From f4c766cd178248e188312b9acd50bca372749528 Mon Sep 17 00:00:00 2001
From: chriscoveries <212684707+chriscoveries@users.noreply.github.com>
Date: Sun, 4 Oct 2026 22:03:35 +0700
Subject: [PATCH] fix(docx): skip non-element XML nodes

Signed-off-by: chriscoveries <212684707+chriscoveries@users.noreply.github.com>
---
 docling/backend/msword_backend.py |  2 ++
 tests/test_backend_msword.py      | 22 ++++++++++++++++++++++
 2 files changed, 24 insertions(+)

diff --git a/docling/backend/msword_backend.py b/docling/backend/msword_backend.py
index dfcb50c99c..5c11e5fafb 100644
--- a/docling/backend/msword_backend.py
+++ b/docling/backend/msword_backend.py
@@ -1105,6 +1105,8 @@ def _walk_linear(
     ) -> tuple[DoclingDocument, list[RefItem]]:
         added_elements = []
         for element in body:
+            if not isinstance(element.tag, str):
+                continue
             tag_name = etree.QName(element).localname
             # Check for Inline Images (blip elements)
             _raw_drawing_blip = self.blip_xpath_expr(element)
diff --git a/tests/test_backend_msword.py b/tests/test_backend_msword.py
index 05a4a63d2b..d0a47b234c 100644
--- a/tests/test_backend_msword.py
+++ b/tests/test_backend_msword.py
@@ -1472,6 +1472,28 @@ def test_malformed_hyperlink_does_not_abort_conversion(tmp_path):
     assert hyperlinks == []
 
 
+def test_non_element_nodes_do_not_abort_conversion(tmp_path):
+    doc = Document()
+    doc.add_paragraph("Before metadata.")
+    doc.element.body.insert(1, etree.Comment("synthetic comment"))
+    doc.element.body.insert(
+        2, etree.ProcessingInstruction("synthetic", 'value="metadata"')
+    )
+    doc.add_paragraph("After metadata.")
+
+    docx_path = tmp_path / "non_element_nodes.docx"
+    doc.save(docx_path)
+
+    in_doc = InputDocument(
+        path_or_stream=docx_path,
+        format=InputFormat.DOCX,
+        backend=MsWordDocumentBackend,
+    )
+    result = in_doc._backend.convert()
+
+    assert result.export_to_markdown() == "Before metadata.\n\nAfter metadata."
+
+
 def test_trailing_whitespace_run_keeps_paragraph_formatting(tmp_path):
     """A whitespace-only trailing run must not overwrite the paragraph's formatting.