fix(corekit)!: to_dict() returns an InfoDict, so a repeated IPv6 extension header survives the dict rebuild (#1453)

[JarryShaw]

• Searched for similar pull requests
• Followed the coding style — isort -l100 -ppcapkit clean; pylint/mypy (Makefile flags) on the 4 modules: no new findings in the changed lines
• make test passes, and a test case covers the change — on 94ae9664c: legs corekit 633, dumpkit 232, protocols/internet 617, project 606, all OK; new module 7 passed, 41 subtests
• 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?

• fix — corrects a defect

---

Description of your pull request and other information

Closes #1453

to_dict() now returns InfoDict, a new OrderedMultiDict subclass, for every Info, so the return type does not depend on the packet. The dict interface, json and the dumpers see the first value per key, and items(multi=True) yields every value in wire order. Info.__update__ given a MultiDict adds to a key it already holds, IPv6 adds each extension header that way, and _lookup_exthdr walks items(multi=True).

Breaking (!), for callers of to_dict():

• type(d) is dict is now False (isinstance(d, dict) still holds).
• d == plain_dict is False when a key repeats.
• yaml.safe_dump(d) raises RepresenterError.
• A repeated header shows the first, not the last, in d[key], info.opts and the dumps.
• Dumping costs 8–11% more on a loaded host (reviewer's figure).

Also touched: the InfoDict docs entry, and pcapkit/dumpkit/common.py, whose object_hook writes any OrderedMultiDict as a list of entries; one condition now sends an InfoDict to the dict branch, so all 92 dumps (23 captures × json/tree/plist/xml) are byte-identical to main at 94ae9664c.

Dict rebuilds before → after: dst-dst 56→64/64, hop-dst-dst 64→72, dst-dst-dst 56→72, Routing-Routing 56→64, dst-rt-dst 56→72, hop-dst-rt-dst 64→80. Byte-exact through Ethernet and Frame too. copy.copy(info) now gets its own __multi__. With main's sources, the new tests/protocols/test_to_dict_repeated_exthdr_1453_unit.py reports 34 failed, 3 passed; with this PR, 7 passed, 41 subtests.

[github-actions]

Coverage: 89.56% (unit tier, Python 3.14, 074e0c89b, Unit Tests run success)

Package	Statements	Missed	Branches	Partial	Cover
pcapkit (top level)	104	4	20	4	93.55%
pcapkit/const	18795	957	2356	870	91.00%
pcapkit/corekit	2107	93	672	27	94.60%
pcapkit/dumpkit	242	2	88	2	98.79%
pcapkit/foundation	2648	98	930	48	94.80%
pcapkit/interface	112	7	40	5	92.11%
pcapkit/protocols	16899	222	4564	187	98.06%
pcapkit/toolkit	539	74	168	3	84.58%
pcapkit/utilities	429	4	122	4	98.55%
pcapkit/vendor	4409	2342	1006	157	43.25%

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

[JarryShaw]

Verdict: NEEDS CHANGES at dd2029c4e (Sonnet cross-review; the author ran on Opus). Two small items.

1. Docs. docs/source/pcapkit/corekit/multidict.rst has no InfoDict entry. The new docstrings in infoclass.py cross-reference pcapkit.corekit.multidict.InfoDict, so those links will not resolve. Add an autoclass block with :show-inheritance: and items.
2. copy.copy(info) shares __multi__. I confirmed this on this head with a dst-dst chain: the copy's __multi__ is the original's object. A later __update__ on the copy therefore leaks into the original (pcapkit/corekit/infoclass.py:447). Copy it per instance and add a test.

Holds:

• All 88 dumps (22 captures × json, tree, plist, xml) are byte-identical to main.
• InfoDict is built only in Info.to_dict, so the options OrderedMultiDict list form from fix(dumpkit): MultiDict and OrderedMultiDict dump groups values by key, losing wire order #1263 is untouched.
• The dict-form rebuild is exact for dst-dst, hop-dst-dst, dst-dst-dst, route-route and dst-route-dst. Wire order is kept.
• {**d}, |, json, copy/deepcopy/pickle and key-set operations all work.
• It merges cleanly with fix(schema): a header cut inside a fixed-width field rebuilds as captured (#1458, #1451) #1465, and both test modules pass.
• Without the fix, 33 tests fail. All legs pass.

Notes:

• The measured dump cost is 8–11% on a loaded host, against the 7% reported.
• type(d) is dict and yaml.safe_dump now fail. Mention both in the PR body under the !.

[JarryShaw]

Verdict: GOOD TO GO at 074e0c89b. Round-2 cross-review on Sonnet; the author ran on Opus.

• copy.copy leak: fixed. __copy__ recreates all three per-instance attributes that __new__ sets (__map__, __map_reverse__, __multi__). Disabling it makes test_info_copy_keeps_its_own_repeats fail. I re-checked myself that a copy shares neither __multi__ nor __map__ with the original.
• Rebuild is byte-exact from both info and info.to_dict(), across ten independent extension-header chains: repeated Destination Options, Routing, and mixed chains starting with Hop-by-Hop. I also checked Destination Options ×2 followed by UDP: 68 of 68 octets on both paths.
• Dumps are unchanged. All 92 dumps (23 captures × json/tree/plist/xml) are byte-identical to main per diff -rq.
• Docs: the InfoDict entry matches the surrounding style.

Optional additions to the breaking-changes list, not blockers:

• json.dumps(info.to_dict()) silently keeps only the first repeated header.
• len(d) counts keys, not values.
• Nested values are also InfoDict.