Table Attr (id, classes, attributes) is silently dropped when tbl-colwidths triggers a SimpleTable round-trip

[eukeni-acebal]

I have:

• searched the issue tracker for similar issues
• installed the latest version of Quarto CLI
• formatted my issue following the Bug Reports guide

Bug description

Bug description

resolve_table_colwidths_scoped (in share/filters/modules/tablecolwidths.lua) applies column widths by round-tripping the table through pandoc.utils.to_simple_table / from_simple_table. SimpleTable has no Attr, so the rebuilt table loses its identifier, classes and attributes.

This is invisible for tables labelled the normal Quarto way ({#tbl-x} in the caption), because crossref carries the id on a wrapping div. It bites when a user Lua filter sets el.identifier (or el.attr) on a Table: the attribute is present in the AST until the pre-scope-resolution step and gone right after it.

With the HTML default tbl-colwidths: auto, the round-trip is triggered by the mere presence of a Link anywhere in the table — the auto branch scans for links and, if it finds one, zeroes the computed widths. So the symptom looks arbitrary from the outside: tables with links lose their id, tables without links keep it.

With tbl-colwidths: false or an explicit list, the attributes are lost for every table, link or no link.

Notes

Related but not the same:

• tbl-colwidths and table's label are incompatible in Quarto Book #7072 (closed): {#tbl-x tbl-colwidths="[…]"} declared in the caption of a book project.
• Non-simple tables with rowspans cannot contain links #13776 (closed by skip processing non-simple tables (#13776) #14096): the same to_simple_table / from_simple_table round-trip, losing a different thing — merged cells. The fix added an is_simple(tbl) guard that bails out when the table has spans, but the round-trip is still lossy for Attr on every other table, so the guard doesn't cover this case.
  A guard for spans and a silent Attr loss suggest the round-trip itself is the wrong tool here; setting the widths on tbl.colspecs directly would avoid both.

Steps to reproduce

filter.lua:

function Table(el)
  el.attr = pandoc.Attr("my-id", {"my-class"}, {{"data-mine", "1"}})
  return el
end

_quarto.yml:

project:
  type: default
filters:
  - filter.lua
format:
  html:
    toc: false

with-link.qmd:

---
title: With link
---
 
| A | B |
|---|---|
| one | [two](https://example.org) |
 
: Caption.

without-link.qmd — the same file, with the cell two instead of the link.

Then quarto render with-link.qmd without-link.qmd --to html.

Actual behavior

<!-- without-link.html -->
<table id="my-id" class="my-class caption-top table" data-mine="1">
 
<!-- with-link.html -->
<table class="caption-top table">

QUARTO_TRACE_FILTERS=trace quarto render with-link.qmd shows the Attr intact through pre-output-location and empty at pre-scope-resolution, which is where resolve_scoped_elements runs resolve_table_colwidths_scoped.

Setting tbl-colwidths: true keeps the Attr in both files — that branch returns nil and never rebuilds the table. Setting tbl-colwidths: false or [50,50] loses it in both.

Expected behavior

Applying column widths should preserve the table's Attr. Something like reassigning result.attr = tbl.attr after from_simple_table, or setting the widths on tbl.colspecs directly instead of going through SimpleTable.

Your environment

Quarto 1.10.18 (also reproduced on 1.5.57)

• macOS
• Not R/Python specific: plain .qmd, no engine.

Quarto check output

❯ quarto check
Quarto 1.10.18
[✓] Checking environment information...
Quarto cache location: /Volumes/macmini external ssd/eukeni_external/Library/Caches/quarto
[✓] Checking versions of quarto binary dependencies...
Pandoc version 3.10.0: OK
Dart Sass version 1.101.0: OK
Deno version 2.7.14: OK
Typst version 0.15.1: OK
[✓] Checking versions of quarto dependencies......OK
[✓] Checking Quarto installation......OK
Version: 1.10.18
Path: /Applications/quarto/bin

[✓] Checking tools....................OK
TinyTeX: (not installed)
Chrome Headless Shell: (not installed)
VeraPDF: (not installed)

[✓] Checking LaTeX....................OK
Using: Installation From Path
Path: /Library/TeX/texbin
Version: 2025

[✓] Checking Chrome Headless....................OK
Using: Chrome found on system
Path: /Applications/Google Chrome.app/Contents/MacOS/Google Chrome
Source: MacOS known location

[✓] Checking basic markdown render....OK

[✓] Checking R installation...........(None)

  Unable to locate an installed version of R.
  Install R from https://cloud.r-project.org/

[✓] Checking Python 3 installation....OK
Version: 3.14.6
Path: /opt/homebrew/opt/python@3.14/bin/python3.14
Jupyter: (None)

  Jupyter is not available in this Python installation.
  Install with python3 -m pip install jupyter

[✓] Checking Julia installation...

