Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.43%. Comparing base (da867f4) to head (ae6a45d).
⚠️ Report is 1 commits behind head on main.

@@           Coverage Diff           @@
##             main    #3687   +/-   ##
=======================================
  Coverage   97.43%   97.43%           
=======================================
  Files          55       55           
  Lines       10003    10005    +2     
  Branches     1837     1838    +1     
=======================================
+ Hits         9746     9748    +2     
  Misses        149      149           
  Partials      108      108           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Followed up on the review:

• reverted the unrelated tests/__init__.py changes so this PR is back to the link-pairing fix plus its regression test
• switched the filtering back to comprehensions
• kept the pairing change as zip(filtered_new_links, filtered_old_links)

Why this fixes the mismatch issue: the old code zipped the raw annotation-derived lists first and only filtered out non-links afterwards. If one side had an extra non-link annotation, the zip alignment could shift or drop a valid link pair. Filtering each side before zipping preserves the relative order of actual link annotations and avoids that offset.

Validation:

• .venv/bin/python -m pytest tests/generic/test_link.py -q
• ruff check pypdf/generic/_link.py tests/generic/test_link.py tests/__init__.py

I just tested with #3667 (comment) and it seems like there is no change before and after this PR. In which real-world example would I see a difference now?

If I understand the code correctly, the condition/logging could be enabled again if moved after the filtering. Then we would see a real effect for the aforementioned example as the URI-based link annotations would be correctly filtered out now, thus having matching sizes afterwards.

By the way: Are there any benefits when using

    new_links = [
        link
        for annotation in new_annotations
        if (link := _build_link(annotation, new_page)) is not None
    ]

instead of

    new_links = list(filter(None, (_build_link(link, new_page) for link in new_annotations)))

as initially proposed?

Followed up in the latest push.

You were right that the previous revision did not visibly change the issue example yet, because I had only changed the pairing logic and had not restored the size check after filtering. I have now moved the size warning to the filtered-link lists instead.

That means there is now a real observable difference for a case like the issue discussion: if one side contains an extra URI / non-GoTo annotation, it gets filtered out first, so we no longer warn just because the raw annotation arrays differ while the actual internal links still match.

A concrete example where behavior changes now:

• old page annotations: [GoTo]
• new page annotations: [URI, GoTo]

Before this change, zipping the raw annotation-derived lists could drop the valid GoTo pair entirely. Now the URI annotation is ignored first, the GoTo links are paired correctly, and no mismatch warning is emitted for that case.

On the comprehension form: there is no strong requirement for the walrus form, but I chose it because is not None is explicit and avoids relying on truthiness. If you prefer the filter(None, ...) variant for style consistency, I can switch that too.

Validation:

• .venv/bin/python -m pytest tests/generic/test_link.py -q
• ruff check pypdf/generic/_link.py tests/generic/test_link.py

The remaining macOS CI failure looks unrelated to the link-pairing changes in this PR.

I checked the failing job and it is currently failing in tests/test_cmap.py::test_text_extraction_fast[...] because tests/pdf_cache/ASurveyofImageClassificationBasedTechniques.pdf is missing on the runner (FileNotFoundError). The link-specific tests added/updated here are passing locally, and the other CI jobs for this PR are green.

I can follow up further on the PR code if needed, but this specific macOS failure does not appear to come from the changes in pypdf/generic/_link.py.

Latest CI rerun is green now, including macOS. Together with the earlier follow-up that moved the size warning after filtering and added the URI/non-GoTo regression case, the current head should now be ready for another look when you have a chance.