comments(examples): state the constraint, drop the citation — the last tree (refs #792)

[Yaraslaut]

The third and last tree of the repository-wide comment cleanup. include/morph
and src went first, then docs/spec, scripts, .github, cmake and
tests; examples/ was what remained, and it was the largest single tree.

523 references removed, 13 left. 466 in the morph#NNN form and 57 in the
bare #NNN / issue #NNN / GitHub-URL forms the first pass's grep never saw.

area	morph# before	bare # before	after
common	38	4	0
bank	51	0	0
bookmarks	10	6	0
crm	18	1	0
kanban	114	26	13
ledger	118	3	0
lims	49	14	0
pastebin	17	0	0
polls	27	2	0
concepts / forms / qt_tls_client	3	0	0
IMPLEMENTATION.md / LADDER.md / TESTING.md / rungs.txt / CMakeLists.txt	21	1	0
total	466	57	13

The 13 are Catch2 tags — [kanban][journal][morph#757] and nine
[kanban][board][morph#751]-shaped siblings in
tests/test_board_post_commit_tail.cpp, plus one [kanban][board][morph#566].
Renaming a tag is a code change, not a comment one, so they stay; .github/ and
scripts/ were grepped and no selector names any of them.

examples/kanban/gui/qml/BoardView.qml and
examples/kanban/tests/test_board_layout.cpp are untouched — morph#794 holds
them.

What this is not

Most of the diff is rewriting, not deleting: +1390/−1514 over 173 files, in
line with the second pass's +2144/−2657 over 179. Removing a citation frees one
clause; restating "before morph#373 this was accepted" as what accepting it
would cost costs about what it saves.

Measured blocks stay whole. A few, so the shape is on the record:

• ledger_model.cpp's ReadTxnGuard: the rollback-journal vs. WAL contention
  contrast, the DataMapperPool::Return consequence, the 60s busy_timeout
  that follows from it, and why turning WAL on to make a snapshot name true is
  a separate decision. Only the ticket left.
• bank's unique_test_database.hpp: 21 of 21 cases failing with
  HY000 (10) - [SQLite]disk I/O error (10) under ctest -j.
• test_bank_gui_format.cpp: the 0.49999999999999994 + 0.5 == 1.0 table, and
  why 0.005 is kept as the half-way input that must still round away from
  zero.
• test_action_driver.cpp: the 0.9^60 = 1.8e-3 sampling argument, and the seed
  sweep that found 314, 779 and 2522 failing.
• crm's GCC 16 -Warray-bounds block: the diagnostic text, the speculative
  devirtualization that reaches it, the 112-vs-136-byte holder sizes, and the
  == 16 bound that makes the suppression lapse rather than accumulate.
• examples/*/include/.clang-tidy: the Variable /ClassConstant /
  ConstexprVariable IgnoredRegexp probe table, which is the whole reason the
  narrower-reading option is the wrong one.
• TESTING.md: the 165 TSan warnings that make a Qt-linked rung unsanitisable,
  the 95.85% examples/common baseline with the four-file split that accounts
  for the drop since, and the ctest -L ladder-lims 85-against-87 count.

A comparison arguing for the current design stays, restated as what each
alternative costs rather than as what was tried when. Where a paragraph's only
content was "a gate used to check this, and it was removed", the rule it was
holding is now written as a rule to keep by hand — which is what is true.

CHANGELOG.md, docs/superpowers/**, Catch2 tags and the audit-finding
vocabularies are exempt and untouched. UAX #44 and the
test_a_nullable_ref_member_keeps_morph189s_numeric_encoding QML function do
not occur under examples/.

User-visible strings changed

Six string literals, each grepped for readers first.

string	what reads it
PoolTransactionAudit's leak diagnostic: "morph#740: a pooled DataMapper was …" → "PoolTransactionAudit: …"	its own self-test, test_pool_transaction_audit.cpp:107, whose contains("morph#740") moves to contains("PoolTransactionAudit") in the same commit. The SQL_ATTR_AUTOCOMMIT still OFF phrase that test also asserts on is untouched.
kanban::BoardEventId::fromRowId's throw: … sentinel (morph#215) → … sentinel	nothing. Grepped for the message text and for unrepresentable; the only assertion on this path is CHECK_THROWS_AS(…, kanban::KanbanError).
polls::OptionId::fromRowId, polls::PollEventId::fromRowId: same shape	nothing, same check.
ledger's static_assert on LedgerRecord::owner: … locking them out of it (morph#382).	nothing — a static_assert message, grepped for locking them out.
test_gui_forms_render.cpp's inline // morph#387 beside a continue	a comment, not a string; the filtered warning text it describes (onOptionsReceived / DynamicForm.qml) is unchanged.

No TEST_CASE was renamed. Every ticket reference that survives in a
TEST_CASE line is in the tag argument, not the name, so no ctest entry name
moved and no -R/-E selector in .github/workflows/ or scripts/ is
affected. (ci.yml:678's -E "OomInjector|morph#108" names nothing under
examples/.)

Non-comment changes, exhaustively

A script over git diff -U0 origin/master...HEAD printing every changed line in
a .cpp/.hpp/.qml/CMake/.clang-tidy file that is not a comment returns
exactly ten lines: the five string-literal pairs above and two trailing
comments. No behaviour change anywhere else.

Gates

gate	result
clang-format --dry-run -Werror (22.1.8) over the 131 changed .cpp/.hpp	clean
clang-tidy-diff.py (22.1.8) against origin/master...HEAD, CI's own filter-unbuilt-sources.py, -path the augmented db	exit 0; 168 of 173 file sections analysed, 49 headers pinned, 0 sources skipped as unbuilt. Compile db: 741 in-workspace sources, 288 under examples/ — above the 600/200 floors.
Doxygen --target doc, WARN_AS_ERROR = FAIL_ON_WARNINGS	exit 0, no warnings
ctest full, linux-everything	3048/3048 pass (3042 + 2 fuzz + 4 bench/soak)
ladder + bank labels	1120/1120 pass

Each label filter was checked with ctest -N -L <label> for a non-zero count
before being trusted: bank 37, ladder-0 159, ladder-bookmarks 130,
ladder-crm 169, ladder-kanban 184, ladder-ledger 162, ladder-lims 140,
ladder-pastebin 60, ladder-polls 79, ladder (prefix) 1083.

clang-tidy-diff first reported one finding —
error: 'test_qml_surface.moc' file not found — which is the configure-only
artefact ci.yml:2904 documents at length. Building
ladder_common_tests_autogen and morph_forms_qml_tests_autogen, exactly as
that job does, cleared it; the re-run is the exit 0 above.

Commits

One per rung or coherent group, so a bad slice can be dropped without redoing
the rest: the 16 .clang-tidy files, then common, bank, ledger, kanban,
crm, lims, then polls/pastebin/bookmarks, then the ladder's own docs, then a
whitespace-only reflow of the paragraphs the citation removals left ragged.

🤖 Generated with Claude Code

https://claude.ai/code/session_01VptDWG2fKr2vBnLSJcgzgW

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!