[ebrahimeolakey]

Issue resolved by this Pull Request:
Resolves #4382

_handle_tables read rows through python-docx Table.rows, which only yields direct w:tr children of w:tbl. A row that Word wrapped in w:customXml or in a content control (w:sdt) was never visited, so its text was lost and num_rows came out short. Cells had the same gap for w:customXml; the w:sdt cell case was already handled by _row_cells (#3951).

Changes:

• New _table_rows() returns a table's w:tr elements in document order, unwrapping w:customXml and w:sdt/w:sdtContent on the way down. _handle_tables builds its _Row objects from that list instead of Table.rows, so grid_cols_before and the rest of the row handling are unchanged.
• _row_cells() now also unwraps w:customXml around a w:tc.
• Regression test builds the four-row table from the issue (plain row, customXml row, sdt row, plain row with a customXml cell) and checks the 4×2 grid cell by cell.

Result on the issue's reproduction: 2 x 2 with the last row showing only plain D1 before, 4 x 2 with every cell in place after.

This is the "extension of _row_cells plus a row-level equivalent" option from the issue. It does not overlap with #4369: that change is in _walk_linear at body level, this one is in _row_cells / _handle_tables, and the two touch different parts of msword_backend.py.

tests/test_backend_msword.py: 66 passed, 1 skipped (LibreOffice not installed). No groundtruth files changed.

Checklist:

• Documentation has been updated, if necessary. (docstrings only; no user-facing docs affected)
• Examples have been added, if necessary. (not needed)
• Tests have been added, if necessary.

[github-actions]

✅ DCO Check Passed

Thanks @ebrahimeolakey, all your commits are properly signed off. 🎉

[mergify]

Merge Protections

🟢 Merge protection satisfied — ready to merge.

Show 1 satisfied protection

🟢 Enforce conventional commit

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

• title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert)(?:\(.+\))?(!)?:

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[chrikrah]

@ebrahimeolakey This fixes #4382 as filed, and I would merge it. Walking w:tbl yourself instead of patching _Row is the right call: docling reads only grid_cols_before and _tr off a row, and both work on a wrapped one, while _Row._index raises AttributeError: 'lxml.etree._Element' object has no attribute 'tr_lst' because the parent is no longer the table.

non-blocking, and it is yours to take or leave: the interaction most likely to regress is a vertical merge whose continuation row is wrapped. It is right on 8bef881 and wrong on the base, so a case would pin what you already fixed. The XML and the two assertions are below if you want them.

$ PYTHONPATH=. ~/.venvs/docling/bin/python repro_4382.py      # my four-row case from the issue
2d5c590c   2 x 2   missing: customXml cell D2, customXml row B1, customXml row B2, sdt row C1, sdt row C2
8bef881    4 x 2   missing: none

$ PYTHONPATH=. ~/.venvs/docling/bin/python repro_4393_edges.py
2d5c590c   vMerge continuation in w:customXml   2 x 2, 'merged' row_span=1, B2 lost
8bef881    vMerge continuation in w:customXml   3 x 2, 'merged' row_span=2, spanning r0-r2

  <w:tr>            <w:tc><w:tcPr><w:vMerge w:val="restart"/> ... merged   <w:tc> ... A2
  <w:customXml w:element="Tagged"><w:tr>
                    <w:tc><w:tcPr><w:vMerge w:val="continue"/> ...         <w:tc> ... B2
  <w:tr>            <w:tc> ... C1                                          <w:tc> ... C2
  assert data.num_rows == 3
  assert next(c for c in data.table_cells if c.text == "merged").row_span == 2

$ PYTHONPATH=. ~/.venvs/docling/bin/python repro_trpr_ins.py
2d5c590c   w:tr/w:trPr/w:ins   2 x 2, both inserted cells read
8bef881    w:tr/w:trPr/w:ins   2 x 2, both inserted cells read

$ ~/.venvs/docling/bin/python -m pytest tests/test_backend_msword.py -q
2d5c590c   65 passed, 1 skipped in 66.36s
8bef881    66 passed, 1 skipped in 57.71s
# 32 of the 35 documents under tests/data/docx/sources are compared here: drawingml, textbox and
# test_emf_docx are skipped locally because LibreOffice is not installed.

$ git show 2d5c590c:docling/backend/msword_backend.py > docling/backend/msword_backend.py
$ ~/.venvs/docling/bin/python -m pytest tests/test_backend_msword.py -k wrapped -q   # your test kept
>       assert data.num_rows == 4
E       AssertionError: assert 2 == 4
1 failed, 1 passed, 65 deselected in 2.51s

$ ~/.venvs/docling/bin/ruff check docling/backend/msword_backend.py tests/test_backend_msword.py
All checks passed!   # ruff 0.15.12

Collated from two runs per commit on this machine, and no CI result is quoted. ty is a pre-commit hook on docling/*.py and was not run, since it is not installed here.

@dolfim-ibm this touches msword_backend.py, which you and @ceberam merge most often. @ebrahimeolakey, want that vMerge case as a patch against your branch, or would you rather write it?

[ebrahimeolakey]

Thanks for running it against both commits. Wrote the vMerge case myself and pushed it, with the same three-row shape as yours: restart row, continuation row inside w:customXml, plain row. It fails on main with 2 rows and passes here with row_span == 2.

[chrikrah]

@ebrahimeolakey Still approve at 8a75684. The wrapped-row test fails on the base backend and passes here, so it guards the merge case from my last review.

$ PYTHONPATH=. ~/.venvs/docling/bin/python -m pytest tests/test_backend_msword.py -q -p no:cacheprovider
8a75684    67 passed, 1 skipped in 71.66s       # 8bef881 was 66 passed, 1 skipped

$ git show 2d5c590c:docling/backend/msword_backend.py > docling/backend/msword_backend.py   # tests kept
$ PYTHONPATH=. ~/.venvs/docling/bin/python -m pytest tests/test_backend_msword.py -q -p no:cacheprovider -k wrapped
E       AssertionError: assert 2 == 4
E       AssertionError: assert 2 == 3
2 failed, 1 passed, 65 deselected in 2.90s

$ git merge-tree --write-tree upstream/main 8a75684        # main at ed04377
CONFLICT (content): Merge conflict in docling/backend/msword_backend.py

Main moved under the docx imports, so GitHub shows a conflict. Could you rebase? Keeping both CharacterStyle and _Row resolves it. @PeterStaar-IBM once that is in, this is ready for a look.