[JarryShaw]

• Searched for similar pull requests
• Followed the coding style (make pylint, make mypy, make isort)
• make test passes, and a test case covers the change
• Added a changelog entry — N/A — centralised in docs(changelog): shared 1.5.0 changelog — long-lived, merges last (#610, #616, #617, #618, #620) #657

What is the purpose of your pull request?

• perf — performance improvements

---

Description of your pull request and other information

Closes #2110. pcap_ct extraction was about 2.4× slower on 3.15. The cause is pcapkit's installed_distributions() in _pcap_backend.py, which runs three times per extraction. On 3.15, importlib.metadata.distribution() → Distribution.from_name wraps its search in _prefer_valid, which parses every candidate's METADATA (importlib_metadata #489). One call costs 115 µs on 3.14 and 397 µs on 3.15. libpcap 1.10.3, pcap-ct 1.3.0b3 and a ctypes call cost the same on both versions.

• Fix: next(iter(importlib.metadata.distributions(name=name)), None) matches on the dist-info directory and reads no file. It finds a package in exactly the cases the old call did; checked on 3.8.20 and 3.15.1 for pcap-ct, pcap_ct, pypcap and pip. The probe() docstring no longer claims a repeat call costs only a sys.modules lookup.

• Measured (Docker images from docs: regenerate the engine benchmark with Python 3.15 measured #2109's harness, 6-packet in.pcap):

  Python	Before (ms/packet)	After (ms/packet)
  3.15.1	0.2617	0.1066
  3.14.7	0.1097	0.1138 (within noise)

• Test: puts an empty pcap_ct dist-info on sys.path and makes any METADATA read raise. Without the fix it fails on both 3.14 and 3.15. The backend, pcap_ct and pypcap engine tests pass on both.

Follow-up: the Extractor still probes three times per extraction (_select_engine calls unsupported_reason twice, plus the constructor). Tracked separately.

[JarryShaw]

GOOD TO GO at b90787270 (Sonnet review).

• Semantics: read in CPython source for 3.8, 3.10 and 3.15. distribution(name) and distributions(name=...) share one lookup and one name normalisation (lowercase, -→_). On 3.15 they differ only in _prefer_valid, which reorders candidates but still returns a dist without METADATA. So found and not-found results are identical, and egg-info and editable installs take the same path. The caller only records the name.
• Attribution: 3.15's _prefer_valid reads .metadata on every candidate. Its docstring cites "importlib_resources#489", most likely meaning importlib_metadata fix(ipv6-route): compute Hdr Ext Len in 8-octet units on both sides #489, whose thread matches.
• Tests:
  • With _pcap_backend.py reverted, the new test fails.
  • test_pcap_backend (14), the pcap_ct and pypcap engine tests and test_core pass.
  • The 10k-iteration benchmark on 3.14 shows no regression (137.6 µs before, 135.8 µs after).
• UNVERIFIED here: the 3.15 speed-up itself, which needs Docker or a 3.15 interpreter; the author measured 0.2617 → 0.1066 ms/packet.

[github-actions]

Coverage: 91.54% (unit tier, Python 3.15, b90787270, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	2	95.16%
pcapkit/const	19599	880	2618	866	92.12%
pcapkit/corekit	2565	71	812	37	96.56%
pcapkit/dumpkit	282	3	102	3	98.44%
pcapkit/foundation	2929	23	1024	25	98.79%
pcapkit/interface	138	4	52	3	96.32%
pcapkit/protocols	17414	219	4574	185	98.14%
pcapkit/toolkit	869	36	298	10	95.20%
pcapkit/utilities	473	7	136	5	98.03%
pcapkit/vendor	4432	1988	1018	167	51.87%

Per-file detail: the coverage-html artifact of this run.

[JarryShaw]

Ready to merge at b90787270.

• Review: the Sonnet review is GOOD TO GO.
• CI: every blocking check passed (Required checks passed SUCCESS). The 3.8–3.10 KTLO legs carry no warnings.
• Ordering: the tests/foundation plain-unittest ordering leg passed on a scratch merge into main: 714 tests, 0 failures.

This closes #2110. Once it merges, I'll re-measure the 3.15 pcap_ct cell in #2109's table.