feat: integrate docling for document data extraction

[Abdeali099]

• https://www.docling.ai/

• Implement docling support

• User can selected PDF Processor from Transaction Parser Settings

• To add a new PDF Processor add it too hook and options of settings

Note

OCR currently disable. Will enable in future with proper package

• no-docs

[vorasmit]

@greptileai

[greptile-apps]

Confidence Score: 4/5

• This PR is safe to merge; the substantive issues from the previous review cycle have all been addressed, with only a minor type annotation inaccuracy remaining.
• The core logic is sound: lazy imports, proper stream management, converter caching, and idempotent patching are all correctly implemented. The one remaining new finding is a misleading str | None return type on FileProcessor.get_content — a cosmetic type annotation issue that has no runtime impact. The two known accepted trade-offs (hard docling dependency, hardcoded dropdown options) are deliberate product decisions. No critical logic or security issues were found.
• No files require special attention beyond the minor type annotation fix in transaction_parser/transaction_parser/utils/file_processor.py.

Important Files Changed

Filename	Overview
transaction_parser/transaction_parser/utils/pdf_processor.py	New file introducing PDFProcessor ABC, DoclingPDFProcessor, OCRMyPDFProcessor, and the get_pdf_processor factory. Most issues flagged in the previous review cycle (lazy imports, stream seek resets, InputFormat enum key, converter caching, status checks) have been properly addressed in this revision.
transaction_parser/transaction_parser/utils/file_processor.py	Refactored to delegate PDF processing to the new PDFProcessor strategy. PDF/spreadsheet logic is now cleanly separated. Minor: get_content is annotated `str
transaction_parser/patches/set_default_pdf_processor.py	Idempotent patch that sets pdf_processor to OCRMyPDF only when no value is currently stored, preventing the unconditional overwrite flagged in the prior review.
transaction_parser/patches/init.py	Empty __init__.py added to make transaction_parser.patches a proper Python package, fixing the previously reported ModuleNotFoundError during bench migrate.
transaction_parser/transaction_parser/ai_integration/parser.py	Tightened get_content return type from `dict
transaction_parser/hooks.py	Adds pdf_processors hook dict mapping processor names to class paths — clean extension point for third-party apps.
pyproject.toml	Adds docling>=2.75.0 as a mandatory install-time dependency. The developer has opted to keep it as a hard dependency (the optional-extras approach was discussed in the previous review and declined).
transaction_parser/transaction_parser/doctype/transaction_parser_settings/transaction_parser_settings.json	Adds pdf_processor Select field with hardcoded OCRMyPDF\nDocling options. The extensibility limitation (hook-registered processors not appearing in the dropdown) is a known accepted trade-off to be addressed in a follow-up PR via Property Setter.
transaction_parser/transaction_parser/init.py	Adds now=frappe.conf.developer_mode to the background job enqueue call so that parsing runs synchronously in development. Deliberate choice documented and accepted by the team.
transaction_parser/patches.txt	Registers the new set_default_pdf_processor patch under [post_model_sync]. The #2 comment on the prior patch and #1 here appear to be ordering notes, though this numbering scheme should be documented if it has semantic meaning.

_{Last reviewed commit: 9f02aaf}

[Abdeali099]

@greptileai

[Abdeali099]

@greptileai

[Abdeali099]

@greptileai

[Abdeali099]

@greptileai

[Abdeali099]

@greptileai

[Abdeali099]

@greptileai

[Abdeali099]

@greptileai

[Abdeali099]

@greptileai

[Abdeali099]

@greptileai

[Abdeali099]

@greptileai

[Abdeali099]

@greptileai

[Abdeali099]

@greptileai

[Abdeali099]

@greptileai