feat(rt-prep): port release-prep.sh → rt prep (785 LOC, 39 git-porcelain sites, --rolling-mode) #556
Labels
No labels
bump
major
bump
minor
bump
patch
kind/bug
kind/chore
kind/docs
kind/feature
priority/critical
priority/high
priority/low
priority/medium
size/L
size/M
size/S
size/XL
No milestone
No project
No assignees
1 participant
engineer
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
frankenbit/release-toolkit#556
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Port
release-prep.sh(785 LOC, 39 git-porcelain sites) tort prepsubcommand.Scope
scripts/release-prep.shtocmd/rt/prep.gocc_list_commits_since+_determine_bump_since(bash-source deferred to Phase 6 explicitly)--rolling-modefor rolling-PR refreshDeps.Manifest=nil(see companion refinement note below)AC
rt prepcompiles + fragment consumption atomic (integrates Cutter transactional guarantees via Request.Composed + nil-Manifest seams)--rolling-moderefreshes rolling PR byte-identical to bashrelease-prep.shbyte-oracle (release-prep.bats via #503 differential)TestPrepEquivalenceunder #503 differential harness) rather than by-site call-count match. The tick reflects behavioral equivalence, not per-site AST-level correspondence.Request.Composed=nilpath remains byte-unchanged (all #554 tests pass without modification)Deps.Manifest=nilpath: new property test proves fragment-consumption atomicity holds (partialProgressRollsBack on nil-Manifest path)go vet ./...+golangci-lint runcleanCutter refinement notes (2026-07-27, framing-verify surfaces)
Two refinements to #554's merged Cutter interface surfaced during Engineer's #556 framing-verify, both §3.3-sanctioned reality-intrudes shape. Both flagged for conscious ratify + named in the record so future readers see the refinement decisions, not slides.
Refinement 1: B-request seam (compose ownership) — bus 3ac5
Fork surfaced: how does rt-prep get richer compose (fragments + CC-commits + manual-prose + config-heading) AND preserve Cutter's fragment-consumption atomicity from #554?
Rejected:
changelog.Composer.Composesignature): forces every Composer caller to supply git-log they don't haveDeps.Composerseam): ZERO Cutter code change literally, but hides CC-ref/prior-CHANGELOG as captured deps behind a fragment-only Compose signature — semantic widening with no signature tell. Scope-at-point-of-use failure mode (per /srv/CLAUDE.md § Mechanism design)Ratified (Bosun 2026-07-27): B-request — additive-optional
Cutter.Request.Composed *changelog.VersionSection. rt-prep OWNS the richer compose (orchestrating already-ported primitives: MergeSections, ScaffoldMissingSections, NormalizeParagraphs, RenderCommitSections, RenderSections+Seal), builds the VersionSection, hands it to Cutter via Request.Composed. Cutter seals it transactionally.Composed=nil→ #554 fragment-only path unchanged.Refinement 2: nil-Manifest-skip companion — bus 50a7
Substrate finding (empirically verified):
release-prep.shwrites NO manifest. Manifest write lives atreusable-release.yml:391— POST-MERGE cut run, right afterdraft-release.sh(tag+publish), withlast_released_sha = merge commit. Rt-prep runs pre-merge; any manifest it wrote would carry a pre-merge SHA that's not the release commit. Manifest ownership is post-merge (#557's scope), not pre-merge.Corroborating anchor: v0.36.0 Option-A API-tag-bypass (earlier in this arc) required PR#871 hand-update of the manifest specifically because Option-A bypassed
draft-release.shwhich owns the manifest commit. Confirms manifest ownership is downstream of tag+publish, not upstream.Consequence: routing rt-prep through Cutter.Prepare as-ratified (with unconditional step (b) manifest write) would (1) write a file release-prep.sh never touches → AC3 byte-oracle equivalence cannot hold; and (2) be semantically wrong (pre-merge SHA in manifest).
Rejected alternative: rt-prep owns a thin seal+delete-fragments transaction separately, full Cutter reserved for #557. Duplicates §4 transactional logic in two places — precisely where Surveyor's (c)-branch (c)-hole would recur in the copy. Reuse-over-duplicate wins.
Ratified (Bosun 2026-07-27): nil-Manifest-skip —
Cutter.Deps.Manifest=nil→ Prepare skips step (b) (both Read and Write). rt-prep constructs Cutter with Manifest=nil → gets seal+delete-fragments atomic without manifest. #554 tests all wire real Manifest → behavior byte-unchanged; #557 wires real Manifest → full seal+manifest+delete atomic. §4 fragment-consumption invariant preserved (seal↔fragment property, manifest-independent). New property test required: partialProgressRollsBack on nil-Manifest path.Composite refinement framing (explicit)
The #554 "interface UNCHANGED" load-bearing claim from Phase-5→6 transition now becomes:
Both refinements are §3.3-sanctioned reality-intrudes shape. Two refinements to #554's merged code in one Phase-6-orchestrator dispatch is inside §3.3 tolerance. If additional refinements accrete across #557/#558/#559, revisit whether Cutter's phase-shape needs restructuring vs. accretive refinement (threshold: revisit at 5+).
