_get_body_paragraphs checks whether a part is present, not whether it produced text, then returns unconditionally, at docling/backend/email_backend.py:270:

if self.mail.text_plain:
    paragraphs: list[str] = []
    for part in self.mail.text_plain:
        paragraphs.extend(self._split_paragraphs(part))
    return paragraphs

A multipart/alternative message can carry a whitespace-only text/plain part beside a real text/html one, and many senders generate exactly that. mail.text_plain is then a truthy [' '], _split_paragraphs(" ") gives [], and the branch returns that empty list instead of falling through to the HTML one.

The whole body is dropped, and the conversion still reports success.

Before

A two-part message whose plain part is a single space, converted through DocumentConverter.convert():

# Blank Plain Part

From: Alice <[email protected]>

To: Bob <[email protected]>

After

The same message, with the body present:

# Blank Plain Part

From: Alice <[email protected]>

To: Bob <[email protected]>

Real content that should appear.

Each branch now returns only once it has produced a paragraph. The final self.mail.body fallback stays as it was.

Tests

$ python -m pytest tests/test_backend_email.py -q
23 passed

$ python -m ruff check docling/backend/email_backend.py tests/test_backend_email.py
All checks passed!
$ python -m ruff format --check docling/backend/email_backend.py tests/test_backend_email.py
2 files already formatted

One new case builds the message as raw bytes and converts it through DocumentConverter.convert(), the path a user takes.

Revert email_backend.py alone, keep the case, and the file gives 1 failed, 22 passed.

No groundtruth changed. No committed .eml fixture carries a blank alternative part, so the suite never reached this line.

Scope

A message whose plain part holds real text behaves as before: the branch still returns on the first paragraph it finds. Only the case where a branch produced nothing changes.

Duplicate search

Two open pull requests touch docling/backend/email_backend.py: #3978 and #3791. Neither mentions _get_body_paragraphs or text_plain.

Closed #4259 touched this file for a different concern, adding isinstance(part, str) guards against non-string parts. Its diff leaves this control flow alone, so a blank string part still reproduces on top of it.

Checklist:

• Documentation has been updated, if necessary.
  No document covers which part supplies the body.
• Examples have been added, if necessary.
  The change corrects existing output rather than adding a capability.
• Tests have been added, if necessary.