pandoc.utils.stringify() behaviour change in pandoc 3.6 (probably correct!)

[wlupton]

I have noticed that pandoc.utils.stringify() can behave differently in pandoc 3.6. Probably it now behaves as documented, and previously was too lax?

The case that I hit was an attempt to stringify a Caption. Previously this worked but I guess it shouldn't, because you're only supposed to be able to stringify AST elements, and captions are objects with long (Blocks) and short (Inlines) fields.

However, should the behaviour have changed? I'm not sure.

---

Here are my tests.

% cat figure.md
![Cat](Cat.png)

% cat caption.lua
function Figure(figure)
    local caption = pandoc.utils.stringify(figure.caption)
    io.stderr:write(string.format('#### caption = %s\n', caption))
end

% pandoc-3.5 figure.md -L caption.lua -t html -o /dev/null
#### caption = Cat

% pandoc-3.6 figure.md -L caption.lua -t html -o /dev/null
Error running filter caption.lua:
table expected, got Caption
stack traceback:
	caption.lua:2: in function 'Figure'

[jgm]

I defer to @tarleb on this.
I don't think there was a change on the Haskell side, but maybe one on the Lua side.

[tarleb]

Captions used to be represented as Lua tables with two keys, long and short (the latter can be missing). The stringify function interpreted that as metadata and was able to stringify it that way. Now that captions are represented as userdata values, this no longer works. We now need a special case for that.

I'll push a fix shortly.