fix: keep key= distinct from a valueless key in TXT records - #1813

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.81%. Comparing base (df947b2) to head (28edc80).
⚠️ Report is 1 commits behind head on master.

@@            Coverage Diff             @@
##           master    #1813      +/-   ##
==========================================
- Coverage   99.81%   99.81%   -0.01%     
==========================================
  Files          33       33              
  Lines        3760     3759       -1     
  Branches      533      533              
==========================================
- Hits         3753     3752       -1     
  Misses          5        5              
  Partials        2        2              

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

Merging this PR will not alter performance

✅ 21 untouched benchmarks

_{Comparing bboe:fix/txt-empty-value (28edc80) with master (df947b2)}

@bluetoothbot review

Previous review — superseded by a newer review below.

@bdraco: @bluetoothbot review

Re-reviewed. Head unchanged at 3f555e7. Fix still correct: sep branch inverts _set_properties exactly. Four nits carry over unresolved — stale key_sep_value=tuple at info.pxd:89, regression history in the info.py:468 comment, test docstring body, and the "decoded properties" comment at test_info.py:681 that passes info.properties. All non-blocking. Approving.

note that CI is red until #1820 but thats waiting for PSR

note that CI is red until #1820 but thats waiting for PSR

Thanks! I was briefly wondering what I might have broke.

@bdraco: note that CI is red until #1820 but thats wai...

Branch now carries that fix. Merge commit 28edc80 pulls master, including df947b2 (#1820). CI should clear on rerun.

PR Review — fix: keep key= distinct from a valueless key in TXT records

Merge-ready — all four prior nits are now resolved and the fix is verified correct end to end.

What's solid:

• info.py:467-471 branches on the partition separator, which is the exact inverse of _set_properties (info.py:416-420): a b"" value encodes as key=, None encodes as a bare key. This is the only partition(b"=") site in the tree (git grep), so the blast radius is contained.
• info.pxd:89-96 now declares key/sep/value as bytes and drops the stale key_sep_value=tuple — the .pxd matches the .py in the same commit, per CLAUDE.md's build convention.
• The comment is trimmed to the RFC 6763 §6.4 invariant, with the regression history left to the PR body.
• The test docstring is a single line naming what it pins, and the round trip now loops over both info.properties and info.decoded_properties — the str form exercises the properties_contain_str encode path, where "" could plausibly have been re-collapsed to a bare key.

Notes, nothing blocking:

• Downstream-visible decode change: key= now yields b"" / "" instead of None. That is the stated intent and restores pre-0.80.0 behaviour, but consumers doing props.get(b"x") is None will see it — worth a line in the release notes.
• Pre-existing and untouched: an empty key (b"=") is still stored rather than ignored per RFC 6763 §6.4. The diff only changes what it stores (None -> b""), and the round trip still reproduces b"=", so no regression.
• 28edc80 merges master, which carries the cython 3.3.0 build fix (fix: build with cython 3.3.0 #1820), so the earlier red CI should clear on the next run.

✅ Resolved since last review (2)

Checklist

• Decode is the inverse of encode (round trip verified against _set_properties)
• No regression for existing None-valued properties
• .pxd locals match the changed function signature
• Comment/docstring conventions per CLAUDE.md
• New behaviour covered by tests (bytes and str APIs, both encode paths)
• Test isolation and teardown
• Diff matches PR description, no scope creep
• No security-relevant surface (no injection, secrets, deserialization, or path handling)

Automated review by Kōan (Claude) HEAD=28edc80 49s