[charloteberlein]

Summary of the issue

When calling set_metadata({}) (or scrub()) on xref-stream PDFs, xref_set_key() does not correctly apply the update to the trailer, and so the metadata is not cleared. Fetching the trailer again before copying the updated dictionary prevents this.

This is a one-line addition in src/__init__.py:

obj = mupdf.pdf_trailer(pdf)

Regression tests and how to reproduce the error

The regression tests I’ve written minimally reproduce the error. One checks set_metadata({}) directly, and the other checks it through scrub() (which calls set_metadata({})).

Running the new tests with python -m pytest -k xref_stream ./tests:

• With the change reverted, they fail:

====================================== short test summary info ======================================
FAILED tests/test_metadata.py::test_clear_metadata_xref_stream - AssertionError: assert not 'Title'
FAILED tests/test_metadata.py::test_scrub_xref_stream - AssertionError: assert not 'Title'
================================= 2 failed, 524 deselected in 0.47s =================================

• With the change applied, they both pass:

tests/test_metadata.py ..                                                                     [100%]

================================= 2 passed, 524 deselected in 0.47s =================================

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[charloteberlein]

I have read the CLA Document and I hereby sign the CLA

[JorjMcKie]

Thanks for spotting this - well done "first contribution" 😎!