fix(docx): keep the text of wrapped runs in plain table cells

docling/backend/msword_backend.py

@@ -2044,6 +2044,18 @@ def _get_paragraph_text(self, paragraph: Paragraph) -> str:
             for text, _format, _hyperlink in self._iter_paragraph_content(paragraph)
         )
 
+    def _get_cell_text(self, cell: _Cell) -> str:
+        """Return a table cell's text, one line per paragraph.
+
+        ``_Cell.text`` only reads runs and hyperlinks that are direct children
+        of a paragraph, so a cell would lose the text of an inline content
+        control, a simple field such as a mail-merge field, a smart tag or a
+        tracked insertion. Read each paragraph the way body paragraphs are read.
+        """
+        return "\n".join(
+            self._get_paragraph_text(paragraph) for paragraph in cell.paragraphs
+        )
+
     def _get_paragraph_elements(self, paragraph: Paragraph):
         """
         Extract paragraph elements along with their formatting and hyperlink
@@ -3301,13 +3313,14 @@ def _handle_tables(
                     continue
 
                 cell = _Cell(tc, table)
+                cell_text = self._get_cell_text(cell)
 
                 # Detect equations in cell text
                 text, equations = self._handle_equations_in_text(
-                    element=cell._element, text=cell.text
+                    element=cell._element, text=cell_text
                 )
                 if len(equations) == 0:
-                    text = cell.text
+                    text = cell_text
                 else:
                     text = text.replace("<eq>", "$").replace("</eq>", "$")
 

tests/test_backend_msword.py

@@ -906,6 +906,53 @@ def test_block_sdt_tables_are_extracted():
     assert phase_2_idx < table_idxs[1]
 
 
+def test_table_cell_keeps_text_of_wrapped_runs(tmp_path):
+    """A plain table cell keeps runs nested in inline wrappers.
+
+    Body paragraphs already read the runs inside content controls, simple
+    fields, smart tags and tracked insertions; a plain cell dropped them, so a
+    form filled in with content controls came out with empty answers.
+    """
+    w = 'xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main"'
+    wrapped_runs = {
+        "Name": (
+            f"<w:sdt {w}><w:sdtPr/><w:sdtContent>"
+            "<w:r><w:t>Jane Doe</w:t></w:r></w:sdtContent></w:sdt>"
+        ),
+        "Customer": (
+            f'<w:fldSimple {w} w:instr=" MERGEFIELD Customer ">'
+            "<w:r><w:t>ACME Corp</w:t></w:r></w:fldSimple>"
+        ),
+        "City": (
+            f'<w:smartTag {w} w:uri="urn:schemas-microsoft-com:office:smarttags" '
+            'w:element="City"><w:r><w:t>Paris</w:t></w:r></w:smartTag>'
+        ),
+        "Amount": (
+            f'<w:ins {w} w:id="1" w:author="Reviewer">'
+            "<w:r><w:t>1200</w:t></w:r></w:ins>"
+        ),
+    }
+
+    document = Document()
+    table = document.add_table(rows=len(wrapped_runs), cols=2)
+    for row_idx, (label, wrapped_run) in enumerate(wrapped_runs.items()):
+        table.cell(row_idx, 0).text = label
+        table.cell(row_idx, 1).paragraphs[0]._p.append(etree.fromstring(wrapped_run))
+
+    doc = _convert_built(document, tmp_path)
+
+    assert [cell.text for cell in doc.tables[0].data.table_cells] == [
+        "Name",
+        "Jane Doe",
+        "Customer",
+        "ACME Corp",
+        "City",
+        "Paris",
+        "Amount",
+        "1200",
+    ]
+
+
 def _table_with_grid_before(
     tmp_path,
     *,