Table caption in DOCX attached to wrong table

[rgaiacs]

Consider table-mwe.docx that has two tables:

Let's convert the .docx to Markdown with pandoc --from docx --to markdown+table_captions table-mwe.docx.

Observed Output

The caption is connected to the first table instead of the second table.

Lorem ipsum

  -----------------------------------------------------------------------
  A                                   B
  ----------------------------------- -----------------------------------
  C                                   D

  -----------------------------------------------------------------------

  : Numbers from 1 to 4

Lorem ipsum

  -----------------------------------------------------------------------
  1                                   2
  ----------------------------------- -----------------------------------
  3                                   4

  -----------------------------------------------------------------------

Lorem ipsum

Expected Output

Lorem ipsum

  -----------------------------------------------------------------------
  A                                   B
  ----------------------------------- -----------------------------------
  C                                   D

  -----------------------------------------------------------------------

Lorem ipsum

  -----------------------------------------------------------------------
  1                                   2
  ----------------------------------- -----------------------------------
  3                                   4

  -----------------------------------------------------------------------

  : Numbers from 1 to 4

Lorem ipsum

Environment

pandoc --version returns

pandoc 3.1.11.1
Features: +server +lua
Scripting engine: Lua 5.4
User data directory: /home/raniere/.local/share/pandoc
Copyright (C) 2006-2023 John MacFarlane. Web: https://pandoc.org
This is free software; see the source for copying conditions. There is no
warranty, not even for merchantability or fitness for a particular purpose.

[jgm]

The xml structure is:

w:tbl
w:p (Lorem ipsum)
w:p (with pStyle "Caption" and w:keepNext)
w:tbl

[jgm]

OK, it looks like the docx reader does the following:

• in bodyToOutput: looks for all the captions among the body paragraphs and puts a list of them in state
• in bodyPartToBlocks for Tbl: gets the list of captions from state, takes the first one, and modifies state to contain the rest

So the captions are assigned to tables in the order they occur, no matter their proximity to the tables. Obviously that's giving bad results in this case, but it is a bit tricky to devise better heuristics.