transforming <p>honk<br /></p> into Markdown yields a stray backslash?

[cpiro]

Hi all! I've been using these tools to turn the HTML output from a tiptap-based editor into Markdown. In particular, I've been seeing weird behavior when the editor document ends in a line break:

1. The editor outputs <p>honk<br /></p>. This seems to be a valid HTML fragment.
2. rehypeParse parses that in the expected way, giving a hast with a p tag containing the text then a br tag.
3. rehype2remark / hast-util-to-mdast transforms that hast into the mdast with a paragraph containing the text then a break.
4. remarkStringify transforms that mdast into the Javascript string "honk\\\n", that is, four characters followed by a single backslash then a LF.

The commonmark.org dingus says that the output of 4. is the Markdown that means the text "honk\" (five characters, the last of which is a backslash). remarkParse agrees that the output of 4. corresponds to the a paragraph containing only the five characters of text "honk\", and indeed the CommonMark spec 6.9 says that

Hard line breaks are for separating inline content within a block. Neither syntax for hard line breaks works at the end of a paragraph or other block element

Clearly I don't want the Markdown output to have a stray backslash character at the end when the editor has no backslashes in it. The problem seems to be that it's OK to have a paragraph that ends in a line break in HTML, but in CommonMark it is not. In 3. hast-util-to-mdast transforms the ASTs in the straightforward way, but is the output a "valid" mdast? Or is it the responsibility in 4. for remarkStringify not to emit a stray backslash for its valid input? Of course I could sanitize the HTML on the way out of the editor to remove trailing brs from a p, but this seems to be valid HTML. Or is there something I'm missing altogether?

Thank you!

Here's the code I used to test:

const html = "<p>honk<br /></p>";
const hast = unified().use(rehypeParse).parse(html);
const mdast = toMdast(hast); // from "hast-util-to-mdast"
const md = unified().use(remarkStringify).stringify(mdast);

const check = unified().use(remarkParse).parse(md);

There's a codesandbox.io doc with the above, and here's a screenshot of the console output (with some of the AST fields stripped out for readability):

[wooorm]

Hi there 👋

Yeah, it’s valid HTML. I think this would be something for hast-util-to-mdast.
The same could be done with headings: <h1>x<br></h1>

I think it would be better to just remove them, instead of the alternative of injecting some non-breaking invisible character after them.

Anyway, probably a filter-trailing-breaks or so utility in hast-util-to-mdast used for both headings andf paragraphs?