How can I avoid unwanted table breaking in PDF (luatex) output?

[alan-isaac]

Description

I am aware of issue 13961. Mine is almost inverted (and I'm not using Typst). In my PDF output, even very short tables are split rather than floating. I want them to be ordinary floating LaTeX tables in PDF output. Is that not the default? If not, can I make it the default? (I do not have any long tables.) Can I set it on a table-by-table basis? Thanks!

[mcanouil]

As you did not provide anything actionable:

• use keep-tex: true to investigate LaTeX code produced
• search online for LaTeX code needed then use that in include-* or as attribute {...}.

I'm not sure why you linked an issue that has nothing to do with yours.

If you want a more direct answer please follow the guidelines / discussion template to provide sufficient information for others to help.

[mcanouil]

You've asked several questions on the board already. When you open a Q&A discussion you have a clear description of what is needed for you to get the help you want.
Here you give no context about the origin of the table and no reproducible example that would help us get context.

• https://github.com/quarto-dev/quarto-cli/blob/main/.github/DISCUSSION_TEMPLATE/q-a.yml

Pandoc is the one emitting longtable table for LaTeX when converting Markdown table and it does not offer direct means to change that.

[alan-isaac]

OK, I see that pandoc uses longtable by default and without a setting. I find this a terrible default, because it breaks even very short tables across pages. Will Quarto 2.0 offer more control? The LaTeX floating tables are very desirable!

[mcanouil]

Quarto is not emitting the longtable from markdown.
I don't believe the team will reimplement Pandoc writers either. (I Might be wrong)

You can choose to emit the table yourself with code (gt, knitr::kable, etc.) or raw LaTeX.

In my personal opinion, the issue is in longtable which does not offer any control on almost everything it does. I don't believe fixing it is the role of third parties.

---

Note that because Pandoc does emit longtable, what Quarto does is a lot of fixup for numbering/counters/cross-references.

A possible workaround, would be for you to write a Lua filter that process the tables the way you want when targeting PDF/LaTeX, but this can have quite some expected consequences.

[alan-isaac]

So Quarto relies on pandoc to convert markdown tables to LaTeX tables, and pandoc currently emits LaTeX tables using only the longtable environment. There is currently no standard way to produce tables for multiple formats where the emitted LaTeX will use the table environment (because e.g. kable emits tabular not table). Post-processing of the LaTeX document might be possible, but simply replacing longtable with table looks risky. Use of a Lua filter to change table processing may be possible but is likely to have unexpected consequences. If the only output format is LaTeX, then of course one may replace markdown tables with raw LaTeX that uses a table environment.

Is that about right?

[cscheid]

Will Quarto 2.0 offer more control?

Not in the short run. Quarto 2 will continue to use Pandoc for writing to non-HTML formats like LaTeX, typst, docx, etc. It is in reading markdown that Quarto 2 (almost) entirely takes over the responsibility away from Pandoc.

To be clear, I share your frustration with the state of table rendering to LaTeX. But we can't simply take on the entire production of LaTeX from markdown for every use case. There are multiple decades of open-source contributions that we leverage, and Quarto is developed by a really small team.

If you have sufficiently strong opinions about how this should work, then you're best served by taking on the responsibility and writing that code yourself using the extension functionality in Quarto/Pandoc (Lua filters and output: asis code outputs from engines).

[mcanouil]

So Quarto relies on pandoc to convert markdown tables to LaTeX tables, and pandoc currently emits LaTeX tables using only the longtable environment. There is currently no standard way to produce tables for multiple formats where the emitted LaTeX will use the table environment (because e.g. kable emits tabular not table). Post-processing of the LaTeX document might be possible, but simply replacing longtable with table looks risky. Use of a Lua filter to change table processing may be possible but is likely to have unexpected consequences. If the only output format is LaTeX, then of course one may replace markdown tables with raw LaTeX that uses a table environment.

Is that about right?

Yes, that's pretty much it as of today.

[cscheid]

Small clarification:

If the only output format is LaTeX, then of course one may replace markdown tables with raw LaTeX that uses a table environment.

table is a cross-referencing environment, not a table-emitting environment. longtable* emits tables; longtable does both (and hardcodes the counters from table in its invocations, which we can't control, and a large reason why longtable causes ongoing issues in Quarto). tabular emits tables.

tabularray is the modern approach, but very few packages and systems use it: https://ctan.math.illinois.edu/macros/latex/contrib/tabularray/tabularray.pdf

[cderv]

There is currently no standard way to produce tables for multiple formats where the emitted LaTeX will use the table environment (because e.g. kable emits tabular not table).

To clarify this, Python and R does have table packages that will help you produce table output for multiple format too. For latex, some use longtable and other don't. This is what Mickael said with "You can choose to emit the table yourself with code (gt, knitr::kable, etc.)" .

And this is currently the only way to really control your table in a cross format manners.

I think your issue is close to this pandoc report

• Prevent tables smaller than one page from spanning two pages in PDF output jgm/pandoc#782

As you can see it is very old, but still active (last comment by the maintainer last March). So maybe Pandoc will change.