not just class.

Normal DOM matching will strip for the purpose of matching selectors, but retain it for the purpose of rendering the class name. We need to build an analog of DOMTokenList or at least implement the whitespace rules, not change the parsing of the attributes itself.

The bug itself is still a great catch since it works in traditional pages and not with scoped styles.

[tanhauhau]

@kreativ-los https://github.com/sveltejs/svelte/blob/master/src/compiler/compile/css/Selector.ts#L211 is where we match the dom nodes with the class name, you can figure out why multiline causes it to not match.

[kreativ-los]

Thanks for your feedback :) I will give @tanhauhau answer a look and try to fix it in the compiler. Will this also fix the "Unused CSS selector" linting error, or is this done separately?

[Conduitry]

That should take care of all of it. The reason for that warning is because the compiler doesn't see that selector as applying to that element. Thanks!