[Bug]: QA chunk preview returns HTTP 500 (IndexError) for sources with zero chunks #45087

Self Checks

• I have read the Contributing Guide and Language Policy.
• This is a bug report, not a usage question.
• I have searched for existing issues, including closed ones.
• I confirm I am using English to submit this report.

Dify version

main at e5acabc3d (2026-10-29)

Cloud or Self Hosted

Self Hosted (Source)

What happened?

In QA chunk mode (doc_form="qa_model"), the chunk preview / indexing-estimate endpoint returns HTTP 500 for any source that yields zero chunks, e.g. a whitespace-only text file or a blank/scanned PDF without extractable text.

api/core/rag/index_processor/processor/qa_index_processor.py:97-104 (preview branch) indexes all_documents[0] without checking that the list is non-empty:

extracted: ["'   \n\n   \n'"]
QA preview: IndexError: list index out of range
Paragraph preview: OK -> []

At the estimate adapter (api/services/knowledge/indexing/adapters/estimate.py:120-129) the same failure is wrapped as IndexingEstimateExecutionError: list index out of range, which the console controllers map to IndexingEstimateError → HTTP 500 Knowledge indexing estimate failed: list index out of range (controllers/console/datasets/error.py:57-60).

Non-preview QA indexing of the same empty source returns [] correctly, and paragraph-mode preview returns an empty preview list, so the failure is specific to QA preview/estimate of an empty document.

Steps To Reproduce

1. Create a dataset with QA chunk mode (doc_form="qa_model").
2. Preview/estimate indexing for a whitespace-only .txt or a blank PDF (or an image-only PDF whose images are not persisted, e.g. URL-sourced).
3. Observe HTTP 500 with list index out of range instead of an empty preview.

A self-contained script using the real TextExtractor / PdfExtractor and QAIndexProcessor.transform(..., preview=True) reproduces it without services (repro_qa_empty_preview.py, repro_qa_blank_pdf.py, qa_estimate_empty.py); happy to attach them.

Expected behavior

200 with an empty preview for a zero-chunk document: preview=[], qa_preview=[], total_segments=0, consistent with paragraph mode and with non-preview QA indexing.

Suggested fix

Guard the preview branch, e.g.:

if preview:
    if not all_documents:
        return {"preview": [], "qa_preview": [], "total_segments": 0}
    ...

and add a regression test for empty/whitespace input in QA preview. Searched open and closed issues/PRs (EN and ZH) for QA preview / empty chunks / IndexError in knowledge indexing: no duplicate; open PR #34236 (format_preview KeyError) and merged #42205 (preview cap) are different.