[mcanouil]

@eukeni-acebal what's this about?
You are saying something about attributes and col widths but your example does not show any of this.
You are instead using a Lua filter to mangle with the AST without accounting for the processing phase.

You also closed the issue "as completed" without any comment or explanation.

Please clarify your issue and more importantly ensure it's a simple example that actually use/show the issue. From what you are saying, no Lua filter is needed.

Your report seems to be a partial copy/paste of something maybe generated by an LLM.

To be clear, I see no attribute loss, so based on your example, that seems to be your own doing.

A link in a cell or not does not change anything.

---
format: html
---

| A | B |
|---|---|
| 1 | 2 |

: Caption {#tbl-one .striped tbl-colwidths="[25,75]" data-whatever="true"}

<div id="tbl-one" class="striped quarto-float quarto-figure quarto-figure-center anchored" data-whatever="true" data-tbl-colwidths="[25,75]">
<figure class="quarto-float quarto-float-tbl figure">
<figcaption class="quarto-float-caption-top quarto-float-caption quarto-float-tbl" id="tbl-one-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
Table&nbsp;1: Caption
</figcaption>
<div aria-describedby="tbl-one-caption-0ceaefa1-69ba-4598-a22c-09a6ac19f8ca">
<table class="table-striped caption-top table">
<colgroup>
<col style="width: 25%">
<col style="width: 75%">
</colgroup>
<thead>
<tr class="header">
<th>A</th>
<th>B</th>
</tr>
</thead>
<tbody>
<tr class="odd">
<td>1</td>
<td>2</td>
</tr>
</tbody>
</table>
</div>
</figure>
</div>

[eukeni-acebal]

Hi @mcanouil, and thank you for taking the time to look at this — you're right on every count, and I'm sorry for the noise. This was my first bug report anywhere, and it shows. Let me go through your points one by one.

On closing the issue "as completed" with no comment. That was pure clumsiness on my part, not a statement about the issue. When I opened it, GitHub told me I had already filed something a month earlier, which confused me; then I found myself in a menu of options that all looked irreversible, and I picked one without understanding what it meant. I did not mean to mark it resolved, and I certainly didn't mean to do it silently. Apologies — I understand now that "completed" is a claim about the issue, not a way of tidying my own list.

On the example not showing what the prose describes. Fair. I wrote the report around what I had inferred from reading tablecolwidths.lua and from a QUARTO_TRACE_FILTERS trace, and then attached the reproduction that had originally bitten me, without checking that the two matched. They don't. The prose talks about identifiers, classes and attributes in general; the example only ever exercises one narrow path.

On the Lua filter and the processing phase. This is the actual answer, and it's my mistake. User filters run before Quarto's filters by default, so my Table function was setting an Attr upstream of resolve_table_colwidths_scoped and then watching Quarto legitimately rebuild the table underneath it. I re-ran the whole thing on 1.10.18 to check:

• filters: [filter.lua] (the default): Attr lost when the table contains a link, kept when it doesn't.
• at: pre-quarto: same — lost with a link, kept without.
• at: post-quarto: Attr kept in both cases.
• filters: [quarto, filter.lua]: Attr kept in both cases.

So there is nothing to fix on my side except where I put the filter, and the "arbitrary" behaviour I described was just me straddling the phase boundary.

On "no Lua filter is needed". I went back and tried to build the filter-free reproduction you asked for, and I couldn't — because there doesn't appear to be a user-visible symptom to reproduce:

• Caption classes survive the round-trip: : Caption. {.my-class} renders as <table class="my-class caption-top table"> both with and without a link in the table, and with tbl-colwidths set to auto, false or an explicit list.
• Crossrefs are unaffected: {#tbl-a} on a table containing a link numbers and links correctly; the id sits on the wrapping quarto-float div, exactly as I'd guessed in the report.
• Arbitrary ids and data attributes can't be set from Markdown at all — : Caption. {#my-id data-mine="1"} leaves them as literal text in the caption, so there's no path by which a user could lose them.

