Using paragraph in tbl-cap or fig-cap cell options do not work and nothing prevents it

[cderv]

---
title: "Example"
format: html
keep-md: true
---

```{r}
#| label: tbl-test
#| tbl-cap: |
#|    Big table 
#|    
#|    Really small

knitr::kable(
  mtcars[1:6, 1:3]
)

I don't think this should be supported - and thus nothing prevents a rendering to a broken doc

![image](https://github.com/quarto-dev/quarto-cli/assets/6791940/5394a089-3aac-40b2-b733-778626def004)

Intermediate Md is 

````markdown
---
title: "Example"
format: html
keep-md: true
---

::: {#tbl-test .cell tbl-cap='Big table

Really small'}

```{.r .cell-code}
knitr::kable(
  mtcars[1:6, 1:3]
)

::: {.cell-output-display}

	mpg	cyl	disp
Mazda RX4	21.0	6	160
Mazda RX4 Wag	21.0	6	160
Datsun 710	22.8	4	108
Hornet 4 Drive	21.4	6	258
Hornet Sportabout	18.7	8	360
Valiant	18.1	6	225

::: :::

See how the new line in YAML will break the line on the fenced div attributes, creating an unsupported syntax

[cderv]

Same for figure by the way

---
title: "Example"
format: html
keep-md: true
---

```{r}
#| label: fig-cap-margin
#| fig-cap: |
#|   MPG vs horsepower, colored by transmission.
#|   
#|   Something else
library(ggplot2)
mtcars2 <- mtcars
mtcars2$am <- factor(
  mtcars$am, labels = c('automatic', 'manual')
)
ggplot(mtcars2, aes(hp, mpg, color = am)) +
  geom_point() +
  geom_smooth(formula = y ~ x, method = "loess") +
  theme(legend.position = 'bottom')

![image](https://github.com/quarto-dev/quarto-cli/assets/6791940/5418f14f-06da-40fe-b1d6-2b58c792792e)

````markdown
---
title: "Example"
format: html
keep-md: true
---

::: {.cell}

```{.r .cell-code}
library(ggplot2)
mtcars2 <- mtcars
mtcars2$am <- factor(
  mtcars$am, labels = c('automatic', 'manual')
)
ggplot(mtcars2, aes(hp, mpg, color = am)) +
  geom_point() +
  geom_smooth(formula = y ~ x, method = "loess") +
  theme(legend.position = 'bottom')

::: {.cell-output-display} ![MPG vs horsepower, colored by transmission.

Something else](index_files/figure-html/fig-cap-margin-1.png){#fig-cap-margin width=672} ::: :::

I believe we should avoid producing such intermediate markdown in both cases

[cscheid]

One trick Rich and I started using is to support base64 encoded strings in computationally-generated content. For example, you can include markdown in HTML tables by using <span data-qmd-base64="yourbase64contenthere"></span>

We could consider doing something similar here.

The main design complication is that I'd like to avoid adding a tbl-cap-base64 to tbl-cap (and similarly to all other such attributes, it would be a big mess to implement).

We could, for example, support the following syntax:

::: {#tbl-test .cell tbl-cap='__quarto__base64__contents__:yourbase64contentshere'}

Then, when we read those attributes in quarto (and we already have to parse those to markdown anyway), we can detect the __quarto__base64__contents__: string prefix and do the right thing.

All we need to do is to come with one such string like __quarto__base64__contents__ that:

• will never be the start of a human-generated attribute
• is clear enough for readers of the intermediate markdown to identify what's going on
• only contains safe ascii characters (to avoid encoding issues)

[cderv]

Oh interesting !

Though at first I was thinking that we do no really support multi paragraph caption. Do we ?

Usually Markdown syntax how do that work ? Only last paragraph below tables is taken into account as caption.

Because for now tbl-cap is placed on the .cell div because it is forwarded there by knitr with no processing of tbl-cap.

We could just disallow new line and paste the strings together if tbl-cap or fig-cap is seen as a string with newlines. 🤷‍♂️

[cscheid]

Though at first I was thinking that we do no really support multi paragraph caption. Do we ?

It's complicated (surprise!).

I don't think there's anything stopping a caption from having multiple paragraphs in a FloatRefTarget ("float"), but we don't have a lot of ways to end up with multiple paragraphs there. The div syntax for floats requires a caption to be a single paragraph: src/resources/filters/quarto-pre/parsefiguredivs.lua:193, and refCaptionFromDiv is

function refCaptionFromDiv(el)
  local last = el.content[#el.content]
  if last and last.t == "Para" and #el.content > 1 then
    return last
  else
    return nil
  end
end

But there are other ways for captions to show up. The most direct one is a Lua filter that constructs a FloatRefTarget explicitly. In that case, caption can be pretty much anything, and we don't currently check for the type of content (it can even be a Blocks({...}) with two paragraphs, for example).

I wouldn't be surprised if there were other ways they can sneak in.