Only first element in `pandoc.Figure` content is shown

[cwickham]

---
title: "Untitled"
format: html
filters:
  - double-image.lua
---

![](image.png)

![Caption](image.png)

double-image.lua

function Image(img)
  return {img, img}
end

The inline images are duplicated, but the image with a caption (that ends up in a pandoc.Figure) is not:

cc @cscheid

[mcanouil]

I think you need to target (also) the cross-reference float using the custom node FloatRefTarget: https://quarto.org/docs/prerelease/1.4/lua_changes.html#support-for-crossreferenceable-elements-in-filters

[cwickham]

If it is actually a cross-reference, it works OK:

---
title: "Untitled"
format: html
filters:
  - double-image.lua
---

![Caption](image.png){#fig-1}

This is just the special case where it gets enclosed in pandoc.Figure

[cderv]

From tracing the filters, this happens in render_pandoc3_figure() as we could have guessed
https://github.com/quarto-dev/quarto-cli/blob/main/src/resources/filters/layout/pandoc3_figure.lua

The process only handles a single Image inside the Figure

quarto-cli/src/resources/filters/layout/pandoc3_figure.lua

Lines 41 to 47 in 23d0172

	local function html_handle_image(figure)
	local image
	_quarto.ast.walk(figure, {
	Image = function(img)
	image = img
	end
	})

So if we need to retrieve all images, while handling caption (on the last image only ?)

quarto-cli/src/resources/filters/layout/pandoc3_figure.lua

Lines 51 to 53 in 23d0172

	if figure.caption.long ~= nil then
	image.caption = quarto.utils.as_inlines(figure.caption.long)
	end

and classes too

quarto-cli/src/resources/filters/layout/pandoc3_figure.lua

Lines 55 to 59 in 23d0172

	for i, v in pairs(figure.classes) do
	if v:match("^margin%-") or v:match("^quarto%-") or v:match("^column%-") then
	image.classes:insert(v)
	end
	end

We currently move them on the expected single image inside the Figure

Aim of this process is to move back to previous syntax

pandoc.Para({pandoc.Image()})

We also remove the caption later in this process - so maybe this is just to pass it around 🤷‍♂

quarto-cli/src/resources/filters/layout/pandoc3_figure.lua

Line 60 in 23d0172

local result = htmlImageFigure(image)

quarto-cli/src/resources/filters/layout/html.lua

Lines 131 to 153 in 23d0172

	function htmlImageFigure(image)
	return renderHtmlFigure(image, function(figure)
	-- make a copy of the caption and clear it
	local caption = image.caption:clone()
	tclear(image.caption)
	-- pandoc sometimes ends up with a fig prefixed title
	-- (no idea way right now!)
	if image.title == "fig:" or image.title == "fig-" then
	image.title = ""
	end
	-- insert the figure without the caption
	local figure = { pandoc.Para({image}) }
	return figure, caption
	end)
	end

Anyhow, this is to think through.

@cwickham you could work around this maybe by duplicating the image later in the process;

---
title: "Untitled"
format: html
filters:
  - at: post-render
    path: double-image.lua
---

![](surus.png)

![Caption](surus.png)

[cscheid]

We need to fix this in 1.8 because we expect that using filters to create responsive figures automatically (inserting two images with .dark-content and .light-content respectively) will be a common pattern.

[cscheid]

We need to decide what happens with captions.

I don't think there's a great answer:

1. if we choose to pick the caption of either of the images, then users might be confused in case they put the caption in the wrong one
2. if we concatenate all captions, then users might be confused if they generate captions for all images (which will happen if they're not careful)

I'm leaning towards 2 because it makes the problem visible and obvious (missing a caption will be harder to spot). I'll also add a warning in this case.

[cscheid]

This particular issue turns out to be quite delicate to fix, because the presence of multiple images in a pandoc.Figure interferes with htmlImageFigure, discoverLinkedFigureDiv, renderHtmlFigure, and so on.