docs(changelog): correct the LICENSE history in the #630 entry

[JarryShaw]

#630 merged with two inaccurate claims in its changelog entry. Both are about the history of LICENSE rather than about the change itself, so LICENSE is untouched here — 2017 was and remains the right answer. This is a changelog-prose correction only.

I wrote the bad claims, and a cross-review caught them after #630 had already been merged. Raising the correction rather than leaving it, because the entry is the permanent record of why the notice changed and it currently argues from facts that are not true.

Error 1 — the pre-BSD history, wrong in every clause as a description of the era

The merged entry says:

bc836cfa2 (2020-05-31) introduced 2018-2020 when the BSD-3-Clause text replaced MPL 2.0 — the file until then was stock MPL text carrying no author notice at all, so 2018 is the first start year the project ever asserted and it was already a year late when it was written

Enumerating all 14 commits that ever touched LICENSE (git log --follow -- LICENSE | wc -l → 14) and reading each blob:

commit	date	licence	author notice
c57f7d0b7	2017-11-07	MIT	Copyright (c) 2017 Jarry Shaw — line 3
d0f308363	2017-11-27	MIT	Copyright (c) 2017 Jarry Shaw — line 3
bc6da0fb8, 9705860e0	2017-12-01	GPL v3	Copyright (C) 2017 JarryShaw
12dd3902e	2018-03-06	GPL v3	Copyright (C) 2017 Jarry Shaw
083c7ad5e	2018-06-30	GPL v3	Copyright (C) 2017 Jarry Shaw
4f3e00d5f	2018-12-08	Apache 2.0	Copyright 2018 Jarry Shaw
1c69341dc	2018-12-08	MPL 2.0	none
bc836cfa2	2020-05-31	BSD-3-Clause	Copyright (c) 2018-2020, Jarry Shaw

So:

• "the file until then was stock MPL text" — MPL only from 2018-12-08. Before it: Apache (same day), GPL v3, MIT. Four licences before BSD, not the one implied. (Each clause quoted here was true of the single blob bc836cfa2 replaced; the error is generalising it across the era.)
• "carrying no author notice at all" — true of the MPL blob alone. MIT carried one on line 3; the GPL "how to apply these terms" appendix was filled in, not left as placeholders (it names the project: "The JSPCAP project is an open source Python program … Copyright (C) 2017 Jarry Shaw").
• "2018 is the first start year the project ever asserted" — flatly false. c57f7d0b7:LICENSE:3 is Copyright (c) 2017 Jarry Shaw. The project asserted 2017 in its very first commit and kept asserting it for the next seven months.
• "already a year late when it was written" — no. The first 2018 was written on 2018-12-08 in an Apache single-year notice, correct for the year it was written. It only became a wrong start year in 2020, when bc836cfa2 carried it into BSD as a range start — by which point it was two years stale.

The passage also contradicted the entry's own opening sentence, which dates the first commit to 2017-11-07 and then claims 2018 was the first year ever asserted.

The true history is a better argument than the one that was written, which is why this corrects rather than deletes: 2017 is the year the project asserted from its first commit, lost in the December-2018 Apache relicence, and never recovered. #630 restores it rather than asserting it anew.

Error 2 — the title line was attributed to the wrong template

The merged entry credits the BSD 3-Clause License title line to "the opensource.org template". It does not carry it: https://opensource.org/license/bsd-3-clause's licence text begins at Copyright <YEAR> <COPYRIGHT HOLDER>, and "The 3-Clause BSD License" is its page heading, not licence text.

The template whose body literally begins with that line is GitHub's choosealicense.com one:

$ curl -sS https://raw.githubusercontent.com/github/choosealicense.com/gh-pages/_licenses/bsd-3-clause.txt \
    | awk '/^---$/{c++; next} c>=2' | head -3
BSD 3-Clause License

Copyright (c) [year], [fullname]

Note it also carries the comma after the year, which is the form this repository's LICENSE uses. Reattributed accordingly.

How the error happened, and what caught it

I checked bc836cfa2^:LICENSE and 1c69341dc:LICENSE, found no author notice in either, and generalised "until then" from that. But git rev-parse bc836cfa2^ is 8f937d7af, whose tree holds 1c69341dc's blob — the two things I checked were the same file, so it was one data point covering only the final 18 months of a 2.5-year pre-BSD period. Enumerating every commit that touched LICENSE, rather than sampling the one before the relicence, is what found it.

What this changes

• docs/source/changelog/1.5.0.rst — the fix(license): start the copyright term at 2017, and drop the end year #630 entry's history passage rewritten to match the evidence, and the title-line attribution corrected. Plus a new **Fixed** bullet recording that the fix(license): start the copyright term at 2017, and drop the end year #630 entry carried these two false claims, since a silent rewrite would hide that the repository once asserted them.
• CHANGELOG.md — regenerated with util/changelog_md.py; --check exits 0. Not hand-edited.

Nothing else. LICENSE is byte-identical to main, and nothing under pcapkit/ is touched. No tests were run: changelog prose only.

Requested implicitly by the review rather than by an issue — no issue is open for it.

