[devYRPauli]

Problem

DictionaryObject.read_from_stream() decides whether a key has already been seen with:

if not data.get(key):
    data[key] = value
else:
    # multiple definitions of key not permitted

The test checks truthiness, not presence. A first value of NumberObject(0) is falsy. On the second occurrence data.get(key) returns 0, so not 0 is True. The branch then treats the key as new. It does not report the duplicate, and the second value replaces the first.

This happens even with strict=True, where the reader is supposed to raise.

Behaviour on current main:

input	result
<< /Count 7 /Count 5 >>	warns, keeps 7
<< /Count 0 /Count 5 >>	silent, keeps 5

The second row is the bug. pypdf does not raise, and it does not log. It also keeps the opposite value from every other duplicate.

Change

if key not in data:

Presence, not truthiness. The falsy case now behaves like every other duplicate: strict=True raises, non-strict warns, and the first value is kept.

input	before	after
<< /Count 7 /Count 5 >>	warns, keeps 7	unchanged
<< /Count 0 /Count 5 >>	silent, keeps 5	warns, keeps 0
<< /Count 0 /Size 5 >>	both keys kept	unchanged

The non-strict result for the second row changes from 5 to 0. That change is the point. It now matches the first row, which has always kept the first value.

Tests

Four tests in tests/generic/test_data_structures.py, all from bytes in memory, no fixture download:

• duplicate key with a falsy first value raises under strict=True
• the same input under non-strict logs the message and keeps the first value
• duplicate key with a non-falsy first value still raises, unchanged
• two different keys where the first value is 0 parse normally, so there is no false positive

Reverting the one-line change fails the first two and leaves the last two passing.

tests/generic/ reports 90 passed with the change and 86 without it. One test, test_files.py::test_embedded_file__basic, fails either way in my environment because a fixture is missing locally.

ruff check passes on both files.

AI usage disclosure

Per the AI policy in CONTRIBUTING.md: I used Claude Code (Anthropic, Opus 5) while working on this. I reviewed the change, reproduced the behaviour myself before and after, and understand and stand behind it. Happy to explain any part of it or adjust the approach.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.00%. Comparing base (2955949) to head (706209c).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #4024   +/-   ##
=======================================
  Coverage   98.00%   98.00%           
=======================================
  Files          57       57           
  Lines       11128    11128           
  Branches     2087     2087           
=======================================
  Hits        10906    10906           
  Misses        124      124           
  Partials       98       98           

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[stefan6419846]

Thanks. I do not know why #69 implemented it like this, but your change seems reasonable.