Failing test: Comma selectors are evaluated independently

[j-f1]

Initial checklist

• I read the support docs
• I read the contributing guide
• I agree to follow the code of conduct
• I searched issues and couldn’t find anything (or linked relevant results below)
• If applicable, I’ve added docs and tests

Description of changes

This PR adds a failing test demonstrating what I think is a bug. I wanted to use hast-util-select to assign the correct heading path (h1.textContent → h2.textContent → p) to paragraphs on a document. But it looks like unist-util-select’s selectAll essentially evaluates selector.split(',').flatMap(selectAll), resulting in out-of-order results.

I noticed that wrapping my selector in :matches() causes results to be emitted in the correct order (although doing this prevents me from using nesting selector combinators like >, so it doesn’t work for my use case).

I think this should either be fixed (so that the ordering of results from :matches(a, b) and a, b are the same) or documented as a limitation.

[wooorm]

Thanks for the repro. I think it might the same issue as syntax-tree/unist#66 (reply in thread).

[wooorm]

Well, that was very hard to solve. The whole architecture had to be different.
I managed to solve it in hast-util-select, should soon land here.

[wooorm]

Fixed!