Repository navigation
docs: close README/skill package-count drift against the manifests - #225
Conversation
Advisor notes not addressed before completionThe contributor's advisor posted these after the final verdict; the agent had no further turn to address them. Recorded here for the reviewer.
— hive contributor relay (#7879)🐝 Hive Agent: |
|
Both advisor notes were correct, fixed in
Comment/doc-only change; no code or tests touched.🐝 Hive Agent: |
castrojo
left a comment
There was a problem hiding this comment.
Reviewed diff: adds check-doc-counts.py to Justfile to prevent package count drift between README and manifests. Branch currently has merge conflict against main (stale index.json/README). Needs rebase by author.
README.md and docs/skills/package-contract.md hand-quoted 61/12/4 for the Bluefin/Utah-additions/unavailable package counts while packages/utah.toml's [unavailable] list had grown to 9 tracked gaps and [parity] to 28 entries -- the docs said 4, the manifest said 9. Recompute both from generate-site-data.py's own build() (58/44/9, matching site/data/packages.json) and add scripts/check-doc-counts.py, wired into 'just check', so the two cannot silently diverge again. Also: - Drop the stale 'pipewire-libs-extra is missing' gap: packages/utah.toml's own header records it as returned to the install contract, and it is one of the 58 packages the manifests now show installed. - Stop the README's top banner and 'Nothing is published' bullet from claiming the installer/offline payload is unimplemented -- the live ISO's bootc-installer already passes an offline, LUKS2-encrypted install in local QEMU validation (docs/verification/README.md). State publication status separately from that local implementation status. - Point docs/building.md's quickstart at the ISO/E2E/rollback-adjacent gates that already exist (just iso, boot-iso, luks-test, try-installed, check-desktop-contract, check-repos) instead of leaving them undiscovered in skills/local-testing.md. Part of #24 (Utah/Dakota release-readiness gaps), specifically #23's documentation-accuracy acceptance criteria. Assisted-by: Claude Sonnet 5 via omp/hive Signed-off-by: Danathar <Danathar@users.noreply.github.com>
- Justfile's iso recipe comment still said installer payload integration was 'the next ISO milestone'; README now documents it as implemented and passing an offline LUKS2 install end to end (this PR's own diff). - docs/building.md credited scripts/update-e2e-readme.py for refreshing docs/verification/README.md and the root README E2E block, but iso/scripts/luks-e2e.sh writes both itself; update-e2e-readme.py is a separate CI-side script just luks-test never invokes.
31908c0 to
53a201f
Compare
|
Rebased onto The conflict, and how I resolved it. Both sides had rewritten the same "Nothing is published" bullet in the README's status banner: On the two advisor notes — both were already closed before the final commit, so the rebase carried them through rather than needing new work. Confirmed at this head:
Gates after the rebase, since the count check is the whole point of the PR and a rebase is exactly when it could go stale: One pre-existing warning, not introduced here and not failing:
|
castrojo
left a comment
There was a problem hiding this comment.
Reviewed diff and CI: adds check-doc-counts.py to Justfile to close drift between README and manifests. All 14 checks pass. Outstanding: two independent maintainer reviews required (+2 policy).
hanthor
left a comment
There was a problem hiding this comment.
Reviewed: checks green, change matches its description. Merging per maintainer triage.
What
Closes the documentation-accuracy slice of #23, one of the linked audit
slices in the #24 release-readiness tracking issue.
docs/skills/package-contract.md's "Current counts" sentence both quoted61/12/4 (Bluefin contract installed / Utah additions / genuinely
unavailable) while
packages/utah.toml's[unavailable]list had grownto 9 tracked gaps and
[parity]to 28 entries. The correct numbers,recomputed from
generate-site-data.py's ownbuild()(the function thatalready writes
site/data/packages.json), are 58 / 44 / 9.scripts/check-doc-counts.py, wired intojust check, so thiscan't drift silently again: it recomputes the three counts from the
manifests and fails if either document disagrees.
pipewire-libs-extrawas missing;packages/utah.toml's own headeralready records it as "returned to the install contract" (ci: check-repos fails on main — firefox, grub2-efi-x64-cdboot, pipewire-libs-extra unresolvable #35), and it is
one of the 58 packages the manifests now show installed.
the "Nothing is published" bullet said "there is no installer payload" —
false: the live ISO's bootc-installer already passes an offline,
LUKS2-encrypted install end to end in local QEMU validation
(docs/verification). Reworded both to state
publication status (nothing is published) separately from local
implementation status (the installer works locally).
docs/building.md's "Build locally" loop stopped atboot-vm; added a"Validate the release-readiness gates" section pointing at
just iso,boot-iso,luks-test,try-installed,check-desktop-contract, andcheck-repos, with a pointer toskills/local-testing.mdfor thebootc rollback/uupdpolicy.Verification
All pass locally.
just check's podman/network-dependent steps(
check-repos,check-desktop-contract, full image builds) were not run inthis sandbox; nothing in this change touches the Containerfile, packages, or
image build path — it is docs plus one new standalone script and its test.
python3 tests/run_suite.pyshows the same 8 failures / 48 errors on thisbranch as on unmodified
main(missingpyyaml/jsonschemain thissandbox, pre-existing); the 4 new tests in
tests/test_check_doc_counts.pyall pass.Out of scope for this PR: the rest of #24's linked slices (CI topology,
VM/rollback lifecycle tests, offline ISO gates, supply-chain attestation,
etc.) are separate, already-tracked issues (#11, #12, #16-#22) with their
own open work.
— hive: backend=omp model=anthropic/claude-sonnet-5 effort=high