Note for whoever merges

This touches the same changelog files as #631 and the other changelog PRs in flight, so expect a conflict. Resolve the .rst keeping every bullet, then regenerate CHANGELOG.md with util/changelog_md.py rather than merging the Markdown by hand.

Merge this one before #631 if there is a choice: it edits the #630 passage that #631 does not touch, so resolving in that order makes #631 a clean append.

[JarryShaw]

Cross-review: GOOD TO GO

Independent cross-review by a subagent on a different model (Claude Haiku), run read-only on its own clones in /tmp, briefed to falsify rather than to bless.

This PR exists because of that review. The reviewer refuted the historical claim that #630 merged with, and it was right — the specific thing it was then asked to hunt for on this PR was whether the correction overcorrects or introduces a fresh error, having already got this wrong once.

Both open PRs are correct. PR 638 fully repairs the two false claims that #630 merged with … Every substantive clause I could check verifies against the blobs, and I found no fresh error introduced by the corrections — which was the specific failure mode you asked me to hunt for.

Every clause of the new passage re-derived from the blobs

• bc836cfa2 (2020-05-31) introducing 2018-2020 when BSD replaced MPL 2.0 — confirmed
• three subsequent hand bumps to 2018-2022, 2018-2023, 2018-2026 (docs: add CITATION.cff and bring the copyright notice up to date #615), each correcting the end year and copying 2018 forward — confirmed via git log -L3,3:LICENSE
• c57f7d0b7:LICENSE line 3 reading Copyright (c) 2017 Jarry Shaw under MIT — confirmed exact
• the GPL v3 appendix being filled in rather than stock — confirmed at 12dd3902e:LICENSE:634-636, which reads "The JSPCAP project is an open source Python program focus on / PCAP parsing and analysis. / Copyright (C) 2017 Jarry Shaw", where stock GPL carries <one line to give the program's name…> / Copyright (C) <year> <name of author>
• 4f3e00d5f relicensing to Apache 2.0 on 2018-12-08 under Copyright 2018 Jarry Shaw, a single year correct for its year — confirmed
• 1c69341dc replacing it with MPL 2.0 the same day, carrying no author notice — confirmed
• the "restores rather than newly asserts" framing — confirmed as the right framing and what the evidence supports
• the "two years stale" characterisation — examined specifically at my request and confirmed as not overcorrecting
• the 14-commit figure — git log --follow --oneline -- LICENSE | wc -l → 14, enumeration matching commit for commit
• git rev-parse bc836cfa2^ → 8f937d7af, holding 1c69341dc's blob, so the "same blob, one data point" diagnosis of how the error happened is itself correct
• "final 18 months of a 2.5-year pre-BSD period" (17.8 months; 2y 6.8m) and "the next seven months" (7.7 months) — both confirmed
• the choosealicense.com body opening BSD 3-Clause License / blank / Copyright (c) [year], [fullname], against opensource.org's text beginning at the copyright line with "The 3-Clause BSD License" as page heading only — confirmed, so Error 2 is correctly diagnosed and correctly reattributed

It also confirmed LICENSE is byte-identical to main (git diff cfb81d3f6 <head> -- LICENSE empty), one commit, no merges, correct authorship, and only the two changelog files touched.

On the shape of the fix

I asked explicitly whether correcting in place was right or whether the original wording should have been left alone:

I would not have left the original text alone. A changelog entry is read as a factual record of why a change was made, not as an append-only log, so leaving a false argument in place and appending a contradiction would mislead anyone who reads only the #630 entry — which is most readers. Correcting the passage and recording that it once said otherwise gives both properties: the entry now argues from true facts, and the fact that the repository once asserted the false ones is not erased.

Two copy-edits it raised, both applied

1. "five licences, not two" sat next to a list of four (the fifth being BSD, which the reader had to supply) while the sentence was scoped "the pre-BSD file was…". Now reads "four licences before BSD, not the one implied".
2. "Every clause of that is wrong" was very slightly overstated — each clause was true of the single blob bc836cfa2 replaced; the error was generalising across the era. Now reads "Every clause of it is wrong as a description of the era, though each was true of the single blob bc836cfa2 happened to replace, which is where the mistake came from".

Advisory, non-blocking

• Merge this PR before fix(packaging): ship CITATION.cff in the source distribution #631. They conflict on the changelog files; this one edits the fix(license): start the copyright term at 2017, and drop the end year #630 passage that fix(packaging): ship CITATION.cff in the source distribution #631 does not touch, so that order makes fix(packaging): ship CITATION.cff in the source distribution #631 a clean append. Resolve the .rst keeping every bullet, then regenerate CHANGELOG.md with util/changelog_md.py rather than merging the Markdown by hand.
• The bullet list in 1.5.0.rst is inconsistent about blank lines between items — pre-existing, not made worse here, and free to tidy while the file is being hand-resolved anyway.

Caveat stated rather than glossed

No tests were run at all, deliberately — a coverage run -m pytest in this repo reached 41.4 GB RSS today and had to be killed. Nothing in this PR needs them: it is changelog prose only.