References in markdown links and spans parsed as citations

[knuesel]

For example:

pandoc -t native
(@a) First case

[link to (@a)](url)

gives

[ OrderedList
    ( 1 , Example , TwoParens )
    [ [ Plain [ Str "First" , Space , Str "case" ] ] ]
, Para
    [ Cite
        [ Citation
            { citationId = "a"
            , citationPrefix =
                [ Str "link" , Space , Str "to" , Space , Str "(" ]
            , citationSuffix = [ Str ")" ]
            , citationMode = NormalCitation
            , citationNoteNum = 1
            , citationHash = 0
            }
        ]
        [ Str "[link" , Space , Str "to" , Space , Str "(@a)]" ]
    , Str "(url)"
    ]
]

The same occurs with spans, e.g. with

(@a) First case

[span with (@a)]{}

It seems to me the [...](...) and [...]{...} semantics should have priority over the citation semantics.

There is also a problem with pandoc-crossref that might be related. Example:

pandoc -F pandoc-crossref -t native
# Title {#sec:title}

[text (@sec:title)]{}
[ Header 1 ( "sec:title" , [] , [] ) [ Str "Title" ]
, Para [ Str "text" , Space , Str "(" , Space , Str "1){}" ]
]

so we get the string text ( 1){}, instead of text (sec. 1) inside a span.

Reproduced with pandoc master (465c28d) on Pop OS based on Ubuntu 21.04.

[jgm]

I agree that this is undesirable in this case, but there's a general problem of ambiguity with citation and example list syntax.

Is this a regression from earlier versions, or has it always behaved this way?

[knuesel]

It was like this already in pandoc 1.13 (the earliest version I tested).

So the current behavior has history on its side. It's causing me some headaches though: it breaks the composability of spans, which makes them a lot less useful.

For context, my use case is a bilingual document with many paragraphs of the form

[some content]{lang=fr}
[some content]{lang=de}
$$some math$$

or

[some content]{lang=fr}
[some content]{lang=de}
$some math$
[more content]{lang=fr}
[more content]{lang=de}

I make one output where both languages appear (with different styles using CSS), and also one output for each language (the other is removed by a Lua filter).

I thought spans where perfect for this but it's not working when I have equation references in the text, which led me to file this issue.

[jgm]

I see the difficulty. I'll see if I can think of a solution.

[knuesel]

Thanks for fixing this!

[jgm]

One nagging worry about the fix: with this fix, [@foo]{.class} will never be parsed as a normal citation. (It will be parsed as a span containing an author-in-text citation.) That's fine with pandoc's markdown as it stands, but if we ever port over the attributes extension (now possible with commonmark), authors will expect to be able to add attributes to normal citations. Not sure how to handle this. There may be a better fix. But at the moment this is fine.