Fix bug and improve detection of incorrect headers

Implements changes suggested by: #17

Fix minor bugs

Fixes:

• #13
• #14

Fix minor bugs

Fixes

• #7 ValueError when parsing arabic numbering
• #8 Don't try to process headings if none were detected

Fix: Source DocumentStreams + Error handling

Makes it possible to pass the file path of the source file or a stream to ResultPostprocessor in order to be read by pymupdf for metadata extraction:

If you run into the PDFFileNotFoundException then your source attribute to DocumentConverter().convert(source=source) has either been of type str or of type DocumentStream so there is the Docling conversion result unfortunately does not hold a valid reference to the source file anymore. Hence the Postprocessor needs your help - if source was a string then you can add the source=source when instantiating ResultPostprocessor - full example:

from docling.document_converter import DocumentConverter
from hierarchical.postprocessor import ResultPostprocessor

source = "my_file.pdf"  # document per local path or URL
converter = DocumentConverter()
result = converter.convert(source)
# the postprocessor modifies the result.document in place.
ResultPostprocessor(result, source=source).process()
# ...

If you have used a DocumentStream object as source you are unfortunately in the situation that you will have to pass a valid Path to the PDF as a source argument to ResultPostprocessor or a new, open BytesIO stream or DocumentStream object as a source argument to ResultPostprocessor. The reason is that docling closes the source stream when it is finished - so no more reading from that stream is possible.

Use also PDF-metadata ToC

New in this release:

• use pymupdf to read ToC from pdf (if it exists in the pdf metadata)
• correct header levels and hierarchy based on this
• best effort attempt to:
  • convert texts and list items to headers if they were parsed incorrectly and appear in the ToC
  • convert header to text items if they were parsed incorrectly and do not appear in the ToC

v0.0.2

Fixes in documentation etc.

Initial minimal release

Initial minimal release to give access to hierarchy inference.