Skip to content

docs(changelog): correct four claims in the published 0.3.0 entry - #149

Merged
ArangoGutierrez merged 1 commit into
mainfrom
fix/b-changelog-accuracy
Aug 19, 2026
Merged

docs(changelog): correct four claims in the published 0.3.0 entry#149
ArangoGutierrez merged 1 commit into
mainfrom
fix/b-changelog-accuracy

Conversation

@ArangoGutierrez

Copy link
Copy Markdown
Owner

Closes #144.

Four statements in the published v0.3.0 entry were not true of the shipped build. Found by review after the tag was cut, so they were already public.

  • "It is now one click in Preferences" ignores the first run. PrivilegedHelperClient.swift comments "user's first click lands here" on a branch returning approvalGuidance, so the real flow is click, approve a switch in System Settings, click again. site/docs.html already described this correctly, so the changelog contradicted the project's own docs.
  • "no other process on the machine can arm the hold" is defeated by sudo coffee-bar-probe arm, which ships in the same image. The pin is a property of the XPC endpoint, not of the machine.
  • "refuses a non-POST request with 405" is wrong for half the socket: /status answers GET with 200 and returns 405 Allow: GET for other methods.
  • The SleepDisabled ceiling tripling to 24 hours went unstated, which is the release's most privilege-relevant bound.

What was deliberately NOT changed

The adjacent clause "the only part of coffee-bar that involves root at all" is true and is preserved byte-identical in both files. ProbeVerb.requiresRoot is true only for arm, report, revert, watchdog and serve, and every one is a lid-closed operation. Two site pages assert the same thing; a disagreement would be worse than the original defect. The issue as originally filed named this clause, and that filing was corrected before the work started.

The 0.1.0 entry's "hard eight-hour cap" also stands: it is true of the build 0.1.0 shipped, and rewriting it would falsify release history.

Testing done

  • swift test rc=0, 1235 tests in 12 suites, verified independently on a clean scratch path rather than taken from the implementer's report.
  • Version headings, dates and the release count are byte-identical to main in both files, checked by diff.
  • Both changelog files move in one commit, as everyReleaseInTheChangelogIsOnTheChangelogPage requires.

Breaking changes

None. Documentation only.

The 0.3.0 entry was tagged and published before review caught four
statements that are not true of the shipped build. CHANGELOG.md's own
header requires every claim to be true of that build, so the entry is
corrected in place rather than left standing with an erratum in 0.4.0.
The version heading, the date and the release facts table are untouched.

- The intro claimed arming is "one click in Preferences". A first run
  takes two: PrivilegedHelperClient returns approvalGuidance on the
  first, because macOS registers the SMAppService job switched off and
  waits for the user to enable it under Login Items. site/docs.html
  already described the flow that way, so the entry contradicted the
  project's own docs page.

- The helper bullet claimed the Team ID and bundle identifier pin means
  "no other process on the machine can arm the hold". The pin binds the
  XPC endpoint, not the machine: sudo coffee-bar-probe arm arms the same
  hold, ships in the same disk image, and SECURITY.md records it as the
  only route available on a Homebrew install. The sentence beside it,
  that lid-closed mode is the only part of coffee-bar involving root at
  all, is TRUE (ProbeVerb.requiresRoot covers arm, report, revert,
  watchdog and serve alone, every one a lid-closed operation) and is
  left exactly as published, matching site/index.html and site/docs.html.

- The 405 bullet claimed the socket refuses "a non-POST request". Only
  the hook channel is POST-only. IngestListener registers /status as a
  read endpoint, answers GET there and returns Allow: GET for anything
  else. The entry lists that read API a few bullets earlier, so it
  contradicted itself.

- The entry named a twenty-four hour ceiling without saying it had
  moved. 54f0058 (#121) tripled JournalRecord.maxTTLSeconds from eight
  hours to twenty-four in this release, and that cap is the most
  privilege-relevant bound the release changed. Older entries state an
  eight-hour cap that was correct for the builds they describe, so the
  change is recorded here rather than by rewriting release history.

Both files move in one commit so
everyReleaseInTheChangelogIsOnTheChangelogPage stays green.

Refs #144

Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>
@ArangoGutierrez
ArangoGutierrez merged commit ef6c640 into main Aug 19, 2026
2 checks passed
@ArangoGutierrez
ArangoGutierrez deleted the fix/b-changelog-accuracy branch August 19, 2026 17:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

The 0.3.0 changelog entry overclaims in four places

1 participant