do not wrap HTML as it was done by default before Pandoc 2.17

[cderv]

as this breaks our parsing process for section references.

This fixes #1301

THis PR is for now limited to the related issue but there are maybe other processing affected.

[yihui]

Thanks!

[cderv]

Thanks for the quick review. I have just added a test so that we can detect some issue in the future. More tests like this would be required IMO. It would be great to detect Pandoc's breakage earlier. 😅

[yihui]

Yes, we definitely need more tests.

[cderv]

Suggested change

	c(pandoc_args, '--section-divs', '--mathjax', '--number-sections',
	if (rmarkdown::pandoc_available("2.17")) c('--wrap', 'none'))
	c(pandoc_args2(pandoc_args), '--section-divs', '--mathjax', '--number-sections')
	)

This also work it seems. Is it better as we already use --wrap elsewhere ?

[cderv]

I have made the change

[cderv]

And tests were failing 😞 ...

Snapshot test are really useful with markdown output as we can see clearly how it looks. However, maintaining several pandoc version is painful as the output will heavily change depending on the default behavior. So I went back to regex matching.
maybe using xml2 parsing would be the safest to target specific item and avoid regex.

I also use pandoc_args2() as it seems more relevant here as we are using it for other conversion in bookdown.

Does it seems better now ?

[cderv]

Does it seems better now ?

I think it is. I am merging.