In PDF output, code listings can float away from their annotations

[cderv]

Following up on feedback from @giabaio and @nichtich at

• In PDF output, code listings can float away from their annotations #5142

Take this example

---
title: Hello World
format: pdf
---

## Hello

{{< lipsum 1-2 >}}

```{#lst-customers .r lst-cap="Customers Query"}


# this is a comment that adds length to the code block
# this is a comment that adds length to the code block
# this is a comment that adds length to the code block

library(tidyverse)

# this is a comment that adds length to the code block
# this is a comment that adds length to the code block
# this is a comment that adds length to the code block

library(palmerpenguins)
penguins |>                                      # <1>
  mutate(                                        # <2>
    bill_ratio = bill_depth_mm / bill_length_mm, # <2>
    bill_area  = bill_depth_mm * bill_length_mm  # <2>
  )                                              # <2>
```
1. Take `penguins`, and then,
2. add new columns for the bill ratio and bill area.

{{< lipsum 3-4 >}}

So it seems the issue from 1.3 is happening again. It was initially fixed by 5e1027b but does not seem to apply anymore

[cderv]

Unfortunately, it seems the initial fix was just lost in #6620 refactoring. We could try re-apply in the right place, or we can find another fix, but it seems here setting [H] option is really what is expected.

[cderv]

It seems our Lua processing is quite differente from before here.

As an id is used in the example, it goes through our FloatTarget processing. This means that setting lst-pos will solve the issue

---
title: Hello World
format: pdf
keep-md: true
keep-tex: true
---

## Hello

{{< lipsum 1-2 >}}

```{#lst-customers .r lst-cap="Customers Query" lst-pos='H'}


# this is a comment that adds length to the code block
# this is a comment that adds length to the code block
# this is a comment that adds length to the code block

library(tidyverse)

# this is a comment that adds length to the code block
# this is a comment that adds length to the code block
# this is a comment that adds length to the code block

library(palmerpenguins)
penguins |>                                      # <1>
  mutate(                                        # <2>
    bill_ratio = bill_depth_mm / bill_length_mm, # <2>
    bill_area  = bill_depth_mm * bill_length_mm  # <2>
  )                                              # <2>
```
1. Take `penguins`, and then,
2. add new columns for the bill ratio and bill area.

{{< lipsum 3-4 >}}

So this is current workaround.

Though, it should really be the default when code-annotation is set.

So ideally we need to find a way to link the two and set the default lst-pos here.

quarto-cli/src/resources/filters/customnodes/floatreftarget.lua

Lines 298 to 303 in 1160258

	_quarto.ast.add_renderer("FloatRefTarget", function(_)
	return _quarto.format.isLatexOutput()
	end, function(float)
	local figEnv = latexFigureEnv(float)
	local figPos = latexFigurePosition(float, figEnv)
	local float_type = ref_type_from_float(float)

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

Lines 513 to 527 in 1c1609c

	-- Computes the figure position for a figure environment
	-- margin figures use offset instead of position
	function latexFigurePosition(el, env)
	if env == kMarginFigureEnv then
	return attribute(el, kOffset, nil)
	else
	local prefix
	if el.t == "FloatRefTarget" then
	prefix = ref_type_from_float(el)
	else
	prefix = refType(el.identifier) or "fig"
	end
	return attribute(el, prefix .. "-pos", nil)
	end
	end

Maybe adding a special case if env is codelisting and the float contains annotated code block

Otherwise, I am seeing in our codebase that it happens we are post processing the LaTeX file instead... 🤔
Example

quarto-cli/src/format/pdf/format-pdf.ts

Lines 845 to 870 in 1c1609c

	const calloutFloatHoldLineProcessor = () => {
	let state: "scanning" | "replacing" = "scanning";
	return (line: string): string | undefined => {
	switch (state) {
	case "scanning":
	if (line.match(/^\\begin{tcolorbox}/)) {
	state = "replacing";
	return line;
	} else {
	return line;
	}
	case "replacing":
	if (line.match(/^\\end{tcolorbox}/)) {
	state = "scanning";
	return line;
	} else if (line.match(/^\\begin{figure}$/)) {
	return "\\begin{figure}[H]";
	} else if (line.match(/^\\begin{codelisting}$/)) {
	return "\\begin{codelisting}[H]";
	} else {
	return line;
	}
	}
	};
	};

Not ideal, but putting that here for later reference