Book: PDF output with SVG images

[pat-s]

I've read #196 and the linked documentation but still do not know how to proceed.

We're currently aiming to switch from a large bookdown book to quarto.

For the PDF output we get errors like

compilation failed- error
LaTeX Error: Cannot determine size of graphic in ./images/learner.svg (no Bou
ndingBox).

See the LaTeX manual or LaTeX Companion for explanation.
Type  H <return>  for immediate help.
 ...                                              
                                                  
l.1636 \includegraphics{./images/learner.svg}

The source references the file via

```{r 02-basics-001, echo = FALSE}
knitr::include_graphics("images/ml_abstraction.svg")
```

This works with bookdown but quarto does not seem to take it directly.

Can quarto already handle such cases dynamically?

[jjallaire]

@cderv Does bookdown do something automatic to handle SVG files in LaTeX output?

[cderv]

I don't think we are doing anything specific. SVG in LaTeX does not play well. A conversion from SVG images to PDF images is usually needed. It could be done automatically somehow using a tool like rsvg-convert or just R package magick. This was discussed in rstudio/bookdown#658. We did not include anything specific in bookdown or knitr and let a use do the conversion itself.

So usually, for HTML + PDF project, what happens is a PNG or SVG version of the image is provided for HTML output, and a PDF version for the LaTeX version. knitr::include_graphics() has a special argument auto_pdf to set to TRUE so that the image extension will be changed to .pdf if it exists. This is not default behavior, either auto_pdf = TRUE in each call to include_graphics() either there is an option knitr.graphics.auto_pdf you can set to TRUE.
With this, this means that in

knitr::include_graphics("images/ml_abstraction.svg")

it will be rendered by knitr as

\includegraphics{./images/learner.pdf}

@pat-s I believe you don't have this feature activated in your new quarto version of the project.
It is working in bookdown for you because:

• You have pdf version of the svg image: https://github.com/mlr-org/mlr3book/blob/main/bookdown/images/learner.pdf
• You are using include_graphics() to include image
• You are setting the knitr option knitr.graphics.auto_pdf = TRUE in your setup.R script: https://github.com/mlr-org/mlr3book/blob/main/bookdown/setup.R#L7
• This setup.R script is loaded in bookdown using before_chapter_script : https://github.com/mlr-org/mlr3book/blob/main/bookdown/_bookdown.yml#L4

I believe this is why it is working in your bookdown version.

So how to fix that in your Quarto project ?

Quarto does not have a before_chapter_script option. One way is to use a setup chunk in each file like this:

```{r, file = "setup.R", include = FALSE}
```

Using a .Rprofile is another option, as it should be taken into account in the R process used during rendering each chapter.

Otherwise, depending on what needs to be set, one can also use special yaml option in Quarto doc. Like knitr options in yaml for opts_knit or opts_chunk. Setting R option should also be possible to set this way

knitr:
  opts_chunk: 
    R.options:
      knitr.graphics.auto_pdf: true

This could be set on a .qmd file YAML, globally in _quarto.yml or in a shared metadata file on folder basis _metadata.yaml

@pat-s I hope this help building the mlr3 book.

@jjallaire if we want to support this better out of the box in Quarto, I think this would require doing a conversion in post processing step or within Lua filter when a SVG image is detected.

Pandoc does it when converting to PDF I believe
https://github.com/jgm/pandoc/blob/ae8ed408b03688f4ddf6a13d0e55fba9138c6eb9/src/Text/Pandoc/PDF.hs#L202-L216
It is using rsvg-convert CLI tool, but there are several variation it seems, including a npm package: https://www.npmjs.com/package/rsvg

I am thinking this could also be printed to PDF probably as we are doing with Diagrams. wouldn't it ?

But last time it was discussed (#196), I am seing we prefer to be done by user directly. So, we may not want to add that built-in.

[pat-s]

@cderv Wow, again impressed by your level of detail when it comes to such replies. Just amazing.

Also you inferred that it's about the mlr3book and digged so deep - huge thanks!

I can happily report that I got it working 🥳 I went with the YAML option for defining knitr.graphics.auto_pdf: true.
I was not aware anymore that we had set this option via before_chapter_script in the bookdown setup. It might have even been me who did this but this is one point that is quickly overlooked.

Feel free to close if everythings has been said - I'll leave it open in case you want to continue discussin with JJ.

[jjallaire]

@cderv It does seem like we should consider doing this when rsvg-convert is available on the path. We'd just need a Lua filter for this that targets PDF/LaTeX/Beamer. I'd think we'd want to do it automatically (as the other alternative is an error message from LaTeX). Could you take a shot at a PR for this? (note that the node module requires native code so not something we can easily port to deno).

cc @cscheid

[cderv]

I'd think we'd want to do it automatically (as the other alternative is an error message from LaTeX).

The other option I see is supporting natively in Quarto an option like knitr.graphics.auto_pdf that would change any image path extension to pdf for LaTeX outputs. This would allow a user to do like in mlr3 book and provide SVG for HTML and PDF for LaTeX outputs. When PNG are provided, it works ok in LaTeX and HTML.

This solution would mean Quarto is not in charge of providing PDF version of images but let a user be responsible for providing several image format.

I'll give a shot though at the Lua Filter to see if we can provide an auto conversion feature.

[jjallaire]

Note that we already have support for multi-format figures (just provide the figure name without an extension and it will bind to the right thing): https://quarto.org/docs/authoring/figures.html#multiformat-figures

[cderv]

Oh perfect. I missed that. We may need to support path without extension in knitr::include_graphics() or maybe set knitr.graphics.auto_pdf = TRUE by default when knitr is used in a Quarto project.

[jjallaire]

I think if we could do this in Lua it would work everywhere (and then we wouldn't have to direct R users to include_graphics, they could just use images in markdown as they are accustomed to)

[fuhrmanator]

Here's a MWE repo of how I get SVG into PDF in a book with no magic (unless it's the latexmk that's doing the magic?). You can ignore the filters for the PlantUML/Graphviz diagrams - I made a sample svg-test.qmd that just links to an SVG file and it renders just fine in PDF. Hope it's somehow helpful.