IndexError in prepare_element for PictureItem with non-empty prov but no embedded image (charts in PPTX)

[kickrunner]

Bug

Follow-up to #2390 / #2401. The fix in #2401 only guards the case where a PictureItem has no embedded image and empty provenance (len(element.prov) == 0). It does not cover the case of a PictureItem that has provenance but no embedded image — which is exactly what happens for native chart objects in PPTX (and presumably DOCX/XLSX) files.

In that case, BaseItemAndImageEnrichmentModel.prepare_element still falls through to the page-cropping branch:

page_ix = element_prov.page_no - conv_res.pages[0].page_no
cropped_image = conv_res.pages[page_ix].get_image(...)

Since SimplePipeline (used for PPTX/DOCX/HTML/XLSX) never populates conv_res.pages, this raises IndexError: list index out of range, which BasePipeline.execute wraps into RuntimeError: Pipeline SimplePipeline failed, failing the entire conversion.

Steps to reproduce

Minimal repro — a PPTX with a single native chart shape (no embedded raster image), converted with do_picture_classification=True (also reproducible with do_picture_description=True):

from docling.document_converter import DocumentConverter, PowerpointFormatOption
from docling.datamodel.base_models import InputFormat
from docling.datamodel.pipeline_options import ConvertPipelineOptions
from docling.pipeline.simple_pipeline import SimplePipeline

opts = ConvertPipelineOptions()
opts.do_picture_classification = True

fmt = {InputFormat.PPTX: PowerpointFormatOption(pipeline_cls=SimplePipeline, pipeline_options=opts)}
converter = DocumentConverter(format_options=fmt)
result = converter.convert("minimal_chart_repro.pptx", raises_on_error=True)

Attached: minimal_chart_repro.pptx — generated with python-pptx, containing a single pie chart shape and no other content.

Error Output

Traceback (most recent call last):
  ...
  File ".../docling/models/base_model.py", line 204, in prepare_element
    page_ix = element_prov.page_no - conv_res.pages[0].page_no
                                     ~~~~~~~~~~~~~~^^^
IndexError: list index out of range

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  ...
  File ".../docling/pipeline/base_pipeline.py", line 98, in execute
    raise RuntimeError(f"Pipeline {self.__class__.__name__} failed") from e
RuntimeError: Pipeline SimplePipeline failed

Root cause

DocumentPictureClassifier.is_processable (and other BaseItemAndImageEnrichmentModel subclasses) accept any PictureItem. For chart shapes converted from PPTX, Docling assigns non-empty prov (page number + bbox), but there is no embedded image, since chart shapes are native XML chart objects, not raster images. The #2401 fix only skips elements with empty prov; elements with non-empty prov but no image still hit the conv_res.pages[0] crop path, which is empty for SimplePipeline.

Suggested fix

Guard on conv_res.pages being empty (or check isinstance(self, ...) for pipelines that never populate pages) before attempting the page crop, in addition to the existing embedded-image check, e.g.:

if isinstance(element, PictureItem):
    embedded_im = element.get_image(conv_res.document)
    if embedded_im is not None:
        return ItemAndImageEnrichmentElement(item=element, image=embedded_im)
    elif len(element.prov) == 0 or not conv_res.pages:
        return None

Docling version

docling-slim: 2.114.0

Python version

Python 3.12

[kickrunner]

For convenience, here's the exact script used to generate the minimal repro PPTX (python-pptx==1.0.2), in case attaching the binary file doesn't come through:

from pptx import Presentation
from pptx.chart.data import CategoryChartData
from pptx.enum.chart import XL_CHART_TYPE
from pptx.util import Inches

prs = Presentation()
slide = prs.slides.add_slide(prs.slide_layouts[5])
slide.shapes.title.text = "Minimal chart reproduction"

chart_data = CategoryChartData()
chart_data.categories = ["Category A", "Category B", "Category C"]
chart_data.add_series("Series 1", (1, 2, 3))

x, y, cx, cy = Inches(1), Inches(1.5), Inches(6), Inches(4)
slide.shapes.add_chart(XL_CHART_TYPE.PIE, x, y, cx, cy, chart_data)

prs.save("minimal_chart_repro.pptx")

Running the repro snippet from the issue against the resulting file reliably raises the IndexError → RuntimeError: Pipeline SimplePipeline failed.

[dosubot]

Hi @kickrunner! I'm Dosu and I'm helping the IBM Docling team.

I can confirm the bug and the root cause analysis. Looking at the current prepare_element in base_model.py, the flow for a PictureItem with non-empty prov but no embedded image (your chart scenario) is:

1. get_image(conv_res.document) returns None ✓
2. len(element.prov) == 0 is False (chart shapes get provenance from the slide) → does not return None early ✓
3. Falls through to conv_res.pages[0].page_no → 💥 IndexError because SimplePipeline never populates conv_res.pages [1]

Your suggested fix is correct — adding or not conv_res.pages to the early-return guard on line 194 would catch this case cleanly. Alternatively the guard could be placed right before the page-cropping block (line 197+) to protect all element types, not just PictureItem:

if not conv_res.pages:
    return None

I didn't find an existing PR addressing this specific gap (#3867). The earlier PR #2401 only covered the empty-prov case, as you noted.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.

[simonschoe]

+1, a hottix would be super valuable!