Roadmap 2.2.0 — proof as the product

[EONRaider]

Tier 4 of the post-1.3.0 roadmap. Target version: 2.2.0 (minor).

The corpus and the property-based fuzzing are already unusual for this
space. This tier pushes them until the guarantees are testable claims
rather than adjectives.

#	Item
#97	Generalize the round-trip property to every protocol
#98	Nightly fuzzing with a moving seed
#99	Prove the browser claim with a Pyodide/WASM CI job
#100	A pcap/pcapng reader that takes bytes, not filenames

What each one closes

• Generalize the round-trip property to every protocol #97 — bytes(decode(x)) == x is Hypothesis-generated for only 4
  of 18 protocols; the other 14 rest on single-example assertions.
• Nightly fuzzing with a moving seed #98 — derandomize=True means CI has run the same 200 inputs
  on every build since the profile was written.
• Prove the browser claim with a Pyodide/WASM CI job #99 — scapy cannot be imported under Pyodide at all
  (unconditional from fcntl import ioctl on every import path). We
  can. Nothing tests it.
• A pcap/pcapng reader that takes bytes, not filenames #100 — no maintained, well-typed, pure-Python pcap+pcapng reader
  exists on PyPI. We already have one hidden in tests/conftest.py.

Note on ordering

All four are independent. #100 is the largest and the only one that
adds public API — it is also what makes memoryview walking in #88
worthwhile, since it creates the large-contiguous-buffer case where
zero-copy actually pays (measured: 1.8×, versus marginally slower for
a single small frame).

If capacity is tight, #99 is a few hours and closes a claim; #100 is
the one to schedule deliberately.

Part of the post-1.3.0 roadmap: #107

[EONRaider]

All four items landed, each in its own PR:

1. Generalize the round-trip property to every protocol #97 — Generalize the round-trip property to every protocol #129: bytes(decode(x)) == x is now Hypothesis-generated for all 18 protocols (up from 4), via tests/strategies.py + test_fuzz.py::TestGeneralizedRoundTrips. No src/ changes.
2. Nightly fuzzing with a moving seed #98 — Nightly fuzzing with a moving seed (#98) #141: a second "nightly" Hypothesis profile (10,000 examples, a real random seed each run) on a new scheduled workflow (.github/workflows/fuzz.yml, 03:00 UTC daily + manual dispatch), kept out of ci.yml deliberately so it can't fire the PR/release gate on a cron. Plus a targeted strategy building well-formed TCP SYN options (MSS/window-scale/SACK/timestamps), which the real corpus never captured.
3. Prove the browser claim with a Pyodide/WASM CI job #99 — Prove the browser claim under a real Pyodide runtime (#99) #140: a CI job that boots a real Pyodide (WebAssembly) runtime and decodes the whole real-capture corpus with the built wheel — not the cheaper sys.meta_path-blocklist alternative the issue also offered. That surfaced a genuine bug: IPv6 decode/encode relied on socket.inet_ntop/inet_pton(AF_INET6, ...), and Pyodide's CPython has AF_INET6 disabled. Fixed with a pure-Python replacement, differentially verified against glibc across 500,000+ addresses.
4. A pcap/pcapng reader that takes bytes, not filenames #100 — A pcap/pcapng reader that takes bytes, not filenames (#100) #142: netprotocols.pcap — read_captures()/read_pcap()/read_pcapng(), taking bytes rather than filenames, auto-detecting format, normalizing every timestamp to nanoseconds. tests/conftest.py drops its private reader (acceptance criterion 4); ~10 dependent test files migrated onto the shipped one. One idea from the issue's own text (the memoryview zero-copy 1.8× figure) was tried and reverted on measurement — see the PR for the numbers.

Every piece is additive — no breaking changes to existing behavior, matching 2.2.0's minor-version scope. Full QA ladder (lint, format, mypy strict, pytest with 98%+ coverage gate, the corpus-throughput benchmark gate) green on all four merges.

---

Generated by Claude Code