Text references don't work with multiline lines

[eliocamp]

Text references don't work with multiline lines

[eliocamp]

Here: https://bookdown.org/yihui/bookdown/markdown-extensions-by-bookdown.html#text-references it states that "The text can contain anything that Markdown supports, as long as it is one single paragraph."

In markdown (at least the flavour used by bookdown) sentences belonging to the same paragraph can be each in its own line. You need two newlines to define a new paragraph. (You can see that in the bad example, the text reference is rendered as a single paragraph).

Also, PDF output does work. This seems to be an issue for HTML (gitbook and epub) and word formats only.

[mschilli87]

@eliocamp: Do you mind linking the documentation you mentioned? I always assumed this was by design. Either way, I can definitely confirm the behaviour you describe for as long as I remember using text references.

[eliocamp]

The issue is around here:

bookdown/R/html.R

Lines 807 to 821 in 52c31aa

	parse_ref_links = function(x, regexp) {
	r = sprintf(regexp, reg_ref_links)
	if (length(i <- grep(r, x)) == 0) return()
	tags = gsub(r, '\\1', x[i])
	txts = gsub(r, '\\2', x[i])
	if (any(k <- duplicated(tags))) {
	warning('Possibly duplicated text reference labels: ', paste(tags[k], collapse = ', '))
	k = !k
	tags = tags[k]
	txts = txts[k]
	i = i[k]
	}
	x[i] = ''
	list(content = x, tags = tags, txts = txts, matches = i)
	}

At this point x has one element per line instead of one element per paragraph; like this:

<p>(ref:caption) Here is a nice figure!
And this is a second sentence.</p>

So grep(r, x) doesn't find it (the line doesn't match r = "^<p>(\\(ref:[-/[:alnum:]]+\\)) (.+)</p>$"). The fix might be to join each p block in its own line, but I don't know if that could break other stuff.

Using --wrap none here actually solves the issue, but it might create others.

bookdown/R/html.R

Lines 80 to 82 in 52c31aa

	pandoc_args2 = function(args) {
	if (pandoc2.0() && !length(grep('--wrap', args))) c('--wrap', 'preserve', args) else args
	}

[eliocamp]

Ok. It seems that --warp=preserve was introduced to fix #504. But that's only related to LaTeX, which doesn't have this problem.
Removing this option only for HTML documents seems to solve the issue in HTML and all tests pass (at least locally on my machine).

That would just pass pandoc_args directly in all HTML functions instead of passing it through pandoc_args2(). For instance, in this line:

bookdown/R/html.R

Lines 59 to 62 in 52c31aa

	config = get_base_format(base_format, list(
	toc = toc, number_sections = number_sections, fig_caption = fig_caption,
	self_contained = FALSE, lib_dir = lib_dir,
	template = template, pandoc_args = pandoc_args2(pandoc_args), ...

As a workaround, one can set

  pandoc_args:
    - --wrap=none

on the output format options in the YAML.

Don't know how you feel about this. If you agree with the fix, I can sent a PR.

[cderv]

Thanks for the investigation. I think we should have used --wrap=none for HTML output instead of --preserve.

This is part of change in HTML output with pandoc 2.17 where wrap has now effect.
I wanted to do that in #1304 but preserve was already set. You reminded us why (#504)

[king-of-poppk]

I've got a similar issue when embedding \n's in fig.cap: the Figure x.y caption prefix is not resolved. Is this related or is it a separate issue?

[cderv]

@king-of-poppk please open a new issue with a reproducible example and we can have a look