Investigate why DoclingDocument.load_from_doctags requires image for bbox extraction

[cau-git]

There are references in the code such as this: https://github.com/docling-project/docling-core/blob/main/docling_core/types/doc/document.py#L5779

Where the extract_bounding_box method is called only if image is not None, which appears to make no sense, since it works without that constraint in other call sites.

E.g. when parsing doctags that include list items, the list_item provenance is not populated.

[dosubot]

The image requirement for bounding box extraction in DoclingDocument.load_from_doctags is inconsistent across different branches of the method. For list items, bounding box extraction and provenance creation are only performed if an image is present, which means list item provenance is not populated when no image is provided—even though the extract_bounding_box function itself does not require an image and can work with just the doctags. Other branches, like text and key-value items, use bounding box extraction regardless of image availability by scaling against fallback dimensions, making the list item logic an outlier. This guard is the reason list items lose provenance when parsing doctags without images, even though the bounding box could still be extracted from the tags alone. You can see this logic in the code here.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}