In other words, my claim that the round-trip "loses its identifier, classes and attributes" was overstated: from Markdown, nothing reachable is lost.

What's actually left, if anything. Only a small internal point, which I'll state once and then leave entirely to your judgement. from_simple_table returns a table with a blank Attr, and the return in resolve_table_colwidths_scoped

return pandoc.utils.from_simple_table(simpleTbl), false

doesn't carry tbl.attr across. Nothing in Quarto's own pipeline seems to depend on that at this point, so it's a robustness nit rather than a bug — a result.attr = tbl.attr before returning, or setting the widths on tbl.colspecs directly, would make the step attribute-preserving and would also make the is_simple guard from #13776 unnecessary. But given that I can't produce a user-facing failure, this doesn't warrant an open issue on its own. Please close it (properly this time!), or tell me and I'll close it myself with a summary of the above so the trail is useful to anyone who lands here from a search.

Thanks again for the correction — and, sincerely, for the lesson in how to be a better citizen here. I'd much rather be told than go on filing reports like this one. Sorry for the time it cost you.

[mcanouil]

All good, thanks for clarifying.

Would you mind share a simple small reproducible example about your last point to get a better idea on the context?

[eukeni-acebal]

Of course — and thanks for the patience.

The last point isn't really reproducible as a rendering symptom, which is exactly why I ended up writing such a confusing report. The cleanest way I can show it is at the Pandoc level, with no Quarto and no document involved. roundtrip.lua:

-- pandoc lua roundtrip.lua
local tbl = pandoc.read("| a | b |\n|---|---|\n| 1 | 2 |\n", "markdown").blocks[1]
tbl.attr = pandoc.Attr("my-id", {"my-class"}, {{"data-mine", "1"}})

local back = pandoc.utils.from_simple_table(pandoc.utils.to_simple_table(tbl))

print("before -> id: '" .. tbl.identifier .. "' classes: " .. #tbl.classes .. " attributes: " .. #tbl.attributes)
print("after  -> id: '" .. back.identifier .. "' classes: " .. #back.classes .. " attributes: " .. #back.attributes)

With the Pandoc 3.10 shipped in Quarto 1.10.18:

before -> id: 'my-id' classes: 1 attributes: 1
after  -> id: '' classes: 0 attributes: 0

That's inherent to SimpleTable, which has no Attr field — not a Pandoc bug, just the shape of the type. The consequence for Quarto is that resolve_table_colwidths_scoped returns

return pandoc.utils.from_simple_table(simpleTbl), false

i.e. a table whose Attr is unconditionally blank, whatever tbl.attr held on the way in.

Where that becomes visible. Only through a filter, and only one running before Quarto's own filters — which is the default, and was my mistake. What made it hard to diagnose from the outside is that the loss is conditional and silent: under the HTML default tbl-colwidths: auto, the round-trip only fires when the table contains a Link (the auto branch scans for links and, if it finds one, zeroes the widths). So with the same filter, an identical pair of tables behaves differently depending on whether a cell happens to contain a link, with no warning. That's what I misread as an attribute bug.

Where it is not visible. I went looking for a Markdown-only path and couldn't find one, so for the record:

• caption classes survive — : Caption. {.my-class} renders as <table class="my-class caption-top table"> with and without a link, and with tbl-colwidths set to auto, false or an explicit list;
• crossrefs are unaffected — {#tbl-a} on a table containing a link numbers and links correctly, since the id lives on the wrapping quarto-float div;
• arbitrary ids and attributes can't be set from Markdown at all — : Caption. {#my-id data-mine="1"} is left as literal text in the caption.

So this is a robustness point, not a user-facing bug: result.attr = tbl.attr before returning, or setting the widths on tbl.colspecs directly, would make the step attribute-preserving — and the colspecs route would also make the is_simple guard added for #13776 unnecessary, since it wouldn't need the round-trip at all. Entirely your call whether that's worth carrying; I'm happy either way, and I'll leave the issue in your hands rather than closing it a second time.