Field-based cross-references instead of hyperlinks in LaTeX2docx conversion

[JPZ4-5]

When using LaTeX2docx conversion, how can we achieve field-based cross-references (for captions and text references) rather than hyperlinks? This is crucial for automatic numbering updates when merging DOCX documents.

Current Behavior:

1. pandoc -o output.docx main.tex can convert \caption{XX} in the table and figure environments in LaTeX to captions in plain text, above tables or below figures. The \ref{} in the text can be converted to hyperlinks.
2. pandoc -F pandoc-crossref -o output.docx main.tex can convert \caption{XX} in the table and figure environments in LaTeX to captions with numbering in plain text, above tables or below figures. The \ref{} in the text can be converted to hyperlinks.
3. pandoc -F pandoc-tex-numbering -o output.docx main.tex can customize captions well, where Figure X-XX: XXXXX has Figure X-XX as a hyperlink and : XXXXX as plain text. The \ref in the text can be converted to hyperlinks. This is also the recommended latex2docx filter in the pandoc-crossref documentation.

Problem:
Haven't found any existing method produces Word field codes (like SEQ fields) for captions/references. Manual conversion (inserting caption fields + replacing hyperlinks with cross-reference fields) is prohibitively tedious.

Questions:

1. Are there mature solutions for automatic field-based conversion?
2. If not, is it better to:
   • Develop a custom Pandoc filter (preferred), or
   • Create VBA macros (undesirable due to lack of VBA expertise)?
   • record a macro in Word and execute it in a loop?

LLM is recommending something like below, wondering correction:

1. During the LaTeX parsing phase, map \caption and \ref to Word's SEQ and REF fields.
2. A filter can be written in Python to intercept AST nodes and insert field codes (e.g., Figure { SEQ Figure \* ARABIC }).

Appreciate any recommendations or implementation insights!

[jgm]

LLM has no idea. "During the LaTeX parsing phase, map \caption and \ref to Word's SEQ and REF fields." Pandoc parses the LaTeX into a format-agnostic Pandoc structure, not something docx-specific.

What we could perhaps do is add a specially marked Span around the generated numbers; this would be enough of a hook for the docx reader to substitute a field.

I'll convert this to an issue with the suggestion.

[badumont]

What we could perhaps do is add a specially marked Span around the generated numbers; this would be enough of a hook for the docx reader to substitute a field.

I think that there is enough information in the link object currently made up from e.g. \ref{fig:myfig}: ``` [ Link ( "" , [] , [ ( "reference-type" , "ref" ) , ( "reference" , "myfig" ) ] ) [ Str "1" ] ( "#myfig" , "" ) ] ```

[jgm]

What about the number in the figure caption itself?

[badumont]

There is none, but it is easy to add an OOXML field at the beginning of the content of the Caption within a Figure, isn't it? However, I must confess that inserting a number in the link but not in the caption looks a bit strange.