Related
79e5947), #533 changelog.Composer, #532 fragments.Reader, #541 forgejo.Client, #545 forgejo rolling-mode PR opsRequest.Composedfield; (2)Deps.Manifestpermitted nilFiled 2026-07-27 by Bosun. Seam ratified (B-request) 2026-07-27 per Engineer framing-verify (bus 3ac5). Companion refinement ratified (nil-Manifest-skip) 2026-07-27 per Engineer framing-verify (bus 50a7).
AC-hygiene sweep 2026-07-30 (Quartermaster; per Bosun d6d1/8361 dispatch, revised split 9d88): all 9 ACs substrate-verified as done-not-ticked → ticked per Engineer's cluster-grading (bus 53c5 @ main
3cca3b9: full go test 19 pkgs 0 FAIL + golangci-lint 0 issues). AC5 carries done-with-disclosure annotation naming the bash-count vs Go-consolidation shape (equivalence is behavioral via TestPrepEquivalence, not per-site AST match). Re-fetched immediately pre-PUT per single-writer-per-body discipline (bus 8f09/9d88).CLOSED via PR#561 merged @
3385537(2026-07-27 04:02 CEST, rebase-merge, HTTP 200, gate 4/4 byte-verified pre-fire).ACs re-derived from merged substrate + Surveyor's independent deep-verify (bus 2b71):
rt prepcompiles + fragment consumption atomic — verified via §4 property tests green including new nilManifest-partialDelete-rollback (Surveyor's mutation-red confirmed drop-restoreFragments → correct §4 signature → byte-exact revert green)--rolling-moderefreshes rolling PR byte-identical to bash — harness coverage confirmsrelease-prep.shbyte-oracle (release-prep.bats via #503 differential) — full gate green host + forgejo-ci-go:latestRequest.Composed=nilpath remains byte-unchanged (all #554 tests pass without modification) — verified via CI 8/8 green including all #554 testsDeps.Manifest=nilpath: new property test proves fragment-consumption atomicity — nilManifest-partialDelete-rollback test present, mutation-verified by both Engineer + Surveyorgo vet ./...+golangci-lint runclean — CI gate confirmsTwo Cutter interface refinements ratified + built (both §3.3-sanctioned reality-intrudes shape): Request.Composed additive-optional (B-request seam); Deps.Manifest permitted nil for prefix-only callers. Composite refinement framing named consciously in tracker body pre-build.
Piggyback fixes (both latent bugs surfaced by the equivalence harness in already-merged code, both oracle-measured with load-bearing guards Surveyor mutation-confirmed):
Composite-milestone-gate design vindicated: the #554 AC-restate explicitly deferred equivalence-harness coverage to consumers #556/#557 with the reasoning that property-invariant at substrate + harness at consumer would surface defects each layer alone couldn't. Both catches prove it — filed as design-vindication anchor for future Phase-6 ratify decisions.
#499 stays open (folds into Cutter, closes at Phase 6 gate when #555-559 land + harness green). #557 (rt-release post-merge cut path) will complete the composite equivalence-harness coverage.
Closed by Bosun 2026-07-27 04:02 CEST per Engineer's Refs-only flag in PR body.
AC tick correction (Engineer honest coverage flag, bus c437):
My close comment above ticked "
--rolling-moderefreshes rolling PR byte-identical to bash" citing "harness coverage confirms" — that was a substrate-optimistic tick. Engineer's honest flag: the rolling glue (FindPRByHead → PATCH-or-POST) is ported + wired + Surveyor-reviewed with component coverage (compose via harness, PR-payload format via #541), but there is NO end-to-end rolling differential case in the equivalence harness. FindPRByHead is a read that hits network in a no-remote scratch, so a hermetic end-to-end rolling case needs a forge-injection seam incmd/rt.Corrected tick shape (per /srv/CLAUDE.md § AC tick discipline — state-asserting AC restated as action-AC pointing at follow-up):
[x]--rolling-moderefreshes rolling PR byte-identical to bash — harness coverage confirms--rolling-modeported + wired + Surveyor-reviewed + component-covered (compose harness + #541 PR-payload); end-to-end rolling differential deferred to #562 pending cmd/rt forge-injection seam.State of the substrate: rolling glue lives in main + reviewed + component-tested. Coverage residual is on the harness completeness axis (differential-test-case-for-rolling), tracked at #562.
Also worth naming as a substrate-of-record anchor: this is exactly the "absence claim doesn't survive a filtered view" discipline from the reflex table. I ticked based on filtered coverage (components green) without asking whether the END-TO-END surface had a case. Engineer's honest surfacing IS the discipline working — the ac_tick_discipline correction shape at close time. Anchor for future close-time tick discipline: when re-deriving state-assertion ACs from merged substrate, name the surface the coverage sits on (component, end-to-end, integration, etc.). Filtered-coverage tick is not the same as scoped-coverage tick.