From 6d1078b5385dd88d1ecbca23b381961be8f9feec Mon Sep 17 00:00:00 2001
From: Jay Xu <mustangxu@users.noreply.github.com>
Date: Thu, 14 Aug 2025 12:14:03 +0800
Subject: [PATCH] fix 'KeyError: "There is no item named 'word/NULL' in the
 archive"' (#9455)

### What problem does this PR solve?

Issue referring to:
https://github.com/python-openxml/python-docx/issues/797
Fix referring to:
https://github.com/python-openxml/python-docx/issues/1105#issuecomment-1298075246

### Type of change

- [x] Bug Fix (non-breaking change which fixes an issue)
---
 rag/app/naive.py | 23 ++++++++++++++++++++++-
 1 file changed, 22 insertions(+), 1 deletion(-)

diff --git a/rag/app/naive.py b/rag/app/naive.py
index 1a6c0b92b..5e946d1c8 100644
--- a/rag/app/naive.py
+++ b/rag/app/naive.py
@@ -22,6 +22,8 @@ from timeit import default_timer as timer
 
 from docx import Document
 from docx.image.exceptions import InvalidImageStreamError, UnexpectedEndOfFileError, UnrecognizedImageError
+from docx.opc.pkgreader import _SerializedRelationships, _SerializedRelationship
+from docx.opc.oxml import parse_xml
 from markdown import markdown
 from PIL import Image
 from tika import parser
@@ -47,8 +49,8 @@ class Docx(DocxParser):
         if not embed:
             return None
         embed = embed[0]
-        related_part = document.part.related_parts[embed]
         try:
+            related_part = document.part.related_parts[embed]
             image_blob = related_part.image.blob
         except UnrecognizedImageError:
             logging.info("Unrecognized image format. Skipping image.")
@@ -62,6 +64,9 @@ class Docx(DocxParser):
         except UnicodeDecodeError:
             logging.info("The recognized image stream appears to be corrupted. Skipping image.")
             return None
+        except Exception:
+            logging.info("The recognized image stream appears to be corrupted. Skipping image.")
+            return None
         try:
             image = Image.open(BytesIO(image_blob)).convert('RGB')
             return image
@@ -360,6 +365,20 @@ class Markdown(MarkdownParser):
             tbls.append(((None, markdown(table, extensions=['markdown.extensions.tables'])), ""))
         return sections, tbls
 
+def load_from_xml_v2(baseURI, rels_item_xml):
+    """
+    Return |_SerializedRelationships| instance loaded with the
+    relationships contained in *rels_item_xml*. Returns an empty
+    collection if *rels_item_xml* is |None|.
+    """
+    srels = _SerializedRelationships()
+    if rels_item_xml is not None:
+        rels_elm = parse_xml(rels_item_xml)
+        for rel_elm in rels_elm.Relationship_lst:
+            if rel_elm.target_ref in ('../NULL', 'NULL'):
+                continue
+            srels._srels.append(_SerializedRelationship(baseURI, rel_elm))
+    return srels
 
 def chunk(filename, binary=None, from_page=0, to_page=100000,
           lang="Chinese", callback=None, **kwargs):
@@ -391,6 +410,8 @@ def chunk(filename, binary=None, from_page=0, to_page=100000,
         except Exception:
             vision_model = None
 
+        # fix "There is no item named 'word/NULL' in the archive", referring to https://github.com/python-openxml/python-docx/issues/1105#issuecomment-1298075246
+        _SerializedRelationships.load_from_xml = load_from_xml_v2
         sections, tables = Docx()(filename, binary)
 
         if vision_model: