fix(vendor): make wrapped const comments render as written (#1846, #1847)

[JarryShaw]

Please follow the guide below

• You will be asked some questions, please read them carefully and answer honestly

• Put an x into all the boxes [ ] relevant to your pull request (like that [x])

• Use Preview tab to see how your pull request will actually look like

• 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. Narrow runs only, listed below.

• Added a changelog entry under docs/source/changelog/ and regenerated CHANGELOG.md, if the change is user-visible: 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?

Tick the commit type your subject line carries.

• fix — corrects a defect
• feat — adds a feature
• perf — changes performance, not behaviour
• refactor — changes neither behaviour nor performance
• test — tests only
• docs — documentation only
• ci — workflows or build tooling
• release — bumps the version or rolls up a distribution
• chore — anything else

---

• Vendor.wrap_comment puts a space after a URL reference that another
  bracket follows, which docutils autolinked into the next citation, and
  backslash-escapes a leading reST enumerator such as (S).
• Vendor.wrap_comment and AppType.tabulate pass break_on_hyphens=False: a
  continued line joins with a space, so a hyphen split rendered "run- time"
  and cut a list-table cell's URL short.
• AppType.wrap_comment takes its lines from the base instead of wrapping on
  its own, so it gets both fixes; nothing it emits trips the fix(vendor): generated comments run a URL reference into the next bracket and start a line with (S), which reST misrenders #1846 cases yet.
• Re-render the affected const comments and AppType tables offline from the
  committed text, with no crawl. Members, values and code are unchanged.
• Add tests/vendor/test_vendor_wrap_comment_unit.py.

Closes #1846
Closes #1847

The const files were re-rendered offline from their committed text, so no crawler ran. Each is AST-identical to main with docstrings blanked, and a member dump of every changed module is byte-identical.

[JarryShaw]

Verdict: GOOD TO GO at db1129257. A Sonnet cross-review (author: Opus) confirmed the following:

• Enumerator escape. Only (S) flag in mh/dhcp_support_mode.py matches the escape rule. Look-alikes such as X.509, AX.25, H.248 and e.g. are untouched, and docutils agrees on 35 probes.
• Hyphen rejoins. Every rejoined word is a single word upstream, checked against the IANA service-names, SCTP, HTTP and protocol-numbers registries and tcpdump's linktypes page.
• Wrapper output. The patched wrappers reproduce every wrapper-made #: block and all the AppType list-tables byte for byte, so the next crawl will not re-diff them.
• Members unchanged. A dump of every enum member in every pcapkit.const module is byte-identical to main.
• Tests. The generator tests fail on main's templates.

Non-blocking:

• the enumerator check looks for a literal space, so a newline or tab after (S) would slip through;
• AppType table cells do not get the URL spacing.
  Neither occurs in the data today.

[JarryShaw]

Verdict: NEEDS CHANGES at db1129257. This replaces the earlier GOOD TO GO. CI's Python 3.14 job fails 18 tests in tests/test_tier_guard.py (job 117260332315), all with the same message: 'docutils' has no MODULE_PROVIDERS entry.

The cause is the new HAS_DOCUTILS gate in tests/vendor/test_vendor_wrap_comment_unit.py. The dependency-gate guard needs every gated module mapped to the distribution and extra that provide it, or, if no extra carries it, an exclusion saying why. The cross-review never ran tests/test_tier_guard.py, so it missed this. A worker is on the fix.

[JarryShaw]

Verdict: NEEDS CHANGES at b20fe2fef. The CI fix is correct: tests/test_tier_guard.py went from 18 failed on db1129257 to 111 passed. A Sonnet cross-review also confirmed the HAS_DOCUTILS exclusion's dark set matches what the guard derives, exactly the test and gate jobs.

One required change, a false comment. tests/_dependency_gates.py (~239) says listing the bare docutils string lets a future extra that declares it close the gap without editing this entry. Measured: with docutils added to an extra, docutils becomes a contested import, and test_mutually_exclusive_imports_matches_its_derivation fails until the entry is cut to the bare name. So reword it to say that, or drop the sentence, and keep both tuple entries. In the same edit, reword "one distribution reached two ways" so it carries no count.

[JarryShaw]

Verdict: NEEDS CHANGES at b671834b7. The reworded MODULE_PROVIDERS['docutils'] comment is now true, and the reviewer re-ran the experiment to check it:

• with docutils declared in the test extra, the entry must be cut to the bare name and the exclusion deleted;
• with docutils declared in the docs extra, only the entry must change.

test_tier_guard.py gives 111 passed.

One change left, the same wording error I've flagged on #1877 and #1881. The new HAS_DOCUTILS reason says it needs "the docs extra that only deploy-pages.yml installs". No workflow installs the docs extra. deploy-pages.yml:167-168 installs docs/requirements.txt (which brings Sphinx, and with it docutils) and .[all,dev]. Reword it to say that.

[JarryShaw]

Verdict: GOOD TO GO at adf8f9c4f. The Sonnet reviewer checked the last change, which touches only the HAS_DOCUTILS reason. Against .github/workflows/, its claims hold:

• no pytest job installs docutils;
• no workflow installs the docs extra;
• deploy-pages gets Sphinx, and with it docutils, from docs/requirements.txt.

The reason states no count, and tests/test_tier_guard.py gives 111 passed. The pre-existing HAS_SPHINX reason, which has the same error, is corrected in #1877.

[github-actions]

Coverage: 90.05% (unit tier, Python 3.15, adf8f9c4f, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	2	20	2	96.77%
pcapkit/const	19595	881	2618	867	92.11%
pcapkit/corekit	2462	69	780	35	96.55%
pcapkit/dumpkit	276	3	100	3	98.40%
pcapkit/foundation	2914	164	1016	34	93.59%
pcapkit/interface	138	7	52	5	93.68%
pcapkit/protocols	17406	224	4580	189	98.10%
pcapkit/toolkit	872	177	298	8	77.18%
pcapkit/utilities	512	2	158	4	99.10%
pcapkit/vendor	4409	2342	1010	157	43.29%

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