MsWordDocumentBackend crashes on .docx files with external image references (TargetMode="External")

[rongo-ms]

Bug

MsWordDocumentBackend._handle_pictures() crashes when processing .docx files that contain image relationships with TargetMode="External" (e.g., images referencing URLs or local file:/// paths instead of embedded in word/media/).

The crash occurs because python-docx's _Relationship.target_part property raises ValueError when accessed on an external relationship, and Docling's code doesn't guard against this.

Error:

ValueError: target_part property on _Relationship is undefined when target mode is External

Docling async API returns:

{
  "status": "failure",
  "errors": [{
    "component_type": "pipeline",
    "module_name": "SimplePipeline",
    "error_message": "target_part property on _Relationship is undefined when target mode is External"
  }],
  "document": {"json_content": null, "md_content": null}
}

Root cause — in docling/backend/msword_backend.py, _handle_pictures() → get_docx_image():

if rId in self.docx_obj.part.rels:
    image_part = self.docx_obj.part.rels[rId].target_part  # CRASHES on external rels

The relationship exists but is external. python-docx's _Relationship.target_part raises ValueError when _is_external is True.

Suggested fix — check rel.is_external before accessing target_part:

def get_docx_image(image):
    image_data = None
    rId = image.get("{http://schemas.openxmlformats.org/officeDocument/2006/relationships}embed")
    if rId in self.docx_obj.part.rels:
        rel = self.docx_obj.part.rels[rId]
        if rel.is_external:  # Skip external image references
            return None
        image_part = rel.target_part
        image_data = image_part.blob
    return image_data

This would cause external images to fall through to the existing if image_data is None: _log.warning("Warning: image cannot be found") path, which already handles missing images gracefully.

Steps to reproduce

1. Obtain a .docx file with external image references (common when saving Wikipedia pages as .docx via a browser). The word/_rels/document.xml.rels will contain entries like:

<Relationship Id="rId139"
  Type="http://schemas.openxmlformats.org/officeDocument/2006/relationships/image"
  Target="file:///C:\path\to\image.png"
  TargetMode="External"/>

2. Convert with Docling:

from docling.document_converter import DocumentConverter
from docling.datamodel.base_models import InputFormat

converter = DocumentConverter(allowed_formats=[InputFormat.DOCX])
result = converter.convert("file_with_external_images.docx")
# Raises: ConversionError: Input document ... is not valid.

Or via the async API:

# Submit to /v1/convert/source/async, poll until "success", then GET /v1/result/{task_id}
# Result: {"status": "failure", "errors": [{"error_message": "target_part property..."}]}

Docling version

docling 2.76.0

Python version

Python 3.12

python-docx version: 1.2.0

[rongo-ms]

Opened a PR with a suggested fix: #3114