refactor: remove unreachable branches, commented-out raises and unused test helpers

Is your feature request related to a problem? Please describe.
These branches and names can never run or be read:

Header.read (pcap): the if _decl is None: raise FileError(...) branch is dead, because the schema's magic_number_callback raises first. Header(io.BytesIO(b'\x00\x01\x02\x03' + bytes(20))) raises ProtocolError: invalid magic number, so the Raises: FileError in the docstring is wrong too.
_read_opt_qs in IPv4, HOPOPT and IPv6_Opts: the final else: raise ProtocolError(... unknown QS function ...) is dead. Since fix(ipv4): an unassigned Quick-Start function code fails the whole header #1322 and fix(ipv6-opts): an unassigned Quick-Start function code fails the whole header #1237 the schema sends an unassigned function to UnassignedOption; I parsed every function nibble from 0 to 15 and none reached it.
_read_opt_calipso (HOPOPT, IPv6_Opts): both length guards are dead, because calipso_pad_len already raises FieldValueError: HOPOPT: invalid CALIPSO option length: 7 for any len < 8 + cmpt_len * 4. The % 8 != 0 clause is redundant even by itself.
Commented-out raise lines sit next to the warn that replaced them: _make_opt_pad (HOPOPT, IPv6_Opts, MH), IPv4._read_opt_sec, HIP._read_param_hip_cipher, the HIP schema's pre_unpack, HTTP/2._read_http_none, PCAP-NG's ts_* properties and _get_* helpers, and the PCAP-NG schema's block-length post_process.
_read_opt_qs (HOPOPT, IPv6_Opts) repeats # pylint: disable=unused-argument on its def line.
examples/generators/dispatch.py: code_name never reads _internet_enum, because no key in Internet.__proto__ is a bare int any more. The code_name docstring still says one is.
tests/project/test_release_tags_the_run_sha_1634.py: the '--json assets --jq .assets[].name' arm of FAKE_GH answers a query nothing makes; util/release_upload.py only sends ASSETS_JQ.
tests/project/test_sweeps_on_latest_only_1732.py: GATING is defined and never read, and test_pypy_worker_memory_2005 does not import it either.
Describe the solution you'd like
Delete each one, change Header.read's Raises: to ProtocolError, and update the comments and docstrings that describe what was removed (code_name, the FAKE_GH note).
Describe alternatives you've considered
Having magic_number_callback raise FileError, so that the documented exception becomes true. That changes the exception that every caller now gets for a bad file.
Additional context
Dead branches make the code look as if it handles cases it never sees, and coverage cannot reach them.

Related: #2300 rewords the magic-number message that Header.read actually raises; #2312 was split out of this list because it needs a ruling.

Blocked until the round-7 fix PRs for #2268–#2316 merge. This change spans the protocol modules those fixes edit, so it starts from the merged tree.

Fixing #2318 turned up a judgment call. Three existing tests reached what the issue calls dead code, but only by hand-fabricating a schema object and calling the private reader directly — bypassing unpack() entirely, which no real parse can do:

a Header with a bad magic number, built via object.__new__ instead of a real byte stream, to reach a FileError branch;
two IPv4/IPv6 cases with a QuickStartRequestOption/CALIPSOOption schema hand-built to disagree with itself.

My recommendation: these are genuinely dead on any real input, so the branches come out and the three tests move to the real path (e.g. Header(io.BytesIO(bad_bytes))), which the PR already does. If you'd rather keep a guard reachable only by a fabricated object — for defence-in-depth against a hand-built schema passed in some other way — say so and I'll restore it with a test that builds it that way instead.

take your recommendations

Ruling noted: taking the recommendation (remove the branches, move the three tests to the real path). Already implemented that way in the committed fix; proceeding to PR.

refactor: remove two unreachable branches, commented-out raises and u…

Closing note for the record: this issue's original "dead code" claim for _read_opt_qs's unknown-QS-function arm was based on parse-only evidence (reachable only by calling the private reader directly). The fix in #2378 found that arm is actually reachable via the public construction path too — a QuickStartRequestOption/QuickStartReportOption built directly with an unassigned func hits it regardless of parse dispatch — so it was kept rather than removed, unlike the other two guards this issue named (Header.read's FileError, _read_opt_calipso's length guards), which are confirmed genuinely dead through both construction and parse routes.

refactor: remove two unreachable branches, commented-out raises and u…

Navigation Menu

Repository navigation

refactor: remove unreachable branches, commented-out raises and unused test helpers #2318

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

refactor: remove unreachable branches, commented-out raises and unused test helpers #2318

Description

Activity

JarryShaw commented on Oct 26, 2026

JarryShaw commented on Oct 27, 2026

JarryShaw commented on Oct 27, 2026

JarryShaw commented on Oct 27, 2026

JarryShaw commented on Oct 27, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions