Can't set metadata for gfm output with lua filter

[qiushiyan]

Description

I am creating a lua filter in Quarto 1.5 to extract all headings from the source document, and want to add a headings field to the output gfm document.

Here is my lua filter and qmd

-- Function to create a slug from a string
local function slugify(s)
    return s:lower()
        :gsub("[^%w%s-]", "") -- remove non-word chars
        :gsub("%s+", "-")     -- replace spaces with hyphens
        :gsub("^-+", "")      -- trim starting hyphens
        :gsub("-+$", "")      -- trim ending hyphens
end

-- Table to store extracted headings
local headings = {}

-- Function to process headers
function Header(el)
    if el.level == 2 or el.level == 3 then
        local title = pandoc.utils.stringify(el.content)
        table.insert(headings, {
            depth = el.level,
            title = title,
            slug = slugify(title)
        })
    end
    return el
end

function Meta(meta)
    quarto.log.output("headings: " .. pandoc.utils.stringify(headings))
    meta["headings"] = headings
    return meta
end

---
format:
  gfm:
    variant: +yaml_metadata_block
    filters:
      - filters/headings.lua
---

# Main Title

## First Section

Some content here.

### Subsection A

More content.

## Second Section

Even more content.

### Subsection B

Final content.

During render, the filter is getting headings correctly and prints

(headings: 2first-sectionFirst Section3subsection-aSubsection A2second-sectionSecond Section3subsection-bSubsection B)

But no headings field is added to the output.

[mcanouil]

Why would you think there will be something? You are adding the headings as metadata not as output.

[qiushiyan]

How could I output the updated metadata? I assumed the output will contain the updated meta object. I found this example on the pandoc lua documentation to update the date field and I thought this should work in a similar way

function Meta(m)
  if m.date == nil then
    m.date = os.date("%B %e, %Y")
    return m
  end
end

[mcanouil]

You need to make the metadata part of the main AST content.

function Pandoc(doc)
  ...
  return doc
end

I think there is a format flag from Pandoc to keep the metadatablock but can't reminder it.

format: gfm+yaml_metadata_block

[qiushiyan]

Thanks for the tip, and I am using gfm+yaml_metadata_block but I still could not get it working after adding this

function Pandoc(doc)
    local meta = doc.meta

    meta.headings = headings
    for _, h in ipairs(meta.headings) do
        print(h.title)
    end
    return pandoc.Pandoc(doc.blocks, meta)
end

From the print results I can tell that the headings are added correctly to the meta object, it is just that they are not pushed to the content. I also tried loop through the doc.meta object and it's empty, there is no field such as title or author. Perhaps this is a problem of how the gfm format deals with the meta object?

[cderv]

Your filter is perfectly fine, as it correspond to the orders filter are run (first Header will be processed, before any Meta)

If you try with pandoc it will work (removing quarto.log.output specific API)

quarto pandoc index.qmd -t gfm+yaml_metadata_block -L headings.lua -s

However, with Quarto, we don't support this use case for now. For some reason, we are taking over +yaml_metadata_block behavior to recreate the feature, but without taking into account any modification by user in Lua filter.

To be clear, very simply, we do get the YAML in the input, and put it back on the output when asked. So currently doing what you want is not possible. We would need to reimplement the logic differently.

That is a current limitation of Quarto

[cderv]

FWIW, I opened an issue to track the idea for our future development

• [FR] Allow to modified Meta in Lua to be write back in Markdown output with +yaml_metadata_block #10686

[qiushiyan]

Thanks for the explanation.