Using `data-qmd` in `div` does not seem to work as expected (gt usecase)

[cderv]

This was found in the context of #7493

From rstudio/gt#146 I understand that using HTML in a table colums to show as HTML in gt table, it requires to use fmt_markdown

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

```{r}
library(gt)
data.frame(A = "Text", B = "**Text**") |> 
  gt() |> 
  fmt_markdown(B)
```

With 1.4 we get a Lua error

Error running filter C:/Users/chris/DOCUME~1/DEV_R/QUARTO~2/src/resources/filters/main.lua:
...2/src/resources/filters/./quarto-pre/parsefiguredivs.lua:425: attempt to index a nil value (field 'classes')

Stacktrace

Error running filter C:/Users/chris/DOCUME~1/DEV_R/QUARTO~2/src/resources/filters/main.lua:
...2/src/resources/filters/./quarto-pre/parsefiguredivs.lua:425: attempt to index a nil value (field 'classes')
stack traceback:
	...DEV_R/QUARTO~2/src/resources/filters/./common/pandoc.lua:52: in function <...DEV_R/QUARTO~2/src/resources/filters/./common/pandoc.lua:46>
	(...tail calls...)
	[C]: in ?
	[C]: in method 'walk'
	...V_R/QUARTO~2/src/resources/filters/./ast/customnodes.lua:86: in function <...V_R/QUARTO~2/src/resources/filters/./ast/customnodes.lua:76>
	(...tail calls...)
	..._R/QUARTO~2/src/resources/filters/./ast/runemulation.lua:40: in local 'callback'
	..._R/QUARTO~2/src/resources/filters/./ast/runemulation.lua:54: in upvalue 'run_emulated_filter_chain'
	..._R/QUARTO~2/src/resources/filters/./ast/runemulation.lua:89: in function <..._R/QUARTO~2/src/resources/filters/./ast/runemulation.lua:86>
stack traceback:
	...V_R/QUARTO~2/src/resources/filters/./ast/customnodes.lua:86: in function <...V_R/QUARTO~2/src/resources/filters/./ast/customnodes.lua:76>
	(...tail calls...)
	..._R/QUARTO~2/src/resources/filters/./ast/runemulation.lua:40: in local 'callback'
	..._R/QUARTO~2/src/resources/filters/./ast/runemulation.lua:54: in upvalue 'run_emulated_filter_chain'
	..._R/QUARTO~2/src/resources/filters/./ast/runemulation.lua:89: in function <..._R/QUARTO~2/src/resources/filters/./ast/runemulation.lua:86>
ERROR: Error
    at renderFiles (file:///C:/Users/chris/DOCUME~1/DEV_R/QUARTO~2/src/command/render/render-files.ts:343:23)
    at eventLoopTick (ext:core/01_core.js:181:11)
    at async renderProject (file:///C:/Users/chris/DOCUME~1/DEV_R/QUARTO~2/src/command/render/project.ts:268:23)
    at async renderForPreview (file:///C:/Users/chris/DOCUME~1/DEV_R/QUARTO~2/src/command/preview/preview.ts:426:24)
    at async render (file:///C:/Users/chris/DOCUME~1/DEV_R/QUARTO~2/src/command/preview/preview.ts:162:22)
    at async preview (file:///C:/Users/chris/DOCUME~1/DEV_R/QUARTO~2/src/command/preview/preview.ts:179:18)
    at async Command.fn (file:///C:/Users/chris/DOCUME~1/DEV_R/QUARTO~2/src/command/preview/cmd.ts:401:7)
    at async Command.execute (file:///C:/Users/chris/DOCUME~1/DEV_R/QUARTO~2/src/vendor/deno.land/x/cliffy@v0.25.4/command/command.ts:1790:7)
    at async quarto (file:///C:/Users/chris/DOCUME~1/DEV_R/QUARTO~2/src/quarto.ts:155:3)
    at async file:///C:/Users/chris/DOCUME~1/DEV_R/QUARTO~2/src/quarto.ts:186:5

Error is from

quarto-cli/src/resources/filters/quarto-pre/parsefiguredivs.lua

Line 425 in 95c6dd6

process_div_caption_classes(div)

from #7449 (cc @cscheid )

The HTML table we get is this one

<table class="gt_table" data-quarto-disable-processing="false" data-quarto-bootstrap="false">
  <thead>
    
    <tr class="gt_col_headings">
      <th class="gt_col_heading gt_columns_bottom_border gt_left" rowspan="1" colspan="1" scope="col" id="A">A</th>
      <th class="gt_col_heading gt_columns_bottom_border gt_left" rowspan="1" colspan="1" scope="col" id="B">B</th>
    </tr>
  </thead>
  <tbody class="gt_table_body">
    <tr><td headers="A" class="gt_row gt_left">Text</td>
<td headers="B" class="gt_row gt_left"><div data-qmd="**Text**"><div class='gt_from_md'><p><strong>Text</strong></p>
</div></div></td></tr>
  </tbody>
  
  
</table>

The div data-qmd has indeed no class.

I would say

• either we always need to test for existence before using includes

   if div.classes and div.classes:include("cell")

  which solve the issue here.

  if we do this fix, it seems there is a bunch of places where we need to do that, looking at our usage of :includes()

• either this is a broader issue where the HTML table parsing is not correctly setting an empty classes attributes. Which pandoc.Div() seems to do by default

   $ pandoc lua -i
   Lua 5.4.4  Copyright (C) 1994-2022 Lua.org, PUC-Rio
   Embedded in pandoc 3.1.9
   > div = pandoc.Div({})
   > div.classes
   List {}
   > div.classes:includes("dummy")
   false

[cscheid]

Thanks! I can repro this.

div.classes should always be a list inside a Div() walk handler. Somehow that's being called with a Blocks instead. It's a bug in the way I implement our custom AST walking code.

[cscheid]

This was a fun, tricky one. Let me document it here for future reference.

We have a utility function called combineFilters that allows us to merge multiple Pandoc filters in one, so they run "in parallel". This function is useful, but has a limitation where filters that are merged are expected to either return nil or a node of the same type that was called.

The crash here is coming from the fact that parse_md_in_html_rawblocks() is called inside quarto_ast_pipeline() in a inside a combinedFilters, but it currently returns Inlines and Blocks from a call to Span and Div handlers.

Ultimately, we need to fix combinedFilters, but that's surprisingly tricky to do correctly without causing performance issues (which is what combinedFilters was created to solve). So we'll work around the issue for now by wrapping the Inlines from parse_md_in_html_rawblocks in a scaffolding Span and Div. We already have code to remove scaffolding in a downstream filter too, so this should be pretty safe.