exporting all fig-alt together with figure numbers

[baptiste]

Description

I'm writing a book with CRC using Quarto + knitr + Latex, and the publishers are asking that the ALT text for each figure be copied into a single Word document. Has anyone written a Lua filter doing something like this?

• grab the fig-alt field of each figure chunk
• dump the text into a new Word document, together with the figure number

Many thanks

[mcanouil]

That should not be too hard.
Do you have a small reproducible example document to use?
(Note that likely to have a post-render script for Word document)

[cscheid]

This is "easy" in 1.4. You'll need:

1. A filter that targets FloatRefTarget nodes
2. ... and that runs after crossrefs are handled. In 1.4 you have better control over when filters run. That's not currently documented (🤦) but you can see an example here. The currently available points for injecting a filter are: pre-ast, post-ast (or its synonym pre-quarto), post-quarto (or its synonym pre-render), and post-render.

pre-quarto and post-render are the old injection points of "pre" filters and "post" filters in Quarto 1.3 and earlier.

This separation clarifies the three major phases in our Lua filters: ast processing handles normalization of the input markdown into a semantic AST; quarto handles what quarto does: resolving citations, references, layouts, subfloats, etc, but does not emit format-specific output. render takes the processed AST and produces a format-specific document.

(This separation is at least somewhat aspirational, and there's still a bit of format-specific output in the quarto phase, specifically in the resolution of references. We'll be cleaning this up progressively in future versions of quarto.)

Specifics

Your filter will look more or less like this:

local captures = pandoc.Blocks({})
return {
  FloatRefTarget = function(float)
    -- float.order is a data structure that stores where the float is found
    -- use quarto.log.output() to inspect its contents, or follow the code in `refNumberOption`
    -- from src/resources/filters/crossref/format.lua
    -- then, use that to create a pandoc.Para
    captures:insert(pandoc.Para(...))
  end,
  Pandoc = function(doc)
    return pandoc.Pandoc(captures) -- replaces the result with a completely new document
  end,
}

Then, in your .qmd you'll use:

---
filters:
  - at: post-quarto
    path: capture.lua
---

Of course, you probably want to consider putting this metadata in a separate .yml file that you conditionally invoke, so that you get your regular content unless you specifically enable the filter.

This is the kind of thing we would love to support better in the future, and why we did the crossref overhaul. One of the things that's pretty janky right now is that we don't expose refNumberOption, etc, or document the order field at all. We'll be improving on this as time goes by, and we could eventually change those fields, but I don't expect we will anytime soon.

EDIT: "Oh, a book"

This is significantly more complicated :( The fundamental problem is that book rendering is multi-file, and so the crossref information is computed differently. For books, we have an additional Pandoc pass that runs in formats that produce a single file (.epub, .pdf, .docx, etc) We'll still need to run the filter above, but only in the final pass that produces the single output.

If this weren't a user filter, you'd use param("single-file-book", false) to check in which pass the filter is getting called; but you don't have access to that :(

Let me talk to the other quarto folks and get back to you.

[baptiste]

Thanks a lot! Yeah, I figured the book aspect might complicate things... Looking forward to hearing more.
I wonder if a custom pandoc Writer might be a way to go? Presumably at that stage quarto is done with its numbering etc., and is letting pandoc do its thing with the output format. The Writer could presumably just call the standard Writer but also output the fig-alt content into a new document in parallel? (Edit: nevermind, I guess that's similar to post-quarto or post-render, but still called in the multiple passes for split documents)

[baptiste]

By the way quarto v1.4.489 seems to not show fig-alt in html in this example. I thought it might be easiest for now to post-process a concrete output document (e.g. HTML, or whatever else keeps the ALT info in the output), grep the figure number and the ALT content, and dump those into a Word document converted via pandoc.